University of Southern California Dissertations and Theses

When things are left unsaid: existential and anaphoric implicit objects in discourse

(USC Thesis Other)

When things are left unsaid: existential and anaphoric implicit objects in discourse

PDF

Download

Open document

Flip pages

Download a page range

Download transcript

Copy asset link

Request this asset

Transcript (if available)

Content

WHEN THINGS ARE LEFT UNSAID:
EXISTENTIAL AND ANAPHORIC IMPLICIT OBJECTS IN DISCOURSE
By
Ana Elisa Piani Besserman Vianna

A Dissertation Presented to the
FACULTY OF THE USC GRADUATE SCHOOL
UNIVERSITY OF SOUTHERN CALIFORNIA
In Partial Fulfillment of the
Requirements for the Degree
DOCTOR OF PHILOSOPHY
(LINGUISTICS)

May 2021

Copyright 2021 Ana Elisa Piani Besserman Vianna
ii
Dedication

To my inner critic: you were wrong!

iii
Acknowledgements

Brace yourselves for a gratitude storm. Here it comes.
First and foremost, a huge thanks to everybody who has participated in my experiments!
Thank you to my advisor, Elsi Kaiser: a most supporting mentor, an incredibly inspiring scholar,
a passionate and dedicated teacher, and above all, a kind and caring human being. Thank you to
Roumi Pancheva, Toby Mintz and Jason Zevin, for your invaluable support, advice and feedback.
I have learned so much from all of you, and I will lovingly carry that with me, always.
I feel grateful to so many students, staff and faculty at USC – too many to fit here. Thank you to
Guillermo Ruiz, who is a lifeline to all graduate students. Thank you to RA extraordinaire Jeffrey
Cho for annotating my data. And thank you to all my grad comrades, especially my cohort and the
psycholinguistics lab, who have been by my side through every stage of this challenging enterprise.
Thank you to my parents, Sérgio and Guida, and my brother, André, for their unwavering love and
support. Also thank you to my irmãs, Patricia, Nina and Julia, and to all my incredible friends.
Thank you to my husband Marshall, my bird, my rock. I could write a million dissertations about
how much I love and appreciate you. Thank you to Buffy and Blade: you deserve all the tuna.

Gratitude storm: over. Look, I think I see a rainbow just around the corner.
Now fasten your seatbelts and get ready for a journey through the life of implicit objects.
iv
Table of Contents
Dedication ....................................................................................................................................... ii
Acknowledgements ........................................................................................................................ iii
List of Tables ................................................................................................................................ vii
List of Figures ................................................................................................................................ xi
Abstract ........................................................................................................................................ xiii
Chapter 1 : Introduction .................................................................................................................. 1
1.1. Implicit objects and context dependency ............................................................................. 5
1.2. Discourse Behavior: Accessibility and Persistence ........................................................... 14
1.3. Processing implicit objects ................................................................................................ 17
1.4. Overview of this dissertation ............................................................................................. 20
Chapter 2 : Existential implicit objects’ accessibility and persistence in discourse: a sluicing
investigation .................................................................................................................................. 23
2.1. Introduction ........................................................................................................................ 23
2.2. Accessibility and Referring Expressions ........................................................................... 27
2.2.1. The accessibility of implicit arguments ...................................................................... 28
2.2.2. What about implicit objects? .......................................................................................... 32
2.2.3. Investigating the accessibility of implicit objects using sluicing................................ 35
2.3. Experiment 1: Are more informative referring expressions better able to refer
back to implicit objects? ........................................................................................................... 41
2.3.1. Methods....................................................................................................................... 42
2.3.2. Results ......................................................................................................................... 44
2.3.3. Discussion ................................................................................................................... 47
2.4. Experiment 2: Does representational complexity affects the accessibility of
discourse-linked implicit objects?............................................................................................. 49
2.4.1. Methods....................................................................................................................... 55
2.4.2. Results ......................................................................................................................... 56
2.4.3. Discussion ................................................................................................................... 59
2.5. Discourse Persistence......................................................................................................... 59
2.5.1. Discourse Persistence and Implicit Arguments .......................................................... 61
2.5.2. Discourse Persistence and the Question under Discussion ......................................... 63
2.5.3. Investigating the discourse persistence of implicit objects using sluicing ................. 64
2.6. Experiment 3: Are implicit objects less discourse persistent than overt objects? ............. 65
2.6.1. Methods....................................................................................................................... 67
2.6.2. Results ......................................................................................................................... 69
2.6.3. Discussion ................................................................................................................... 71
v
2.7. General Discussion ............................................................................................................ 72
Chapter 3 : Existential implicit objects’ accessibility and persistence in discourse: a story
continuation task investigation ..................................................................................................... 75
3.1. Introduction ........................................................................................................................ 75
3.2. Discourse Behavior: Accessibility and Persistence ........................................................... 76
3.2.1. Accessibility ................................................................................................................ 77
3.2.2. Persistence................................................................................................................... 79
3.2.3. The relationship between Accessibility and Persistence ............................................ 80
3.3. Broader discourse factors: prior context, discourse coherence .......................................... 81
3.3.1. Prior context ................................................................................................................ 82
3.3.2. Discourse Coherence .................................................................................................. 83
3.4. Investigating the discourse behavior of implicit objects in a story continuation task ....... 84
3.4.1 Methods........................................................................................................................ 85
3.4.2. Results and Discussions .............................................................................................. 94
3.5 General Discussion ........................................................................................................... 116
Chapter 4 : Implicit objects in comprehension: Effects of context-dependency ........................ 122
4.1. Introduction ...................................................................................................................... 122
4.2. Context-dependency and Implicit Arguments ................................................................. 124
4.2.1. Existential and Anaphoric implicit objects in English and Brazilian Portuguese .... 130
4.3. Investigating effects of context-dependency during the comprehension of implicit
objects in a self-paced reading task ........................................................................................ 138
4.3.1. Methods..................................................................................................................... 139
4.3.2. Results and Discussions ............................................................................................ 143
4.4. General Discussion .......................................................................................................... 152
Chapter 5 : Implicit objects in comprehension: Garden-path sentences ..................................... 158
5.1. Introduction ...................................................................................................................... 158
5.2. The comprehension of Garden-Path sentences with a temporary object/subject
ambiguity ................................................................................................................................ 163
5.2.1. Garden-Path reanalysis: implicit object interpretation ............................................. 165
5.3. Investigating how implicit object interpretation affects the comprehension of
garden-path sentences in a self-paced reading task ................................................................ 178
5.3.1. Methods..................................................................................................................... 180
5.3.2. Results ....................................................................................................................... 184
5.4. General Discussion .......................................................................................................... 196
Chapter 6 : General Discussion and Conclusions ....................................................................... 205
References ................................................................................................................................... 215
Appendix 1 .................................................................................................................................. 227
vi
Appendix 2 .................................................................................................................................. 230
Appendix 3 .................................................................................................................................. 231
Appendix 4 .................................................................................................................................. 233
Appendix 5 .................................................................................................................................. 234
Appendix 6 .................................................................................................................................. 241
Appendix 7 .................................................................................................................................. 246
Appendix 8 .................................................................................................................................. 254

vii
List of Tables
Table 2.1 Experiment 1: Conditions and Target item examples ................................................... 43
Table 2.2 Experiment 1: Raw means and standard error of participants’ acceptability scores by
object type and wh-phrase type (5-point scale) ............................................................................ 45
Table 2.3 Experiment 1 Results: Main effects and interactions (“*”: p < 0.05) ........................... 45
Table 2.4 Table 2.4. Experiment 1 Results: Planned pairwise comparisons (“*”: p < 0.05)........ 46
Table 2.5 Experiment 1: Raw means and standard error of participants’ acceptability scores in
ungrammatical control comparison (5-point scale) ...................................................................... 46
Table 2.6 Experiment 1 Results: Ungrammatical control comparison (“*”: p < 0.05) ................ 47
Table 2.7 Exp. 2: Conditions and target items examples .............................................................. 56
Table 2.8 Experiment 2: Raw means and standard error of participants’ acceptability scores by
object type and context set type (5-point scale) ............................................................................ 57
Table 2.9 Experiment 2 Results: Main effects and interactions (“*”: p < 0.05) ........................... 58
Table 2.10 Experiment 2: Raw means and standard error of participants’ acceptability scores in
ungrammatical control comparison (5-point scale) ...................................................................... 58
Table 2.11 Experiment 2 Results: Ungrammatical control comparison (“*”: p < 0.05) .............. 58
Table 2.12 Experiment 3: Conditions and target item examples ................................................. 68
Table 2.13 Experiment 3: Proportion of object responses and SE by object type and QUD ....... 70
Table 2.14 Experiment 3 Results: Main effects and interactions (“*”: p < 0.05) ......................... 70
Table 3.1 Examples of target items and conditions ...................................................................... 87
Table 3.2 Examples of continuations produced by participants and how they were annotated (the
first mention of the prompt object appears underlined in bold).................................................... 89
viii
Table 3.3 Proportion and SE of references to prompt object by condition and sentences
(S1/S2/S3) ..................................................................................................................................... 95
Table 3.4 Proportion and SE of references to prompt object by condition................................... 97
Table 3.5 Results (Persistence): Main effects and interactions (“*”: p < 0.05) ............................ 97
Table 3.6 Results (Persistence): Planned pair comparisons (“*”: p < 0.05) ................................. 97
Table 3.7 'Persistence' vs. 'Talking about/Mentioning the object' proportions .......................... 100
Table 3.8 Proportion and SE of continuations that talked about prompt object by condition .... 101
Table 3.9 Results (Overall mentions): Main effects and interactions (“*”: p < 0.05) ................ 101
Table 3.10 Proportion and SE of prompt object pronominalization by condition ...................... 104
Table 3.11 Results (Accessibility: pronominalization rate): Main effects and interactions (“*”: p
< 0.05) ......................................................................................................................................... 105
Table 3.12 Proportion and SE of prompt object references realized in Subject position by
condition ..................................................................................................................................... 106
Table 3.13 Results (Accessibility: Promotion to subject): Main effects and interactions (“*”: p <
0.05) ............................................................................................................................................ 107
Table 3.14 Distribution of all continuations by condition and coherence relation +-1 SE ......... 109
Table 3.15 Distribution of references to the prompt object by condition and coherence relation +-
1 SE ............................................................................................................................................. 111
Table 3.16 Proportion of references to the prompt object in Elaboration and Occasion relations
+-1 SE ......................................................................................................................................... 112
Table 3.17 Results (Exploratory Analysis: Persistence in Elaboration coherence relations): Main
effects and interactions (“*”: p < 0.05) ....................................................................................... 114
ix
Table 3.18 Results (Exploratory Analysis: Persistence in Elaboration coherence relations):
Planned pair comparisons (“*”: p < 0.05) ................................................................................... 114
Table 3.19 Results (Exploratory Analysis: Persistence in Occasion coherence relations): Main
effects and interactions (“*”: p < 0.05) ....................................................................................... 114
Table 3.20 Results (Exploratory Analysis: Persistence in Occasion coherence relations): Planned
pair comparisons (“*”: p < 0.05)................................................................................................. 114
Table 4.1 Summary of the types of implicit objects allowed per verb type. .............................. 133
Table 4.2 Example target items (BP on top, English translation below; ∅ indicates an IO. The
critical region for the main question is underlined. The critical region for the exploratory
question begins at “guardou” and includes the threw following words) .................................... 142
Table 4.3 Mean RTs (in milliseconds) and SE for words in the critical region by verb and
antecedent type............................................................................................................................ 145
Table 4.4 Reading Time Results for words in the critical region, “No Antecedent” inquiry (“*”: p
< 0.05) ......................................................................................................................................... 145
Table 4.5 Planned Comparisons for words in the critical region, “No Antecedent” inquiry (“*”: p
< 0.05) ......................................................................................................................................... 146
Table 4.6 Mean RTs in milliseconds and SE for words in the exploratory critical region
(Existential implicit object antecedent inquiry) by verb type and antecedent type .................... 148
Table 4.7 Reading Time Results for words in the exploratory critical region, “Existential implicit
object antecedent inquiry” inquiry (“*”: p < 0.05) ..................................................................... 149
Table 4.8 Proportion of ‘yes’ responses +-1 SE by verb type and antecedent ........................... 151
Table 4.9 Comprehension Question Results (“*”: p < 0.05) ...................................................... 151
x
Table 5.1 Example target items (BP on top, English translation below; ∅ indicates an implicit
object) ......................................................................................................................................... 182
Table 5.2 Mean RTs and SE for words in the critical region by antecedent and definiteness for
UNAMBIGUOUS (with comma) conditions ............................................................................. 188
Table 5.3 Table 5.3. Mean RTs and SE for words in the critical region by antecedent and
definiteness for AMBIGUOUS (no comma) conditions ............................................................ 188
Table 5.4 Reading Time Results: Main Effects for words in the critical region (“*”: p < 0.05) 189
Table 5.5 Reading Time Results: 2-way interactions for words in the critical region (“*”: p <
0.05) ............................................................................................................................................ 189
Table 5.6 Reading Time Results: 3-way interaction for words in the critical region (“*”: p <
0.05) ............................................................................................................................................ 189
Table 5.7 Planned Comparisons for Verb position (“*”: p < 0.05) ............................................ 190
Table 5.8 Planned Comparisons for Spill 2 position (“*”: p < 0.05) ......................................... 190
Table 5.9 Proportion of ‘yes’ responses +-1 SE by ambiguity, antecedent availability and
definiteness of NP ....................................................................................................................... 194
Table 5.10 Comprehension Question Results: Main effects and interactions (Indefinites only;
“*”: p < 0.05) .............................................................................................................................. 195
Table 5.11 Comprehension Question Results: Planned Pairwise comparisons (Indefinites only;
“*”: p < 0.05) .............................................................................................................................. 195

xi
List of Figures

Figure 2.1 Experiment 1: raw means and +-1 SE by object type and wh-phrase type (5-point scale)
....................................................................................................................................................... 46
Figure 2.2 Experiment 1: raw means and +-1 SE in ungrammatical control comparison ............ 47
Figure 2.3 Experiment 2 raw means and +-1 SE by object type and wh-phrase type (5-point scale)
....................................................................................................................................................... 58
Figure 2.4 Experiment 3: Proportion of object responses +-1 SE by object type and QUD type 71
Figure 3.1 Proportion of references to prompt object +-1 SE by conditions and sentences
(S1/S2/S3) ..................................................................................................................................... 96
Figure 3.2 Proportion of references to prompt object +-1 SE by context type and prompt object
type ................................................................................................................................................ 97
Figure 3.3 Proportion of continuations that TALK ABOUT/MENTION the prompt object (with
any referring expression) +-1 SE by context type and prompt object type ................................ 101
Figure 3.4 Proportion of REFERENCES relative to ALL MENTIONS by context type and prompt
object type (Mention includes all instances of talking about the prompt object with any referring
expression; Reference includes only the instances of talking about the prompt object with given
expressions)................................................................................................................................. 102
Figure 3.5 Proportion of pronominalizations +-1 SE by context type and prompt object type .. 104
Figure 3.6 Proportion of references to the prompt object realized in Subject position +-1 SE by
context type and prompt object type ........................................................................................... 107
Figure 3.7 Distribution of all continuations by condition and coherence relation +-1 SE ......... 110
Figure 3.8 Distribution of references to the prompt object by condition and coherence relation +-
1 SE ............................................................................................................................................. 111
xii
Figure 3.9 References in Elaborations ........................................................................................ 113
Figure 3.10 References in Occasions .......................................................................................... 113
Figure 4.1 Word-by-word reading times +-1 SE by verb and antecedent type (English translation:
The Clara [entered/selected a book], [read/flipped through/sat] on couch of living-room, then put-
away on shelf and left) ................................................................................................................ 145
Figure 4.2 Proportion of ‘yes’ responses to comprehension question +-1 SE ............................ 151
Figure 5.1 Word-by-word reading times +-1 SE by ambiguity, antecedent and definiteness
(English translation: After that I [received a letter and] read [,] [a/the] book fell on my foot left).
..................................................................................................................................................... 189
Figure 5.2 Proportion of ‘yes’ responses to comprehension question +-1 SE ............................ 195

xiii
Abstract

This dissertation aims to expand our knowledge on implicit linguistic phenomena, their discourse
properties and how they participate in comprehension processes. In particular, this dissertation
looks at the underexplored domain of direct objects that are introduced into the discourse
implicitly, such as what Ana was writing in the sentence “Ana was writing.”
Chapters 2 and 3 focus on the discourse status of existential implicit objects (such as the implicit
object of ‘writing’ introduced above), which have been claimed to be “discourse inert”. Chapter 2
presents two acceptability judgement tasks and one forced-choice sentence completion task using
sluicing constructions. Findings revealed that existential implicit objects are less accessible to
reduced expressions, but they can be as discourse persistent as overt objects. Chapter 3 expands
on this line of inquiry with a story continuation task, which reveals that although relatively less
accessible and persistent, existential implicit objects do participate in the structuring of subsequent
discourse and they are sometimes pronominalized (i.e., they are not completely discourse inert).
Crucially, Chapter 3 also reveals that a prior context can affect implicit objects’ discourse
persistence, but not their accessibility.
Chapter 4 investigated how the context-dependency properties of implicit objects affects their
comprehension. A self-paced reading task in Brazilian Portuguese (a language with both context-
dependent and -independent implicit objects, i.e., anaphoric/existential) shows that the language
processor is able to quickly identify which types of implicit objects are allowed by each verb.
Moreover, this task shows that the comprehension of anaphoric implicit objects with no available
antecedents in context leads to processing difficulties, however no difficulties occur when there is
at least an implicit antecedent available. In contrast, the comprehension of existential implicit
objects does not suffer in the absence of antecedents.
xiv
Chapter 5 investigated how the interpretation of implicit objects in sentences with temporary
subject/object ambiguities (garden-paths) affects sentence re-analysis and the occurrence of
lingering misinterpretations. A self-paced reading task in Brazilian Portuguese shows that an
anaphoric interpretation of the implicit object in garden path reanalysis leads to a large reduction
in the proportion of lingering misinterpretations.
In sum, this dissertation shows that existential implicit objects are not entirely “discourse inert”
and do participate in people’s mental models of the discourse. Moreover, investigations on the
comprehension of context-dependent and -independent implicit objects show that the language
processor is able to quickly recruit information from various domains (e.g., verb argument
structure, prior context, general syntactic parsing biases) when dealing with these implicit
elements.

1
Chapter 1 : Introduction

People are able to quickly and effortlessly decode language input, even when the input may seem
incomplete. Speakers often produce utterances in which part of the content is elided or implicit,
such as “Nicole wrote something, but Richard doesn’t know what [Nicole wrote].” or “Nicole was
reading [?] in the kitchen.” (content within brackets is unpronounced; the question mark denotes
content that is unspecified, i.e., best paraphrased with “something”). In the first case, called
sluicing, the entire last clause is omitted except for the wh-phrase, but the meaning of the elided
clause can be reconstructed from the first clause. In the second case, the direct object of the verb
“reading” is omitted, i.e., it is an unspecified implicit object. Despite these ‘missing pieces’, most
of the time we are able to somehow reconstruct and comprehend the intended message.
Linguists have long been concerned with how, exactly, humans are able to put together
these missing pieces and what their linguistic properties are. We now know that there are many
different kinds of implicit phenomena in language, and the ways in which they behave
linguistically can vary drastically. In my dissertation, I focus on the category of implicit arguments:
these are nominal or verbal arguments, such as the ones specified in a verb’s argument structure,
that are not realized overtly in speech or writing. For instance, the verb ‘to read’ has two argument
slots: one for the agent (the reader) and another for the theme (what is being read). In a sentence
such as “This book has been read.”, the agent is omitted. In a sentence such as “Lisa was reading.”,
it is the theme that is missing. This dissertation is centered around the latter case, broadly known
as implicit objects. Crucially, implicit objects remain a core component of the event’s meaning
even when they are omitted: for example, in a reading event, there must be someone doing the
reading and there must be something being read, even if they are not overtly mentioned.
2
Although implicit arguments are a frequent part of communication, they pose challenges
for theories of language processing and in particular for the crucial phenomenon of reference
tracking: i.e., keeping track of who and what is being talked about. Reference tracking is a
fundamental part of language use, without which successful communication cannot occur. For
example, when people encounter a pronoun (e.g., “she”, “it”, “they”), they must decipher to what
or to whom the pronoun is referring in order to successfully interpret the message.
Entities that have already been overtly mentioned are often easily referred to with
pronouns: In (ex.1.1a), ‘it’ refers back to the cake Lisa baked. The relationship between the noun
phrase ‘a cake’ and the subsequent pronoun ‘it’ is attributed to how entities are represented in
people’s mental discourse models. My mental discourse model has “pegs” for mentioned referents
such as Lisa and cake, and I am able to refer back to these pegs by using pronouns (e.g., Heim
1982).
1

(1.1) a. Lisa baked a cake last night. Itcake tasted delicious.
b. Lisa baked last night. ?? It tasted delicious.
However, things are less clear with implicit objects, as in (ex.1.1b). Can the pronoun ‘it’
refer to what Lisa baked last night? Empirically, the answer is not straightforward and prior work
disagrees (e.g., Resnik 1983, Koenig & Mauner 1999, Williams 2015). Theoretically, the answer
is significant for our understanding of reference tracking and how people build their mental
representations of the discourse: If the answer is yes, it suggests that implicit objects are
represented in the discourse model similarly to overt referents. What does this mean for the mental

1
This is an extremely simplified take on pronoun resolution and coreference; this topic will be further discussed in
subsequent sections of this dissertation under the lens of accessibility theory.
3
representations of other kinds of unmentioned information? If the answer is no, should we
conclude that implicit objects are simply not represented in the discourse model at all? Or perhaps
they are represented in the discourse, but are not sufficiently accessible/prominent to be picked up
with a pronoun?
To make matters more complicated, there are also various kinds of implicit objects, which
may differ in their discourse functions. For instance, implicit objects like the one associated with
“baked” in (ex.1.1b) are not able to refer back to other entities; that is, they are not able to point to
specific ‘pegs’ in people’s mental discourse models (ex.1.2a). In contrast, other types of implicit
objects are able to refer back to previously-mentioned entities, similarly to pronouns ((ex.1.2b);
both of these
2
will be further discussed in the subsection below about implicit objects and context-
dependency).

(1.2) a. *The lasagna was raw, so Lisa baked. [meaning: Lisa baked the lasagna that was
raw.]

b. The car broke down, so Ana pushed. [meaning: Ana pushed the car that broke
down.]

Notably, these differences indicate that the language parser and its reference tracking
system must be nuanced and powerful enough to handle two distinct phenomena that are identical

2
Note, however, that none of the examples so far (ex.1.1a-b, ex.1.2a-b) are considered cases of null pronouns/pro-
drop, which are phonologically null but syntactically present. It is unclear whether the implicit objects in these
examples are represented in the syntax at all; this issue is briefly discussed later in this dissertation, both in regard to
implicit objects in English and in Brazilian Portuguese. Nonetheless, this dissertation focuses on discourse-level
properties of implicit objects and does not further investigate their syntactic status.
4
in their form - both are unpronounced/invisible implicit objects - but have drastically different
discourse functions. How does the language parser identify various implicit objects and their
discourse properties?
The answer lies at least partially in the fact that implicit objects cannot be disentangled
from their verbs. Implicit objects are part of a verb’s argument structure. This compositional
nature makes them a particularly interesting phenomenon from the perspective of reference
tracking. Most work on this domain has been on independent expressions like overt and covert
pronouns and noun phrases, whereas not much has been done on phenomena such as implicit
objects, which are intrinsically connected to their verbs and may differ across verb types and across
languages. For example, the verbs “to cook” and “to bake” allow implicit objects in English,
whereas the verb “to fry” does not. Furthermore, in English, verbs like “to cook” and “to bake”
only allow for one kind of implicit object. In Brazilian Portuguese, a verb like cozinhar (“to cook”)
can have two types of implicit objects: one that is interpreted anaphorically in relation to another
referent (unlike English), and one that cannot be interpreted anaphorically (like English). In
contrast, a verb like assar (“to bake”) can only have one type of implicit object, the anaphoric
kind.
In addition, the verb also contributes to how people interpret the implicit object: in
(ex.1.1b) it is understood that Lisa baked something, and moreover, that this something is a typical
baked good like cake or croissants, but most likely not something like ham and potatoes (even
though these can technically be baked as well; see Condoravdi & Garwon 1996, Martí 2011, Glass
2020). A verb like “eat” can have an implicit object compatible with interpretations in which what
was eaten was a meal or type of food typically eaten, but not an unusual object such as “paper” or
“nails” (even if these objects could, technically, be eaten).
5
The examples above illustrate the importance of the verb in the interpretation of implicit
objects. Information about the verb is typically known to be quickly accessed by the language
parser and plays a crucial role during comprehension (e.g., Boland et al. 1990, Trueswell et al.
1993, Boland 1993). When it comes to implicit objects, the parser needs to be able to use
information associated with verbs to determine which (if any) types of implicit objects are allowed,
how they can be interpreted and how they behave in discourse. I.e., in the case of implicit objects,
the verb plays a crucial role for comprehension and reference tracking.
This dissertation investigates the discourse behavior and comprehension of two types of
implicit objects: one that does not have a specific referent (like the implicit object of “bake” in
ex.1b) and one that is able to point to a specific referent. Although their form is identical (and
‘invisible/unpronounced’), these implicit objects have drastically different functions and each
presents its own challenges to language processing and referent tracking. The subsections below
introduce and discuss in more detail the two types of implicit objects, as well as features of
discourse behavior and issues in comprehension that are relevant for the studies presented here.
1.1. Implicit objects and context dependency

There are various kinds of implicit arguments, which vary in their syntactic, semantic and
discourse behavior (for an overview, see Bhatt & Pancheva 2006). Broadly speaking, however, the
interpretation of implicit arguments – and more generally, any expression – can be construed as
belonging in one of two main categories: context-dependent or context-independent (e.g.,
Condoravdi & Gawron 1996, Gillon 2006). To illustrate this point, let’s consider two overt
expressions, such as “a cat” and “her”:
6
An indefinite noun phrase, like “a cat”, can be effortlessly interpreted by English speakers
without the need to obtain any additional information from the context. That is, an out-of-the-blue
utterance such as “I saw a cat!” requires no additional context for successful interpretation.
3
Such
expressions are considered context-independent.
In contrast, some expressions, like the pronoun “her”, require further context for successful
interpretation. That is, the meaning of an expression like “her” is obtained from context (e.g.,
Condoravdi & Gawron 1996; see also Cresswell 1973, Kaplan 1989, Lewis 1981, Stalnaker 1972).
An out-of-the-blue utterance such as “I saw her!” requires the comprehender to resolve to whom
the pronoun “her” refers to. The value/referent used to resolve the meaning of “her” can typically
be obtained from context, including previous discourse and accessible referents in the common
ground. For instance, if the entire context of the utterance had been “Roxane Gay arrived. I saw
her!”, the meaning of “her” can be interpreted as coreferring with the previously introduced entity
“Roxane Gay”. Expressions like “her” are considered context-dependent.
Although implicit objects all share the same unpronounced/invisible form, some are
interpreted independent of any context whereas some are interpreted as coreferential with a
referent provided by the context (e.g., Fillmore 1986, Gillon 2006, Anderbois 2012).
In English, implicit objects like the one for “bake” (ex.1b) are context-independent: an
utterance such as “I am baking.” can be interpreted successfully without any additional
information. The implicit object of “baking” does not get its meaning from a value or referent
supplied from context; this implicit object is interpreted independently of context, as a non-specific
‘something/stuff’. Such implicit objects are known as existential implicit objects because they
entail the existence of an object, without specifying its identity (e.g., Fillmore 1986, Anderbois

3
Other than knowing the interlocutor, time and location of utterance.
7
2012, Williams 2015). That is, an utterance such as “I am baking.” entails the existence of
something being baked, although the identity of what is being baked is not specified.
While the precise identity of these existential implicit objects is unknown, the lexical
semantics of the verb “bake” and the contexts in which such an expression is commonly used do
supply a strong inference that the “something” being baked is a typical baked good, such as muffins
or bread, and not something less typical, like ham – let alone something completely atypical in the
context of baking such as nails (e.g., Condoravdi & Gawron 1996; Martí 2011, Glass 2020).
Existential implicit objects generally arise within a class of verbs called Optionally
Transitive (OPT) verbs, like the ones in (ex.1.3a-f). These verbs typically express conventional
activities, those that are part of routines in the communities they are used (e.g., Martí 2011, Glass
2020). When OPT verbs are used without an overt direct object, they introduce an entailment about
the existence of an unspecified implicit object associated with that activity. Some examples of
OPT verbs in English can be seen below in (ex.1.3a-f); implicit objects are represented with “∅”:

(1.3) a. Roxane baked ∅ yesterday.
b. Nnedi was writing ∅ in her car.
c. Tressie is planning to study ∅ at the library.
d. I think Nora is painting ∅.
e. Toni has been reading ∅ for hours.
f. Have you eaten ∅ already?
8
Beyond existential implicit objects, English also has instances of implicit objects that are
context-dependent: that is, their meaning is a value that must be acquired through context. One
example is the verb “to notice”: without a value provided by context, the use of such a verb without
an overt object is infelicitous (ex.1.4a) because there is no way to recover the meaning of the
implicit object (e.g., Fillmore 1969, Anderbois 2012). The same occurs with other verbs with
context-dependent implicit objects, such as “to push” and “to apply” (ex.1.4c, 1.4e): when
presented with no prior context to provide a meaning for the implicit object, infelicity arises. This
contrasts directly with existential implicit objects (ex.1.3a-f), which do not require any previous
context.
(1.4) a. #Mia noticed ∅.
b. Pearl got a new haircut. The next morning, Mia finally noticed ∅.
c. #Stevie pushed ∅.
d. David started the car and Stevie pushed ∅.
e. #He applied ∅.
f. Jack heard about a job opening at the University. Later, he applied ∅.
When prior context does supply a value that can provide a meaningful interpretation to
context-dependent implicit objects, these objects are felicitous and interpreted anaphorically. Thus,
implicit objects of this kind are generally known as anaphoric implicit objects. Existential and
anaphoric implicit objects are also known as indefinite and definite implicit objects, respectively;
while the former introduces an unspecified new referent and is better paraphrased with an
9
indefinite noun phrase (ex.1.5a), the latter refers back to a referent that has been previously
introduced in the discourse and is best paraphrased with a definite noun phrase ((ex.1.5b); see
Fillmore 1986, Condoravdi & Gawron 1996, Anderbois 2012).

(1.5) a. Julia baked ∅. => Julia baked a baked good.

b. David started the car and Stevie pushed ∅. => Stevie pushed the car David started.

The clause with the anaphoric implicit object in (ex.1.4b) can be interpreted as “Mia finally
noticed the new haircut/that Pearl got a new haircut”; the one in (ex.1.4d) as “Stevie pushed the
car David started”, and (ex.1.4f) as “He applied for the job opening at the University”.
The contrast between anaphoric/context-dependent and existential/context-independent
implicit objects becomes even clearer when these verbs are inspected under negation. For instance,
(ex.1.6a) is interpreted as a denial of the existence of an implicit object, i.e., as “Roxane didn’t
bake anything yesterday”. If Roxane had baked any single baked good the day before, the
statement in (ex.1.6a) would be false. In (ex.1.6b), although there is a potential referent to supply
a meaning for the implicit object – “a layered cake” – the implicit object of “bake” is still
interpreted existentially, so that its negation means that Roxane did not bake anything, not the
layered cake nor any other baked good. In this case, if Roxane had baked a croissant the day before,
the statement in (ex.1.6b) would still be considered false.

(1.6) a. Roxane didn’t bake ∅ yesterday.
b. Roxane had planned on making a layered cake but she didn’t bake ∅ yesterday.
10
c. #Stevie didn’t push ∅.
d. David started the car but Stevie didn’t push ∅.
On the other hand, the implicit object of a verb like “to push” is always interpreted
anaphorically – leading to infelicity when no supporting context is available (ex.1.6c). Even under
negation, the implicit object of “push” is still interpreted in reference to the value provided by
previous context: (ex.1.6d) is interpreted as “David started the car but Stevie didn’t push the car
David started.” If Stevie had pushed a different car, the statement in (ex.1.6d) would remain true
because the implicit object is interpreted anaphorically to a specific prior referent, i.e. the car David
started. In sum, context-dependent implicit objects are always interpreted anaphorically, even
under negation. In contrast, context-independent implicit objects are interpreted existentially, so
when they appear under negation, the existence of the implicit object itself is negated.
In English, the distribution and interpretation of context-dependent and -independent
implicit objects is quite asymmetrical. Verbs with context-dependent implicit objects in English
like (ex.1.4a-f) are fewer and more varied within themselves. Specifically, they vary in what value
is recovered to supply the meaning of the implicit object: in some cases, it seems to be an entire
proposition (ex.1.4b); in others, a prepositional phrase (ex.1.4f) or a noun phrase (ex.1.4d). On the
other hand, context-independent implicit objects occur with a wider range of verbs and generally
represent noun phrases, i.e. the verb’s theme (ex.1.3a-f).
In order to systematically investigate differences regarding context-dependency in the
comprehension of implicit objects, this dissertation probes into a context-dependent implicit object
different from those seen in English. More specifically, this work investigates anaphoric implicit
objects in Brazilian Portuguese (also commonly referred to simply as “null objects”; see Farrell
1990, Kato 1993, Cyrino 1996, Cyrino & Lopes 2016). Brazilian Portuguese (henceforth BP)
11
allows for a more direct comparison between context-independent and context-dependent implicit
objects. In BP, unlike in English, both types of implicit objects generally play the semantic role of
themes and are typically best represented by noun phrases.
BP has the same kind of existential implicit objects as English, which occur with optionally
transitive verbs describing common, routine activities. Although there is great overlap between
most optionally transitive verbs in the two languages (e.g., (ex.1.7b-f)), not every OPT verb in
English is also an OPT verb in BP, and vice-versa. For example, “to bake” allows an Existential
implicit object in English but not in BP (ex.1.7a).

(1.7) a. #Roxane assou ∅ ontem.
Roxane baked ∅ yesterday.
b. Nnedi estava escrevendo ∅ em seu carro.
Nnedi was writing ∅ in her car.
c. Tressie está planejando estudar ∅ na biblioteca.
Tressie is planning to study ∅ at the library.
d. Eu acho que Nora está pintando ∅.
I think Nora is painting ∅.
e. Toni está lendo ∅ há horas.
Toni has been reading ∅ for hours.
f. Você já comeu ∅?
Have you eaten ∅ already?
12
Just like in English, existential implicit objects in BP are context-independent. In contrast,
BP anaphoric implicit objects – like the name suggests – are interpreted anaphorically, in relation
to a referent introduced by the context. They can occur with any transitive verb as long as there is
an available referent. Consequently, in addition to existential implicit objects, optionally transitive
verbs in BP can also have context-dependent, anaphoric implicit objects: in (ex.1.8a), the implicit
object of the OPT verb lendo (“reading”) is interpreted as the previously mentioned livro sobre
penguins (“book about penguins”). Crucially, for OPT verbs in BP, it is the availability of an
antecedent (as well as the absence of an overt direct object) that determines whether the verb will
yield an existential implicit object (ex.1.7e) or an anaphoric one (ex.1.8a).
(1.8) a. Eu dei um livro sobre penguins para a Toni e ela está lendo ∅ há horas.
I gave a book about penguins to Toni and she has been reading [∅ = it/the book
about penguins] for hours.
Another type of verb, namely obligatorily transitive ones (OT verbs) like folhear (“to flip-
through”
4
), can only have context-dependent, anaphoric implicit objects. Thus, they are infelicitous
when no antecedent is present (ex.1.9a) but felicitous when one is available (ex.1.9b).
(1.9) a. #Toni está folheando ∅ há horas.
Toni is flipping-through ∅ for hours.
b. Eu dei um livro sobre penguins para a Toni e ela está folheando ∅ há horas.
I gave a book about penguins to Toni and she is flipping-through [∅ = it/the
book about penguins] for hours.

4
As in flipping through the pages of a book or magazine.
13
Finally, it is worth pointing out that existential and anaphoric implicit objects may also
differ in regard to their syntax. The syntactic nature of anaphoric implicit objects in BP is not well
established, although there is agreement that they are represented in the syntax: they may be covert
pronouns (e.g., Farrell 1990, Kato 1993) or cases of DP ellipsis (e.g., Cyrino 1996, Cyrino & Lopes
2016). On the other hand, the syntactic status of existential implicit objects is unclear. They are
unable to exert control other syntactic elements (like PRO subjects of non-finite clauses; see Rizzi
1986). This property has been described as Bach’s generalization: “English objects need to be
structurally overt to control non finite clauses” (Bach 1979).
5
Moreover, as will be further
discussed in Section 1.2, existential implicit objects are often considered infelicitous antecedents
for pronouns. These properties (and others – see Williams 2015 for a review) have been argued to
suggest that existential implicit objects are not represented in the syntax at all (e.g., Martí 2011,
Williams 2015). However, another possibility is that existential implicit objects simply have a
‘weaker’ syntactic representation: for example, Landau (2012) proposes two types of implicit
arguments, strong ones (“SIAs”) – like anaphoric implicit objects – and weak ones (“WIAs”). SIAs
are comprised of a ‘full’ syntactic representation, i.e., a determiner phrase (DP) and a bundle of
features. WIAs, in contrast, lack the determiner, and this small difference may help explain the
patterns introduced above without completely discarding WIAs’ role in the syntax.
Although investigating the syntactic status of implicit objects goes beyond the scope of this
dissertation, I will return to this matter in Chapter 6 – General Discussion to reflect on whether the
findings reported in this dissertation may also shed light on implicit objects’ syntactic nature.

5
There are recent patterns in English that go counter Bach’s generalization, such as the implicit object of the verb
‘to counsel’. This is a recent development in English and seems to be an exception restricted to this one particular
verb (Rudanko 2015).
14
1.2. Discourse Behavior: Accessibility and Persistence

Although existential and anaphoric implicit objects are identical in form – both are
unpronounced/unwritten direct objects – they exhibit strikingly different characteristics,
particularly in respect to how they participate in people’s mental models of the discourse.
Notably, the entire wider class of existential implicit arguments – including agents,
instruments and themes (henceforth referred to as implicit objects) – are generally considered to
be discourse “inert” (e.g., Williams 2015). This reputation is due to two main reasons: (i)
existential implicit arguments cannot refer back to an antecedent (as seen, for instance, in (ex.1.2a))
and (ii) existential implicit arguments are not themselves felicitous antecedents for reduced
expressions like pronouns. This latter observation is supported by work on implicit agents of short
passives, which shows that when the agent is implicit, a pronoun like “he” cannot felicitously refer
back to said implicit agent (e.g., (ex.1.10a-b) from Koenig & Mauner 1999):

(1.10) a. The ship was sunk. #He wanted to collect the insurance.

b. The ship was sunk by someonei. Hei wanted to collect the insurance.
It remains unclear whether other existential implicit arguments, like implicit objects,
exhibit the same discourse behavior. Examples often elicit diverse judgements of acceptability: for
instance, Resnik (1993) provides (ex.1.11a) as a felicitous instance of the pronoun “it” referring
back to an existential implicit object (the implicit object of the verb “cook”). In contrast, Williams
(2015) presents (ex.1.11b) as an instance of infelicity regarding the use of the pronoun “it” referring
back to the implicit object of “cook”.
15
(1.11) a. Did you cook ∅ today?
Yes, and it came out delicious.
b. #Al cooked ∅ before it has thawed.
Given their poorly understood discourse status, this dissertation aims to take a closer look
into the discourse behavior of existential implicit objects. Specifically, the current work explores
two aspects of referential behavior in discourse: their accessibility and persistence. In this section,
I introduce both of these notions.
The notion of discourse accessibility is used to refer to referents’ degree of activation in
people’s mental model of the discourse, and how easy or effortful it is to retrieve a given referent
from memory. Recently mentioned referents, for example, are highly activated and easier to
retrieve, relative to referents last mentioned long ago, which are lower in activation and
consequently harder to retrieve (e.g., Prince 1981, Givón 1983, Ariel 1990).
A general pattern has been noted between referents’ degree of activation and the
expressions used to refer back to them. Expressions that are more informative, such as noun
phrases (e.g., “the book”), are generally used to refer back to referents low in activation, whereas
less informative expressions such as pronouns (e.g., “it”) are often used to refer back to highly
activated antecedents (e.g., Ariel 1990, Gundel et al. 1993). As described by Ariel (2001:1), “The
basic idea is that referring expressions instruct the addressee to retrieve a certain piece of given
information from his memory by indicating to him how accessible this piece of information is to
him at the current stage of the discourse.” Thus, pronouns would be viewed as indicating to the
language user that they should find a highly accessible antecedent, and definite noun phrases would
instruct the language user to find an antecedent with a lower degree of accessibility. By assessing
how often and under which conditions implicit objects can be felicitous antecedents for reduced
16
expressions, the current work aims to analyze how their implicitness and other factors contribute
to implicit objects’ accessibility in discourse.
The notion of discourse persistence pertains to how likely referents are to be mentioned
again in subsequent discourse, regardless of the form/referential expression used for this re-
mention. Prior work has observed that not all referents mentioned in discourse are equally
persistent (e.g., Givón 1983, Chiriacescu & von Heusinger 2010). For example, sentence subjects
– which are also often discourse topics – are overall more likely to be re-mentioned in subsequent
discourse than objects (e.g., Givón 1983). In addition to grammatical position, a noun phrase’s
form can also affect the likelihood of its re-mention in subsequent discourse: research on English
indefinite noun phrases suggests that the difference between indefinites headed by ‘a/an’ (e.g.,
“Lisa was reading a book”) and indefinites headed by ‘this’ (e.g., “Lisa was reading this book”) is
that indefinite this signals the speakers’ intentions to further elaborate on that referent in
subsequent discourse (e.g., Prince 1981, Givón 1987, Gernsbacher and Shroyer 1989, Ionin 2006).
This hypothesis has been corroborated experimentally in a sentence continuation task: ‘this’
indefinites are more persistent than ‘a/an’ indefinites (e.g., Chiriacescu 2014).
These patterns illustrate that referents vary in regard to how strong an influence they have
in shaping the direction in which the discourse unfolds. Because of that, persistence has been
proposed to be a core component of referents’ discourse structuring potential (e.g., Chiriacescu
and von Heusinger 2010): a measure of referents’ potential to affect and participate in subsequent
discourse.
This dissertation aims to uncover a more thorough picture of implicit objects’ discourse
behavior by looking at not only their accessibility but also their persistence in discourse.
Accessibility represents a ‘backwards-looking’ approach to discourse: e.g., what kinds of referring
17
expressions are able to felicitously refer back to a previous discourse referent? Prior work on
implicit arguments in discourse has mainly focused on their accessibility, i.e., on this backwards-
looking measure of discourse behavior. On the other hand, persistence conveys a ‘forward-
looking’ perspective in discourse: e.g., what is the likelihood that a referent will be talked about
again as discourse moves forward (regardless of the expression used)? By taking this forward-
looking approach, one can better understand referents’ discourse structuring potential (e.g.,
Chiriacescu & von Heusinger 2010).
Crucially, by adopting both backwards- and forward-looking approaches, Chapters 2 and
3 aim to deepen our understanding about how implicit objects participate in people’s mental
models of the discourse and reveal more about how these various discourse measures interact with
one another. Moreover, these chapters explore the discourse behavior of implicit objects to shed
light on whether they fit the discourse “inertness” label applied to existential implicit arguments
across the board.
1.3. Processing implicit objects

This dissertation also explores how the occurrence of implicit objects affect how people
comprehend language: In particular, I take a closer look at how the interpretation of both context-
independent/existential and context-dependent/anaphoric implicit objects participate in online and
offline processes related to syntactic parsing.
There are several processes that presumably need to take place during the comprehension
of implicit objects, yet very little is known about them. For instance, the language processor
presumably must identify whether a transitive verb has an object that is overt or not. Studies on
English garden-path sentences with object/subject temporary ambiguities like (ex.1.12a) have
18
shown that when a suitable noun phase follows an optionally transitive verb, the processor tends
to interpret said noun phrase as an overt direct object (at least until there is new information
indicating that this is the incorrect parse; e.g., see Late Closure bias, Frazier 1978, Frazier &
Rayner 1982). That is, these garden-path studies reveal that whenever possible the parser favors
processing the verb as having an overt object (e.g., “a book” in (ex.1.12a)) rather than stipulating
that the verb has an existential implicit object.

(1.12) a. After I read a book fell on my foot.

b. Depois que eu recebi a carta e li um livro caiu no meu pé.
After I received a letter and read a book fell on my foot.

However, it is unclear what happens when the implicit object can be interpreted
anaphorically: in the Brazilian Portuguese example (ex.1.12b), the object of li (“read”) can
potentially be interpreted as an overt object (e.g., “a book”) or as an anaphoric implicit object (e.g.,
referring back to “a letter”). Unlike in (ex.1.12a), in (ex.1.12b) both interpretations offer a specific
referent that can be interpreted as the object of “reading”. In this case, does the overt object
interpretation still take precedence over the anaphoric implicit object interpretation? General
processing biases such as Late Closure suggest that the answer is yes – and this is confirmed
experimentally in Chapter 5, which further explores other aspects related to how the interpretation
of implicit objects affects the comprehension of garden-path sentences.
Beyond identifying whether a verb has an overt object or not, the language processor must
also be able to determine whether the verb permits an implicit object interpretation, and which
19
kind. As both existential and anaphoric implicit objects share an ‘invisible form’ despite their
vastly different functions, it is the verb itself that provides this crucial information to the language
parser. In this sense, implicit objects (and implicit arguments more broadly) are quite different
from other expressions in the discourse, such as pronouns: their comprehension is inextricably
linked to their verbs.
Thus, the parser must use information about the verb to assess whether an implicit object
is allowed by that verb, and if so, which one: as previously discussed, in BP, the verb fritar (“to
fry”) allows only anaphoric implicit objects, whereas the verb cozinhar (“to cook”) allows both
existential and anaphoric implicit objects. Furthermore, for a verb like cozinhar, an implicit object
is interpreted anaphorically when there is an available referent; otherwise it is interpreted
existentially. Thus, the language processor must also be able to identify whether or not a suitable
referent has been introduced.
How is the comprehension of implicit objects affected when no suitable referent is
identified? A parallel can be drawn in the world of pronouns: personal singular pronouns like
“he/she” differ from the pronoun “they” in this respect; both can be used to indicate specific
referents, however the former requires a referent to be specified in the context whereas the latter
can also be used in a more generic sense (e.g., institutional “they”; Sanford 2008, Filik et al. 2008).
Studies have shown that when a pronoun like “she” is used within a context that does not provide
a suitable referent, comprehension difficulties ensue (e.g., Van Berkum et al. 2004, Nieuwland and
Van Berkum 2006). In contrast, no difficulties arise when a pronoun like “they” occurs without a
specific referent available for reference (e.g., Sanford 2008, Filik et al. 2008).
Accordingly, if a verb permits existential implicit objects, it may behave more like “they”
pronouns: comprehension should not be affected when no referent is available because these
20
objects can be interpreted completely independently of context. On the other hand, if a verb permits
only anaphoric implicit objects, it may behave more like she/he pronouns: comprehension
difficulties should presumably ensue due to the inability of extracting a value from context to
supply meaning to the implicit object. Chapters 4 and 5 investigate these and related issues
regarding the comprehension of implicit objects.
1.4. Overview of this dissertation

This dissertation investigates the underexplored domain of implicit objects’ discourse
behavior and comprehension. This work’s aim is to assess previous claims about their discourse
“inertness” and shed light on how these formally ‘invisible’ but with varied context-dependency
properties phenomena participate in people’s mental models of the discourse and related
comprehension processes.
Chapters 2 and 3 explore claims related to existential implicit objects’ discourse “inertness”
by taking a closer look at both their accessibility and persistence in discourse. More specifically,
Chapter 2 looks at implicit objects within the specific phenomena of sluicing/sprouting, i.e.,
clauses in which the entire content has been elided with the exception of a wh-phrase that refers
back to content from the previous clause (e.g., “Uthred was reading but I don’t know which book
[he was reading].”). Two acceptability judgement tasks investigated which expressions (e.g.,
“which one” or “which book”) were considered more acceptable when referring back to an
existential implicit object through the lens of accessibility theory. A sentence continuation forced-
choice task then assessed how often people chose to include a reference to a previously-introduced
existential implicit object (relative to a reference to an overt subject); i.e., a measure of implicit
objects’ persistence in sluiced sentences. Chapter 3 continued this line of inquiry by going beyond
21
sluices and using a story-continuation task in which people could freely choose whether or not to
refer to the implicit object. When they did refer to the implicit object, they could also freely choose
any referring expression to do so. Moreover, Chapter 3 also investigated whether a prior context
that helps ‘narrow down’ the identity of an existential implicit object affects its accessibility or
persistence in discourse.
These studies found that more reduced and relative uninformative expressions such as
“which one” are considered less acceptable when referring back to an object that is implicit.
Likewise, people tend to use reduced, ‘uninformative’ expressions such as pronouns less often
when referring back to implicit objects. These findings indicate that existential implicit objects are
indeed less accessible than overt objects. However, these findings also suggest that existential
implicit objects are not entirely “inert” in the discourse: they are still sometimes referred to with
pronouns, and depending on previous context and the expression used, they can be as persistent in
discourse (i.e., likely to be mentioned again) as overt objects. In sum, although existential implicit
objects do seem relatively less accessible and persistent in discourse relative to overt objects, they
are not entirely ‘inert’, and can potentially contribute to the structuring of subsequent discourse.
Furthermore, their discourse persistence can be modulated by other factors, such as previous
context.
Chapters 4 and 5 investigate the comprehension of implicit objects, particularly in regard
to their context-dependency properties. Both of these chapters present experiments conducted in
Brazilian Portuguese, a language with both existential and anaphoric implicit objects. Chapter 4
explores the comprehension of sentences with verbs that have differing requirements regarding
their implicit objects: either the verb allows anaphoric implicit objects only, or the verb allows
both existential and anaphoric implicit objects. Findings from a self-paced reading task reveal that
22
in the first case, comprehension difficulties arise when no suitable referent can be found in context.
In contrast, for the latter case, no difficulties occur. These patterns reveal that the language
processor is able to quickly identify verb subcategorization information regarding the verb’s
implicit object(s) and use this information during comprehension. Chapter 5 looks into how the
interpretation of existential and anaphoric implicit objects affects the comprehension of garden-
path sentences with subject/object ambiguities. A self-paced reading task reveals that the garden-
path effect occurs both with existential and anaphoric implicit objects, i.e., the overt object route
is always preferred – presumably because of Late Closure bias (e.g., Frazier 1978, Frazier &
Rayner 1982). However, once reanalysis occurs, cases in which the implicit object can be
interpreted anaphorically drastically reduce the occurrence of lingering misinterpretations relative
to when the implicit object must be interpreted existentially.
Lastly, Chapter 6 presents a general discussion on the findings and insights introduced in
Chapters 2-5, painting a more thorough picture of how these ‘unsaid’ objects participate in
people’s mental models of the discourse and comprehension processes. This final chapter also
discusses possible future paths for research that may contribute to a deeper understanding of
implicit arguments’ properties and how they affect language processing.

23
Chapter 2 : Existential implicit objects’ accessibility and persistence in
discourse: a sluicing investigation
2.1. Introduction

Keeping track of who and what is being talked about allows people to build mental representations
of the discourse and is a crucial part of language comprehension. For instance, when people
encounter pronouns like “she” or “it”, they must figure out to whom or what these pronouns are
referring. However, many times entities can be introduced to the discourse implicitly via
unpronounced/invisible linguistic phenomena, which can pose a challenge to reference tracking
and comprehension.
In particular, amongst many types of implicit arguments (e.g., Bhatt & Pancheva 2006,
Gillon 2006), verbs can have what are known as existential implicit arguments: these are
arguments specified by the verb’s argument structure that can be omitted. For example, the verb’s
agent is omitted in a passive such as “The food has been eaten” (i.e., by someone – the agent) and
the verb’s theme is omitted in “They already ate” (i.e., they ate something – the theme). Although
these arguments are not explicitly mentioned, their existence is entailed by the verb: the event of
eating entails that there is something being eaten.
The status of such existential implicit arguments in people’s mental models of the discourse
is unclear. It has been suggested that existential implicit arguments are “discourse inert” (e.g.,
Williams 2015), an idea supported by work showing that implicit agents are not accessible enough
in discourse to be picked up by subsequent reference with reduced expressions like “he” (e.g.,
Koenig & Mauner 1999). However, existential implicit arguments are not a heterogeneous
phenomenon; implicit objects differ from implicit agents in various ways, including their syntactic
and semantic statuses. Thus, their discourse accessibility may also differ. In addition, there is more
24
to referents’ life in discourse than their accessibility: other discourse measures such as persistence
can shed light on how referents, including implicit ones, participate in the structuring of discourse.
Thus, this chapter focuses on the underexplored domain of how implicit objects behave in
discourse. In doing so, it aims at a better understanding of how existential implicit objects
contribute to comprehenders’ discourse updating and management, i.e., how we keep track of
entities and events that are introduced and re-mentioned in a conversation.
Specifically, the current chapter investigates two key aspects of referent’s behavior in
people’s mental models of the discourse: accessibility and persistence. Accessibility is typically
seen as a measure of how salient/available a given referent is at a particular point in the discourse
(e.g., Prince 1981, Givón 1983, Ariel 1990). Assuming that there is a relation between referent
accessibility and choice/interpretation of referring expressions, we can ask: once an implicit
argument is introduced in the discourse, what kinds of referring expressions refer back to it?
Traditionally, research on the discourse status of referents has focused on their accessibility
properties. This line of inquiry conveys a ‘backwards-looking’ perspective on discourse relations:
how and when are different referring expressions like pronouns and definite noun phrases able to
refer back to different kinds of previously-mentioned antecedents?
In contrast, discourse persistence is a measure of how likely a given referent is to be
mentioned again in subsequent discourse. Based on how a referent is introduced in the discourse
– e.g., via an overt indefinite noun phrase, a definite noun phrase, or an implicit one –, we can ask:
what is the likelihood that this referent will come up again in the discourse? This line of inquiry
conveys a ‘forward-looking’ perspective on the discourse status of these referents: are speakers
likely to mention them again moving forward in the discourse? Discourse persistence has been
proposed to be a key ingredient of a referents’ discourse structuring potential (e.g., Chiriacescu &
25
von Heusinger 2010). Referents that are highly likely to be mentioned again, as well as likely to
become topics, have a high discourse structuring potential – they are most often the ‘shapers’ of
where the discourse is going. For example, sentence subjects – which are often discourse topics –
are overall more likely to be re-mentioned in subsequent discourse than objects (Givón 1983). In
other words, their higher-ranking grammatical position grants them a higher persistence.
Crucially, accessibility and persistence do not always go hand in hand – an issue that will
be further explored in this chapter as well as in Chapter 3. Some highly persistent referents are not
highly accessible (e.g., Fukumura & van Gompel 2009, Kaiser 2010, Chiriacescu 2014, Brocher
et al. 2016). Here, I combine a backwards-looking approach focusing on accessibility with a
forward-looking approach investigating discourse persistence to capture a more comprehensive
picture of how implicit objects participate in comprehenders’ discourse update and management.
In doing so, Chapters 2 and 3 also aim to provide a deeper understanding of how these different
measures of referents’ discourse status interact and contribute to comprehension processes.
The current work makes use of a particular construction to study implicit objects’ discourse
behavior: sluicing. Sluices are constructions in which an entire clause is elided/omitted except for
a wh-phrase: e.g., “I want to read a book, but I’m not sure which one [I want to read]” (e.g., Ross
1967, 1969, Chung et al. 1995, Merchant 2001). This wh-phrase refers back to a correlate in the
initial clause: for instance, “which one” refers back to “a book”. This link between the wh-phrase
and its correlate allows for the interpretation of the elided content. Relevant to the current work’s
purposes, there are three main reasons that make sluices fruitful constructions for an investigation
on implicit objects’ discourse status: firstly, the wh-correlate can be overt or implicit, which allows
for a comparison between the two possibilities. Secondly, the wh-phrase itself can be reduced,
similarly to a pronoun, or informative, similarly to a noun phrase (e.g., “which one” vs. “which
26
book”). By manipulating the informativity of the wh-phrase, one can obtain a measure of the
correlate’s accessibility to referring expressions in discourse. Finally, sluices display a locality
effect, i.e., people exhibit a bias towards interpreting the wh-phrase as referring back to the most
local correlate (e.g., Frazier and Clifton 1998, Carlson et al. 2009, Harris et al. 2015). Thus, in
sluices we find a construction that generally favors referring back to direct objects relative to
subjects, which is not typically the case with overt pronouns.
This chapter is structured as follows: Section 2.2 introduces the concept of discourse
accessibility and how it relates to referring expressions, including implicit arguments. Section 2.2
also discusses how sluicing constructions can be used to investigate the accessibility of implicit
objects. Section 2.3 presents Experiment 1, an acceptability judgement task that investigates
whether more informative referring expressions are better able to refer back to implicit objects
relative to reduced referring expressions – i.e., a measure of implicit objects’ accessibility in
discourse. Section 2.4 presents Experiment 2, another acceptability judgement task aimed at
investigating if previous context (more specifically, the representational complexity of a set
associated with the implicit object) affects implicit objects’ accessibility in discourse. Section 2.5
introduces the concept of discourse persistence and how it relates to implicit arguments. It also
discusses the association between accessibility and persistence. Section 2.6 presents Experiment
3, a forced-choice sentence continuation task which examines implicit objects’ likelihood of being
chosen to be re-mentioned in subsequent discourse (relative to overt objects) – a measure of their
persistence. Finally, section 2.7 presents a general discussion on the findings reported in this
chapter.

27
2.2. Accessibility and Referring Expressions

The term accessibility is often used to refer to the degree of activation associated with a
piece of information in the comprehender’s mental discourse model and how effortful it is to
retrieve this information. For example, referents that have just been mentioned in the discourse are
high in activation and are thus easier to retrieve, relative to referents that had been mentioned much
earlier on, which are low in activation and are consequently harder to retrieve (e.g., Prince 1981,
Givón 1983, Ariel 1990).
In turn, different referring expressions are associated with different degrees of activation
of their antecedents. “The basic idea is that referring expressions instruct the addressee to retrieve
a certain piece of given information from his memory by indicating to him how accessible this
piece of information is to him at the current stage of the discourse” (Ariel, 2001:1).
For example, referring expressions such as the overt pronouns ‘she’, ‘he’ and ‘it’ in English
are most often associated with antecedents that have a high degree of activation, such as a very
recently mentioned entity in a grammatically prominent position. On the other hand, referring
expressions like definite noun phrases, such as ‘the girl’, ‘the boy’ and ‘the thing’ are more
typically associated with referents that have lower degrees of accessibility, i.e., these referring
expressions are generally used to refer back to entities that have lower activation in the discourse
model. Prior work in this area has revealed an overarching pattern in the association of referring
expressions and the accessibility of their antecedents: namely, that reduced, ‘uninformative’
6

forms like null and overt pronouns are most often used to refer back to highly accessible referents,
while heavier, more informative expressions such as noun phrases tend to be used for less

6
In this work, I refer to these expressions as ‘uninformative’ for ease of comparison to other relatively more
informative expressions.
28
accessible referents
7
.
The precise underpinnings of which factors contribute to the perceived accessibility of a
referent and how people are able to determine a certain referent’s degree of activation in order to
choose the most appropriate referring expressions remain an active and fruitful field of research.
The current chapter investigates where implicit objects fall on the accessibility spectrum by
assessing which referring expressions are better able to refer back to them. Moreover, this chapter
also investigates whether other factors, such as prior context, may modulate implicit objects’
accessibility in discourse.
2.2.1. The accessibility of implicit arguments

To what degree is information that is not overtly expressed activated in people’s mental
models of the discourse? Intuitively, entities that are introduced implicitly seem less accessible
than those introduced overtly. Consider Partee’s famous marbles example (cited in Heim 1982):
(2.1) I dropped ten marbles and found only nine of them. #It is probably under the sofa.

(2.2) I dropped ten marbles and found all of them, except for one. It is probably under the
sofa.

In (ex.2.1), the pronoun it cannot felicitously refer back to the one marble that was not
found – despite the fact that the one unfound marble ‘feels’ highly salient. However, notice that

7
Here I use the terms ‘reduced’ and ‘heavier’ as traditionally used in the field of linguistics to refer to an expression’s
“size”: expressions that are short (typically zero or one word) are considered reduced, while expressions that are
generally longer and can be expanded with modifiers are considered heavier. Stress and other prosodic patterns can
also play a role in this definition, but that is beyond the scope of this paper (see Ariel 2000 for a review).
29
this one unfound marble was not overtly introduced into the discourse. Once that marble is
mentioned overtly, as in (ex.2.2), it can be felicitously referred to with a pronoun in the next
sentence.
These examples suggest that even when a referent may seem prominent in discourse, if it
was not overtly linguistically expressed, it is not accessible enough for subsequent pronominal
reference. However, not all implicitly-introduced entities are equally inaccessible to pronominal
reference. Cornish et al. (2005) note that in (ex.2.3), the implicit entity related to being pregnant
can be easily referred to with a pronoun.

(2.3) Barbara is six months’ pregnant, and she’s already knitted a bonnet and gloves for it.

One of the reasons why we may see differences in the behavior of implicit arguments and
how accessible they are in the discourse is that the label ‘Implicit Arguments’ – henceforth IAs –
is an umbrella term for various kinds of nominal and verbal arguments that are non-overt, i.e., not
pronounced/written (see Bhatt & Pancheva 2006 for a review). The unfound marble and the unborn
baby share their implicitness in discourse but may differ in many other features including their
semantic, syntactic and discourse statuses.
The current work focuses on a particular kind of implicit arguments, introduced in Chapter
1: existential IAs, which are those that are introduced to the discourse through an existential
entailment of a predicate.
8
Existential IAs are traditionally viewed as ‘discourse inert’. Evidence

8
Remember that existential IAs contrast with anaphoric IAs (e.g. “Lisa was anxious. Marge noticed Ø.), which do
not introduce anything to the discourse, but rather refer anaphorically to something/some event that has been
previously introduced. Existential and anaphoric IAs are also referred to as indefinite and definite IAs (see Fillmore
1986, Condoravdi & Gawron 1996, Anderbois 2012). This chapter and Chapter 3 focus solely on existential/indefinite
implicit objects, however Chapters 4 and 5 of this dissertation examine both types.
30
for this comes from experimental work on implicit agents of short passives (ex.2.4a), which reveals
that despite being able to syntactically control the PRO subject of a non-finite clause (ex.2.4b),
these agent IAs cannot felicitously serve as antecedents of intra-sentential pronouns ((ex.2.4c);
examples from Koenig and Mauner 1999):

(2.4) a. The ship was sunk.

b. The ship was sunk to collect the insurance.

c. The ship was sunk. #He wanted to collect the insurance.

In (ex.2.4c), despite the fact that the existence of an implicit agent is entailed by the passive
(i.e. there is an x such that said x has sunk the ship), this implicit agent is generally considered to
be an awkward antecedent for the subsequent pronoun ‘he’. This is especially surprising
considering that the implicit agent of passives is able to control non-infinitival clauses, displaying
some degree of syntactic presence/activation.
Compare (ex.2.4c) to its long passive paraphrase with an overt agent in (ex.2.4d):

(2.4) d. The ship was sunk by someonei. Hei wanted to collect the insurance.

Here, “he” felicitously refers back to the agent overtly introduced by “someone”. The
striking difference between (ex.2.4c) and (ex.2.4d) has contributed to the belief that these implicit
31
arguments are discourse inert. Indeed, Koenig and Mauner (1999) proposed that implicit agents
belong in a class termed “A-definites.”
According to Koenig and Mauner, a-definites crucially differ from both indefinite and
definite noun phrases in that, unlike the others, a-definites lack the ability to introduce a new
referent to the discourse model (also sometimes referred to as introducing a new ‘discourse marker’
in the sense of Heim 1982). Consequently, a-definites are inaccessible for future reference, or at
least, for future reference with a pronoun (a marker of high accessibility), as shown in (ex.2.4c)).
Koenig and Mauner also propose that reference to implicit arguments can be achieved more
felicitously with definite noun phrases (rather than pronouns) via inference/accommodation: the
morphosyntactic cue provided by the definite determiner, along with the informative content of
the noun, allows reference to these less accessible referents (in this case, implicit arguments).
Interestingly, not all expressions that have been proposed to be A-definites are implicit: the
French subject clitic on (‘one’) and predicate modifier noun phrases in Hungarian also appear not
to generate a discourse referent, at least not one that can felicitously be referred back to with a
pronoun (e.g., Koenig 1999, Farkas 1997).
Based on Koenig and Mauner (1999)’s findings, one may wonder: do all implicit
arguments lack the ability to be referred back to with a pronoun, despite other discourse factors?
Are all implicit arguments a-definites, in the sense that they simply can’t introduce a highly
accessible referent in the discourse?
As previously discussed, the label ‘implicit arguments’ is used for a wide variety of
linguistic phenomena which share the single defining characteristic of not appearing overtly (e.g.,
Bhatt and Pancheva 2006). It is possible that this implicitness itself renders existential implicit
arguments inaccessible to referring expressions that typically require their antecedent to be highly
32
accessible (i.e., pronouns in English); if so, research into other types of implicit arguments beyond
implicit agents should reveal that they too are not acceptable as the antecedents of reduced, less
informative referring expressions. On the other hand, examples like (ex.2.3) suggest that there’s
at the very least some variability in the degree of accessibility that various implicit arguments can
display and how acceptable they are as the antecedents of pronouns. Implicit objects differ from
implicit agents in various respects; for example, the verb “bake” provides more information about
its implicit object (e.g., it’s likely a typical baked good, not any other type of food/object) than its
implicit agent (e.g., it’s someone who bakes, but baking is a common activity). It is possible that
these differences – e.g., the degree to each their identities are ‘narrowed down’ – contribute to a
difference in their degrees of accessibility in people’s mental models of the discourse.
By extending this line of inquiry beyond implicit agents, I aim to more clearly identify the
role of implicitness in determining existential referents’ statuses in the discourse, as well as
identifying other discourse factors, such as prior context, that may contribute to implicit
arguments’ discourse status.
2.2.2. What about implicit objects?

The current work expands the breath of research on the discourse status of implicit
arguments by focusing on the underexplored domain of implicit direct objects. In English, these
object IAs arise within a class of verbs known an Optionally Transitive verb such as “read”,
“bake”, “eat”, etc. These verbs can happily appear with overt direct objects (ex.2.5a, 2.6a, 2.7a) or
with implicit ones (ex.2.5b, 2.6b, 2.7b; as in Chapter 1, the symbol Ø will continue to be used
throughout this dissertation to indicate an implicit argument).

33
(2.5) a. Lisa was reading a book.

b. Lisa was reading Ø.

(2.6) a. Marge baked her famous apple pie last night.

b. Marge baked Ø last night.

(2.7) a. Bart is eating the last chocolate chip cookie in the kitchen.

b. Bart is eating Ø in the kitchen.

Implicit objects are often grouped with implicit agents and all other implicit arguments due
to their apparent ‘invisibility’, but the members of this group do not necessarily display
homogenous semantic, syntactic or discourse statuses. There are various ways in which implicit
objects and agents differ that may affect their accessibility in the discourse model.
One previously mentioned difference is in their syntactic presence
9
: in English, while
implicit agents are able to control other syntactic elements (e.g., PRO; see (ex.2.4b)), implicit
objects are thought not to share this ability. For example, the object of the verb lead needs to be
overt in order to control; see the contrast between ((ex.2.8a-b), from Rizzi 1986):

9
As noted in Chapter 1, there is still debate as to what the ability to control PRO means exactly in regard to the
syntactic status of these elements and whether they are structurally present in the syntax or just somehow syntactically
active; nonetheless, the crucial difference is that implicit agents differ from implicit objects in this regard.
34
(2.8) a. *This leads Ø [PRO] to conclude what follows.

b. This leads people [PRO] to conclude what follows.

This property has been described as Bach’s generalization: English objects need to be
structurally overt to control non finite clauses (Bach 1979). This has been taken to suggest that
implicit agents of passives, but not implicit direct objects, have a syntactic presence.
10

Another notable difference between implicit objects and implicit agents is that only the
former is restricted to a limited set of verbs. Whereas existential object IAs arise only with
Optionally Transitive verbs in English, implicit agents appear with any verb in a short passive
construction. This is a relevant property because experimental studies on other topics (not on
implicit arguments) have shown that arguments that are specific to restricted sets of verbs – such
as instrument arguments for verbs that obligatorily require instruments, like “behead” – are
activated as part of the verb’s argument structure and facilitate later reference to the argument
(e.g., Koenig at al. 2003). This suggests that existential implicit objects, which are restricted to a
specific set of verbs in English, may be more strongly activated relative to implicit agents.
Finally, it is worth mentioning that the semantic content of implicit objects itself is
restricted, unlike the semantic content of implicit agents. Any animate human can potentially be
the agent of a short passive, but the interpretative possibilities of an implicit object are lexically
restricted: for example, “I ate” entails “I ate something typically edible/a meal”. Anything that is
not typically edible/a meal, such as “succulents”, “bears” or “nails” would be a highly implausible

10
There are recent patterns in English that go counter Bach’s generalization, such as the implicit object of the verb ‘to
counsel’. This is a recent development in English and seems to be an exception restricted to this one particular verb
(Rudanko 2015).
35
interpretation for the implicit object of eating (e.g., Fillmore 1986, Mittwoch 1982; see Martí 2011
for more on the typicality of implicit objects in English and their similarity to noun incorporation
structures). In a sense, although existential entailments are underspecified by default – at a
minimum, they just entail the existence of an x, nothing about its identity or features – implicit
agents may be even more underspecified than implicit objects.
By extending the research on implicit arguments’ discourse status to implicit objects, we
aim to shed light on how implicitness itself, as well as other factors that vary amongst different
kinds of implicit arguments, may affect how these referents participate in comprehenders’ mental
discourse model.
2.2.3. Investigating the accessibility of implicit objects using sluicing

To probe into implicit objects’ status in the discourse model, this chapter steps away from
the most common tools on accessibility research – pronouns – and towards a possibly better suited
construction for our purposes: sluicing.
Prior work on accessibility has often relied on pronouns to assess referents’ accessibility,
and they have become a useful staple in this domain’s experimental tool shed. However, pronouns
– in accordance with their association with highly accessible antecedents – have been known to be
most often associated with preceding subjects (e.g., Chafe 1976, Crawley & Stevenson 1990,
Kaiser 2011), a grammatical position which itself indicates topicality and high accessibility in
English (e.g., Chafe 1976, Givón 1983, Brennan et al. 1987). Pronouns do sometimes display an
increased object bias, particularly in parallel constructions in which the antecedent and the pronoun
are both objects (e.g., Grober et al. 1978, Chambers & Smyth, 1998). However, even in these cases
36
implicit antecedents seem to be awkward pronoun antecedents, as in “Lisa cooked Ø last night. I
ate it this morning” or “Lisa was reading Ø on the balcony. Maggie had written it.”
Thus, English pronouns may not be the most informative tool to investigate the
accessibility of object referents: we may see differences amongst various object referents be
washed away due to comprehenders’ general preferences to see pronouns linked to the subject, not
the object. Luckily, English offers a new tool that allows us to tap into the accessibility of object
referents without fear of a subjecthood bias: sluicing constructions (see Chapter 3 for an
investigation that does look into pronouns and other referring expressions).
Sluicing is a syntactic phenomenon involving ellipsis, in which an entire clause is
elided/not overtly expressed except for a wh-phrase, called ‘the remnant’ (e.g., Ross 1967, 1969,
Chung et al. 1995, Merchant 2001). Example (2.9) contains a sluicing construction: the clausal
complement “what he was eating” is mostly elided, and all that remains of the clause is the wh-
phrase “what”.

(2.9) Homer was eating something but I don’t know what [Homer/he was eating].

How are comprehenders able to ‘reconstruct’ the elided component of this sentence? That
is, how do we understand that in (ex.2.9) the thing that the speaker does not know is what Homer
was eating? The exact underpinnings of how comprehenders are able to ultimately interpret sluices
is still under investigation. However, it is uncontroversially accepted that the crucial component
that allows comprehenders to ‘reconstruct’ the missing parts of the sluice is its relationship with
the antecedent clause (e.g., Ross 1967, 1969, Chung et al. 1995, Merchant 2001). In (ex.2.9), the
antecedent clause is “Homer was eating something”. When comprehenders arrive at the sluiced
37
constituent – “…but I don’t know what” – they can identify that the wh-remnant what has a
correlate antecedent in the antecedent clause: “something”. It is this identity-matching relationship
between the wh-phrase in a sluice and its correlate in the antecedent clause that allows us to
reconstruct the missing pieces: “what” must be “what [Homer was eating].”
Crucially for our purposes, this wh-phrase can have either an overt correlate on the
antecedent clause (like in (ex.2.9)), or it can have an implicit correlate ((ex.2.10); sluicing with
implicit antecedents is also known as ‘sprouting’):

(2.10) Homer was eating but I don’t know what [he was eating].

One main property of sluices makes them particularly suited for investigating the
accessibility of object referents: sluices are known to display a locality effect. That is, when faced
with ambiguous sluices – in which the wh-remnant phrase could refer back either to a subject or
an object in the antecedent clause – comprehenders tend to choose the most local antecedent, i.e.
the object (e.g., Frazier and Clifton 1998, Carlson et al. 2009, Harris et al. 2015). For example, in
(ex.2.11) (from Harris et al. 2015) the wh-remnant “which” or “which ones” could refer back to
either the subject “a few linguists” (option a) or the object “some silly examples” (option b), and
people display a bias towards the latter option.

(2.11) [A few linguists] gave [some silly examples], but I don't remember which (ones).

a. …I don't remember which linguists. [Subject correlate]

38
b. …I don't remember which examples. [Object correlate]

Hence, sluices may be a more appropriate construction to look into the discourse status of
object referents, particularly if there are any fine-grained distinctions in the accessibility degree of
different types of objects (for example, between overt and implicit objects).
Another property of sluices that makes them a suitable tool for investigating accessibility
is the possibility of manipulating the nature of the referring expression, the wh-phrase. As
previously discussed, accessibility work suggests that reduced, less informative expressions like
pronouns are most often used for highly accessible antecedents. There is only one kind of wh-
phrase in English that can vary in informativity: which phrases. Unlike their other wh-cousins,
which phrases can be less (ex.2.12a)
11
or more (ex.2.12b) informative:

(2.12) a. Homer was reading a book but I don’t know which one [Homer was reading].

b. Homer was reading a book but I don’t know which book [Homer was reading].

Thus, which phrases allow us to use sluicing minimal pairs in investigating the accessibility
of implicit objects: We can compare a less informative referential expression like which one to a
more informative expression such as which noun. Which one phrases resemble pronouns in that
they do not provide any semantic information about the identity of the referent
12
, whereas which

11
Which phrases can also be bare, as in “Homer was reading a book but I don’t know which.” – they will not be further
discussed in this work, but see Harris (2015) for experimental work involving these.
12
The expression which one involves one anaphora, which differs from ‘regular’ pronouns in various respects (see
e.g. Hankamer and Sag 1976, Gagnon 2013). The details of the broader semantic and syntactic behavior of one
anaphora are beyond the scope of the present study. Crucially, Gagnon (2013) shows that one anaphora occurs with
overt as well as implicit antecedents, and Harris (2015) shows that which phrases containing one anaphora also display
the locality effect in sluices.
39
noun phrases are more informative as they identify the semantic category of the referent (e.g.,
book).
Using sluicing constructions with which phrases also allows us to exploit another
phenomenon to investigate implicit objects’ accessibility: discourse-linking (d-linking). Which
phrases are a particular case of wh-phrases that are interpreted in reference to the discourse context:
comprehenders tend to interpret which phrases in reference to a salient context set (e.g., Pesetsky
1987, 2000).

Discourse-Linking Implicit Arguments

By virtue of using discourse-linked sluicing constructions, the current work also taps into
another backward-looking component of implicit objects’ discourse status: namely, whether they
are able to anaphorically refer back to a previously-mentioned referent. So-called existential
implicit objects in English are generally thought to not be able to anaphorically co-refer with an
antecedent. In fact, this is the key distinguishing feature between existential implicit arguments,
like the ones studied in this work, and anaphoric implicit arguments such as the ones that can
occur with verbs like ‘notice’ and ‘find out’ (see Fillmore 1986, Condoravdi & Gawron 1996,
Anderbois 2012 for more on the differences between existential and anaphoric implicit arguments,
see also Chapter 1). For example, when the object of ‘eat’ is overt (ex.2.13a), it can corefer with
the previously mentioned referent “my banana”, but not when it is implicit (ex.2.13b).
(2.13) a. A: Where’s my bananak? B: Lisa ate itk.

b. A: Where’s my bananak? B: #Lisa ate Øk.
40

In d-linked constructions, however, there is a different kind of anaphoric relationship
between a referring expression and the previous context: not a direct, coreferential relationship of
identity as in (ex.2.13a), but rather a domain restriction on interpretation. In (ex.2.14a-b), the set
‘some books’ is introduced in the first sentence and then the which phrase in the sluice establishes
a d-linking relation to the previously-mentioned set of books (e.g., Pesetsky 1987, 2000). That is,
people can interpret which one/book in this context in reference to the discourse salient set of
“some books”.

(2.14) a. I gave Homer some books. Later, I saw him reading a book, but I’m not sure
which one/which book.

b. I gave Homer some books. Later, I saw him reading Ø, but I’m not sure which
one/which book.

I assume that, as a consequence of d-linking, the antecedent of the d-linked which-phrase
is restricted to the domain of the salient context set. This may seem surprising given implicit
arguments’ resistance to anaphoric interpretations. However, overt indefinites also resist anaphoric
interpretations, i.e., they generally introduce a new discourse referent (e.g. Heim 1982). Despite
this resistance to anaphoric coreference, examples like (ex.2.14a-b) suggest that both overt
indefinites and existential implicit arguments can have their interpretation restricted to a salient
context set. This assumption, to the extent of my knowledge, has not yet been empirically tested.
To verify it, a secondary task was included in Experiment 2.
41
This d-linking property allows us to control the specificity of the object and keep the
constructions we are comparing as close to a minimal pair as possible: in these discourse-linked
sluicing constructions, both the overt object and the implicit object are interpreted in reference to
the salient context set (e.g., as “one of the books”), so there is little difference in their specificity,
only in their implicitness/overtness.
In sum, Experiments 1 and 2 investigate the accessibility of implicit objects using
discourse-linked sluicing constructions. These constructions provide a novel, well suited tool for
this task because (i) correlates can be overt or implicit; (ii) unlike pronouns, sluice remnants have
a locality bias in regards to their antecedents, i.e., an object preference in English and (iii) which
sluices allow the use of two types of referring expressions to test the accessibility of their
antecedents: a pronoun-like, uninformative which one and a noun-phrase-like, more informative
which noun. Moreover, by using discourse-linked sluices, the interpretation/specificity of overt
and implicit objects can be kept as uniform as possible.
2.3. Experiment 1: Are more informative referring expressions better able to refer back to
implicit objects?

This experiment aims to investigate implicit objects’ accessibility in discourse. To do so,
it tests whether the acceptability of implicit objects as antecedents in sluicing constructions is
modulated by the form of the referring expression. More specifically, this experiment investigates
whether more informative referring expressions are better able to refer back to implicit objects.
Previous research on pronoun resolution and accessibility suggests a strong relationship
between (i) the accessibility of a referent and (ii) the form of the referring expression used to
retrieve said referent (aka the antecedent). Generally, referents that are deemed low in accessibility
are referred to with more informative expressions; and conversely, referents that are high in
42
accessibility are typically referred to with less informative expressions (e.g., Givón 1983, Ariel
1990, Gundel et al. 1993). Therefore, if implicit objects are lower in accessibility relative to overt
objects, I expect that comprehenders may find more informative expressions (e.g. noun phrases)
more acceptable than less informative expressions (e.g., pronouns) for referring back to implicit
objects.

Participants

57 native speakers of English participated in this experiment. 2 participants were excluded
from analysis for not being native English speakers, and 5 additional participants were excluded
for poor performance on comprehension questions (below 80% accuracy). Responses from the 50
remaining participants were analyzed.
2.3.1. Methods

The current experiment uses discourse-linked ‘which’ sluices as a tool to assess the
accessibility of object antecedents. The experiment was programmed on Qualtrics and conducted
online through Amazon Mturk. Participants were asked to read sentences and evaluate how
acceptable each item was in a 5-point scale from 1 – “Completely Unacceptable” to 5 –
“Completely Acceptable”.

43
Materials and Design

The experiment was comprised of 50 items: 20 target and 30 fillers items. The 30 filler
items were designed to vary in acceptability so that participants would most likely use the full
range of the scale.
13
In addition, 10 comprehension questions unrelated to the critical
manipulations were presented throughout the experiment to verify that participants were indeed
reading the items.
In target conditions, (i) object type (overt or implicit) and (ii) wh-remnant phrase type
(“which one” or “which noun”) were manipulated. 20 target items were designed, using a set of
20 optionally transitive verbs in English such as “read”, “bake”, “study”; see complete list in
Appendix 1). All target items started with a sentence introducing a context set (e.g., “some
books”); then a sentence in which an OPT verb would appear either with an overt (e.g., “reading
a book”) or an implicit object (e.g., “reading”); and ended in a discourse-linked sluice with either
an informative wh-phrase (e.g., “which book”) or an uninformative one (e.g., “which one”). See
Table 2.1 for examples of target items in each condition.

Table 2.1 Experiment 1: Conditions and Target item examples
CONTEXT SENTENCE: Bob gave Sarah several books to read.
Implicit + which one Later, he saw her reading, but he doesn't know which one.
Implicit + which noun Later, he saw her reading, but he doesn't know which book.
Overt + which one Later, he saw her reading a book, but he doesn't know which one.
Overt + which noun Later, he saw her reading a book, but he doesn't know which book.
Ungrammatical control Later, he saw her sneezing, but he doesn't know which one.

13
A later analysis revealed that this was indeed the case (ratings for filler items varied from 1 to 5 in the expected
fashion).
44
In addition to the manipulations, an ungrammatical control condition was included to
provide a baseline of unacceptability. In this condition, the critical verb was changed from an
optionally transitive verb to a true intransitive verb, i.e., verbs that generally have no slot for a
direct object (overt or otherwise) in its argument structure (e.g., “sneeze”, “sleep”; see full list in
Appendix 2). In these ungrammatical controls there is no available correlate for the wh-sluice
remnant in the antecedent clause; consequently, I expect that people will consider these sentences
unacceptable.
2.3.2. Results

Raw acceptability ratings are shown in figures and tables for ease of presentation (Table
2.2, Figure 2.1). However, statistical analyses were conducted on z-scores. Participants’ ratings
were transformed into z-scores to minimize possible issues such as scale compression or skew
(e.g., Schütze and Sprouse 2014). Linear mixed effects analyses of z-scores were performed with
lmer using R (R Core Team, 2017). Object type and Wh-remnant type (and an interaction term)
were included as fixed effects. Object type was coded -1 (implicit), +1 (overt) and Wh-remnant
type was coded -1 (‘which one’), +1 (‘which noun’), with random slopes for subjects and items
when supported by the data (e.g., see Barr 2013).
Findings reveal that implicit objects are indeed less accessible than overt objects for
future discourse reference made with reduced expressions such as “which one”. Participants
rated conditions with implicit objects as less acceptable than conditions with overt objects, but
only when the wh-phrase was reduced/uninformative (e.g., “which one”; see Table 2.2, Figure
2.1).
45
Strikingly, when the wh-phrase was informative (e.g., “which book”), there was no
difference between participants’ ratings for implicit and overt objects. That is, overt and implicit
objects do not differ in their accessibility as antecedents of informative referring expressions.
Statistical analyses reveal significant main effects of object and wh-phrase type, as well as
a significant interaction (Table 2.3); planned pairwise analyses show a significant main effect of
object type on conditions with uninformative wh-phrases, but not with informative ones (Table
2.4).
These findings contribute to a body of work examining the discourse status of existential
implicit arguments by taking a closer look at the underexplored domain of implicit objects. As
with other implicit arguments, implicit objects were found to be relatively less accessible than their
overt counterparts.

Table 2.2 Experiment 1: Raw means and standard error of participants’ acceptability scores by object type
and wh-phrase type (5-point scale)

which one which noun
Acceptability Score SE Acceptability Score SE
Implicit object 3.85 0.083 4.45 0.060
Overt object 4.66 0.048 4.59 0.053

Table 2.3 Experiment 1 Results: Main effects and interactions (“*”: p < 0.05)
Object Type Wh-Phrase Type Object Type*Wh-Phrase Type
b SE t b SE t b SE t
0.144 0.028 4.990* 0.076 0.021 3.467* -0.098 0.019 -5.201*

46
Table 2.4 Table 2.4. Experiment 1 Results: Planned pairwise comparisons (“*”: p < 0.05)
Contrasts
Object Type
b
SE t
Uninformative wh-phrase
“which one”
0.024 0.044 5.462*
Informative wh-phrase
“which noun”
0.045 0.028 1.594

Figure 2.1 Experiment 1: raw means and +-1 SE by object type and wh-phrase type (5-point scale)

An additional analysis contrasting the ungrammatical control condition and conditions with
implicit objects and uninformative wh-phrases was also conducted: ungrammatical controls were
rated lower in acceptability when compared to items in which implicit objects were antecedents of
uninformative which phrases (Tables 2.5, 2.6 and Figure 2.2).

Table 2.5 Experiment 1: Raw means and standard error of participants’ acceptability scores in ungrammatical
control comparison (5-point scale)
Acceptability Score SE
Implicit object + “which one” 3.85 0.083
47
Ungrammatical control 1.27 0.048

Table 2.6 Experiment 1 Results: Ungrammatical control comparison (“*”: p < 0.05)
Contrasts b SE t
Ungrammatical control vs.
Implicit object +
“which one”
-0.771 0.024 -32.121*

Figure 2.2 Experiment 1: Raw means and +-1 SE in ungrammatical control comparison

2.3.3. Discussion

Experiment 1 investigated implicit objects’ accessibility in the discourse model by
comparing the acceptability of overt and implicit objects as antecedents of informative and
uninformative referring expressions in sluicing constructions.
In sluices with uninformative remnants (namely which one), implicit objects are rated less
acceptable than overt objects. This finding goes hand in hand with previous observations that
48
implicit objects generally do not seem to be acceptable antecedents for pronouns, which are also
‘uninformative’ referring expressions.
In light of the broad generalization that reduced, less informative referring expressions are
most felicitous when referring back to highly accessible referents, these findings indicate that
implicit objects are not highly accessible referents, or at least not as highly accessible as overt
objects.
Notably, however, when the referring expression is informative – such as “which book” –
people judged implicit object antecedents to be just as acceptable as overt antecedents. This
supports the idea that more informative which-expressions are better able to refer back to implicit
arguments. In addition, it further suggests that implicit and overt objects may not differ drastically
in accessibility, as no differences were found in this condition; i.e., the data shows no cost for
reference to implicit antecedents.
One may be concerned that conditions with informative remnants could be ‘too repetitive’:
Prior work suggests that sometimes the repetition of redundant information can lead to processing
difficulties; for example, repeating a name when a pronoun would suffice can produce a burden in
comprehension (“repeated name penalty”; e.g., Gordon et al. 1993, Almor 1999, Garrod et al.
1994). The condition with overt objects and informative wh-phrases contains name repetition, i.e.,
the correlate noun is repeated in the wh-phrase (e.g., “a book… which book”). In principle, it is
possible that due to the repeated name penalty, this condition could have been judged as having a
lower acceptability than it would if the name repetition had been avoided. This possibility is not
supported by the results: sentences with overt objects and uninformative wh-phrases (e.g., “a
book… which one”), which do avoid the name repetition, do not significantly differ in
acceptability.
49
Why don’t we see a repeated name penalty in the acceptability of sluices in which the wh-
phrase repeats the correlate noun? Dickey & Bunger (2011:75) argue that “any penalties associated
with repeating the full form where ellipsis is possible may be outweighed by the processing
advantages derived from encountering that overt material”. Their results suggest that in cases like
sluicing – where the ellipsis site must be reconstructed for interpretation – this ‘redundant’ material
is actually advantageous during comprehension.
2.4. Experiment 2: Does representational complexity affects the accessibility of discourse-
linked implicit objects?

Experiment 1 revealed that implicit objects are not as acceptable antecedents for
uninformative wh-remnants (which one) as overt objects are, but that this asymmetry disappears
with informative wh-remnants (which noun). These findings suggest that implicit objects’ degree
of accessibility is more compatible with the use of informative referring expressions. In
Experiment 2, I investigate whether it is possible to increase the degree of accessibility of implicit
objects by manipulating the representational complexity of the context set associated with the
interpretation of the implicit object. Let us consider why one might expect that an increase in
representation complexity might have this effect:
Prior work on long distance filler-gap dependencies suggests that the representational
complexity of an antecedent is linked to the acceptability and ease of processing the referring
expression: the more representationally complex, the more acceptable and easy to retrieve is the
antecedent (e.g., Hofmeister 2011). An expression is more complex than another when the
semantic and syntactic information of the latter is a proper subset of the former. For instance,
“some rare old books” is more representationally complex than “some books”, because the latter
is a subset of the former semantic and syntactic-wise. The core idea behind this work is that more
50
complex referents require more effort in the initial encoding and are thus more strongly encoded
in memory, which in turn makes them easier to access at subsequent retrieval cues.
For example, Hofmeister (2011) compared (ex.2.15a-c) in a self-paced reading experiment
and found that reading times immediately following the retrieval site (the gap after the verb
“banned”) were faster for more complex antecedents (ex.2.15c) than those of medium complexity
(ex.2.15b) which in turn were read faster than simple antecedents (ex.2.15a).

(2.15) a. It was a communist who the members of the club banned from ever entering the
premises.

b. It was an alleged communist who the members of the club banned from ever
entering the premises.

c. It was an alleged Venezuelan communist who the members of the club banned
from ever entering the premises.

While Hofmeister (2011) offered a memory-based account of the effects found in their
data, Frazier and Clifton (2011) present a different perspective. They looked at filler-gap
dependencies like (ex.2.16a-b), in which there is a discourse-linked dependency but no interveners
between the filler (in this case, the wh-phrase) and the gap (the object position inside the elided
constituent in the sluice). Representational complexity of the wh-phrase was manipulated by
comparing bare what with more representationally complex which noun phrases. They found the
same representational complexity advantage as Hofmeister: sentences were rated as more
51
acceptable in the complex which condition than in the bare what condition. However, because
there are no interveners between the filler and the gap in these dependencies, they argue these
findings are possibly not due to memory effects, but rather other reasons, such as an increased
accessibility in the discourse for more complex phrases.

(2.16) a. Britney likes this guy who destroyed a new vehicle but she didn’t reveal what.

b. Britney likes this guy who destroyed a new vehicle but she didn’t reveal which
one/which vehicle
14
.

Frazier and Clifton (2011) further suggest that discourse-linked which phrases may be
generally preferred over other wh-phrases precisely due to their discourse-linking property, which
may raise the referent’s accessibility in the discourse. However, not all of Hofmeister’s studies
used which phrases (as shown in (ex.2.14a-c)). Thus, it remains unclear whether memory effects,
discourse prominence, or both are at play.
In Experiment 2, representational complexity is manipulated to investigate whether it
modulates the acceptability of implicit objects as antecedents of uninformative anaphoric
expressions such as which one. As prior work has shown that increased representational
complexity can facilitate the processing and increase the acceptability of discourse-linked
dependencies (through increased discourse accessibility or other mechanisms), I hypothesize that

14
Frazier and Clifton’s study used both uninformative which one and informative which noun phrases in their complex
condition without controlling for informativity, but they report post-hoc analyses revealing the patterns for each of
these to be the same.
52
it may also make the accommodation process between the uninformative wh-phrase which one and
the implicit object antecedent less burdensome and more acceptable to comprehenders.
Crucially, the very nature of implicit objects makes it impossible to manipulate their
representational complexity directly, as they cannot be modified by other linguistic material (e.g.,
adjectives, relative clauses, etc.). However, in this set of experiments we are dealing with implicit
objects that are part of a discourse-linking relationship with a previously mentioned, salient context
set. Thus, the representational complexity of the context set itself was manipulated. In harmony
with conditions set in Hofmeister (2011), representational complexity was manipulated by taking
simple noun phrases (e.g. “some books”) and adding two adjectival modifiers (e.g. “some rare old
books”), making the former a proper semantic and syntactic subset of the latter.
If representational complexity of the context set affects the resolution of discourse-linked
dependencies that refer to the set, I expect that conditions with complex sets will increase
acceptability across the board for both implicit and overt object antecedents.

Secondary Task: Assessing the interpretation of discourse-linked objects

In Experiment 2 a secondary, forced-choice task was included after the main task to
confirm whether comprehenders ultimately do reach a discourse-linked interpretation of the
critical sentences. That is, this task was designed to assess whether participants inferred that the
implicit and overt objects of the main verb (e.g., “reading”) is a member of the salient context set
(e.g., “several books”; see (ex.2.17) below). We present the results of this secondary task before
turning to the results of the main experiment in order to address potential concerns regarding the
53
interpretation of target items. As will become clear below, the secondary task confirms that
participants interpreted the items as intended.
In addition to allowing us to check participants’ interpretation of the overt and implicit
objects in these discourse-linked constructions, this task lets us to take a closer look at another
‘backwards-looking’ property of implicit arguments: their capacity to interact with prior context.
As mentioned above, the implicit arguments investigated in this work are all existential, i.e., they
generate an existential entailment. As such, one of their properties is that they resist coreferential
anaphoric interpretations: implicit objects are generally infelicitous when coreferential with a prior
referent (as described in Chapter 1).
However, in the discourse-linked constructions used here, the implicit object is anaphoric
in a different sense: its interpretation is restricted in reference to a previous salient context set. In
this secondary task, we confirm whether this discourse-linked interpretation does indeed occur for
implicit objects, and whether it is comparable to overt objects’ discourse-linking.
After participants completed the main part of the experiment (i.e., acceptability judgments,
described below), they completed the secondary task. It consisted of the same target items from
the main experiment accompanied by a new set of 30 filler items. After reading the items,
participants answered comprehension questions such as in (ex.2.17):

(2.17) Bob gave Sarah several books/several rare old books to read.
Later, he saw her reading/reading a book, but he doesn't know which one.

Q: What was Sara reading?
[ ] one of the books Bob gave her [‘context set’ response]
54
[ ] something else [‘not context set’ response]

‘Context set’ responses are compatible with d-linked interpretations, while ‘not context set’
responses are incompatible with such interpretations. I hypothesized that both overt and implicit
objects, if interpreted in a d-linked fashion as expected, would yield mostly ‘context set’ responses.
A control condition in which the overt object in the target item was explicitly not a member of the
context set, such as “Later, he saw her reading a magazine...” was also included. For the controls,
we expected a majority of ‘not context set’ responses, as the object is clearly not a member of the
salient context set.
Results from secondary task

Participants selected the context set response (e.g., “one of the books Bob gave her”) the
majority of times, at a rate of 95.6% and 94.2% for implicit objects with simple and complex
context sets respectively, and 98.2% and 98.6% for overt objects with simple and context sets.
Logistic regression analyses (glmer family binomial on R) revealed no significant differences
between these conditions (effect of object type: b = 0.150, SE = 0.096, z = 1.566; effect of set
complexity: b = -0.015, SE = 0.096, z = -0.160; interaction: b = 0.036, SE = 0.096, z = 0.377).
Upon encountering the control condition (e.g. “reading a magazine”), participants selected
the context set response only 40% of the time, which is significantly lower than in the condition
with implicit objects and complex context set (b = -3.736, SE = 0.287, z = -12.98*).
This secondary forced-choice task revealed that comprehenders did in fact associate the
implicit argument of the verb with the previously introduced context set, as intended. Although
implicit objects cannot be interpreted anaphorically in the strict sense (with a coreferential
55
antecedent), the proportion of context set interpretations for implicit arguments did not
significantly differ from the proportion of context set interpretations for overt indefinites. These
findings suggest that comprehenders are able to interpret implicit objects in a discourse-linked
fashion, i.e., in reference to a previously-mentioned, salient set. Crucially, this confirmation sets
an important foundation for interpreting the main results of Experiment 2.

Participants

58 native speakers of English participated in this experiment. 3 participants were excluded
from analysis due to poor performance on comprehension questions (below 80% accuracy).
Responses from the 55 remaining participants were analyzed.
2.4.1. Methods

Experiment 2 was programmed on Qualtrics and conducted online through Amazon
MTurk. As in Experiment 1, participants were asked to read and evaluate how acceptable each
item was in a 5-point scale from 1 – “Completely Unacceptable” to 5 – “Completely Acceptable”.

Materials and Design

In Experiment 2 I manipulated (i) context set complexity (simple or complex) and (ii)
object type (implicit or overt) to test whether representational complexity of the context set affects
implicit arguments’ acceptability as antecedents in discourse-linked constructions with
uninformative which one phrases.
56
Target items were virtually identical to those in Experiment 1, with a few exceptions. Since
this experiment investigates whether the accessibility of implicit objects as antecedents of
uninformative referring expressions can be improved, in Experiment 2 all target items end with
which one remnants (i.e., the informativity of the wh-phrase was not manipulated). In addition,
representational complexity of the context set was manipulated so that items had either simple sets
(e.g., “some books”) or complex ones (e.g., “some rare old books”). As Hofmeister (2011)
indicates that adjectives that frequently co-occur with nouns are more likely to boost effects
(effects disappear when adjectives are ‘incompatible’ with the noun), the adjectives for our target
items were selected from the most frequent co-occurrence results in the Corpus of Contemporary
American English (Davies 2008-). See example items in Table 2.7 (Appendix 3 contains a list with
all simple and complex sets used in target items).

Table 2.7 Exp. 2: Conditions and target items examples
Simple NP +
Implicit object
Bob gave Sarah several books to read.
Later, he saw her reading, but he doesn't know which one.
Complex NP +
Implicit object
Bob gave Sarah several rare old books to read.
Later, he saw her reading, but he doesn't know which one.
Simple NP +
Overt object
Bob gave Sarah several books to read.
Later, he saw her reading a book, but he doesn't know which one.
Complex NP +
Overt object
Bob gave Sarah several rare old books to read.
Later, he saw her reading a book, but he doesn't know which one.
Ungrammatical control
Bob gave Sarah several rare old books to read.
Later, he saw her sneezing, but he doesn't know which one.

2.4.2. Results

Experiment 2 analyses were conducted in a parallel way to Experiment 1. Object type and
Set type (and an interaction term) were included as fixed effects. Object type was coded -1
57
(implicit), +1(overt) and Set type was coded -1 (simple), +1 (complex), with random slopes for
subjects and items when supported by the data.
Set complexity did not affect the accessibility of implicit objects in discourse:
participants rated conditions with implicit objects as less acceptable than conditions with overt
objects regardless of set complexity (see Table 2.8, Figure 2.3).
Statistical analyses reveal a significant main effect of object type, but no main effect of set
complexity nor interaction (Table 2.9).
These results replicate findings from Experiment 1: when the referring expression is
uninformative (which one), people find implicit objects less acceptable antecedents than overt
objects. Moreover, results show that increased set complexity does not play a role in people’s
assessment of implicit objects’ accessibility to reduced referring expressions, even in the
discourse-linked dependencies investigated here.
Finally, as in Experiment 1, the ungrammatical control condition was considered
significantly less acceptable when compared to the implicit object with a complex set condition
(see Tables 2.10, 2.11).

Table 2.8 Experiment 2: Raw means and standard error of participants’ acceptability scores by object type
and context set type (5-point scale)

‘some books’ ‘some rare old books’
Acceptability Score SE Acceptability Score SE
Implicit object 3.82 0.072 3.80 0.074
Overt object 4.42 0.053 4.34 0.057

58
Table 2.9 Experiment 2 Results: Main effects and interactions (“*”: p < 0.05)
Object Type Set Complexity Object Type*Set Complexity
b
SE t b SE t b SE t
0.179 0.025 14.145* 0.016 0.014 -1.191 -0.008 0.014 -0.571

Figure 2.3 Experiment 2 raw means and +-1 SE by object type and wh-phrase type (5-point scale)

Table 2.10 Experiment 2: Raw means and standard error of participants’ acceptability scores in
ungrammatical control comparison (5-point scale)
Acceptability Score SE
Implicit object + complex set 3.80 0.074
Ungrammatical control 1.24 0.036

Table 2.11 Experiment 2 Results: Ungrammatical control comparison (“*”: p < 0.05)
Contrasts b SE t
Ungrammatical control
vs.
Implicit object +
Complex set
-0.792 0.045 -17.370*
59
2.4.3. Discussion

As in Experiment 1, people considered implicit objects to be less acceptable antecedents
of uninformative which one expressions relative to overt objects. This effect is replicated
regardless of context set complexity.
The representational complexity of the context set was manipulated to investigate whether
it could affect the acceptability of these discourse-linked dependencies. Surprisingly, the context
set manipulation yielded no effects. This may be due to two major differences between the current
study and prior work: firstly, Hofmeister (2011) and Frazier and Clifton (2011) focused on filler-
gap dependencies, while the current study addresses a different type of dependence, between a
sluice remnant and its antecedent clause. Secondly, because of the nature of implicit arguments, in
the current study we could not manipulate the complexity of the actual antecedent. Thus, we
investigated the role of complexity of the salient context set, which may not be as relevant.
2.5. Discourse Persistence

Having taken a closer look at implicit objects’ accessibility, this section aims to get a fuller
picture of implicit objects’ discourse status by addressing a ‘forward-looking’ question: how
likely are implicit and overt object referents to be mentioned again in the discourse (regardless of
the form/referential expression that this re-mention is done with)?
It has been observed in prior work that not all referents mentioned in discourse are equally
persistent (e.g., Givón 1983, Chiriacescu & von Heusinger 2010). For example, sentence subjects
– which are also often discourse topics – are overall more likely to be re-mentioned in subsequent
discourse than objects (e.g., Givón 1983).
60
In addition to grammatical position, a noun phrase’s form also affects the likelihood of its
re-mention. For example, research on English indefinite noun phrases suggests that the difference
between indefinites headed by ‘a/an’ (e.g., “Lisa was reading a book”) and those headed by ‘this’
(e.g., “Lisa was reading this book”) is that indefinite this signals the speakers’ intentions to further
elaborate on that referent (e.g., Prince 1981, Givón 1987, Gernsbacher and Shroyer 1989, Ionin
2006). This idea has been corroborated experimentally in a sentence continuation task (e.g.,
Chiriacescu 2014). A similar case occurs in Romanian, a language in which indefinite noun
phrases can appear with a specificity marker called pe-marking
15
: a sentence continuation task
revealed that indefinites with pe-marking yield higher rates of re-mention than those without it,
similarly to English indefinite this (e.g., Chiriacescu and von Heusinger 2010).
These patterns illustrate that some referents seem to exert a stronger influence than others
in shaping the direction discourse unfolds. Because of that, persistence has been proposed to be a
core component of referents’ discourse structuring potential (e.g., Chiriacescu and von Heusinger
2010): a measure of referents’ potential to affect and participate in subsequent discourse
16
.
At first glance, persistence and accessibility may seem to be tightly connected: topics
(typically highly accessible referents) tend to be re-mentioned most often (e.g., Givón 1983).
Moreover, prior work looking into the discourse status of source and goal referents in transfer of
possession events (e.g., “Lisa[source] gave the book to Maggie[goal]”) show that participants mention
the goal again more often than the source in a story continuation task (e.g., Arnold 2001).
Crucially, in the same task, participants also used pronouns most often to refer back to goal
referents than to source referents. In light of these findings, Arnold (2001) proposed that referent

15
Pe-marked indefinites differ from non-pe-marked indefinites in Romanian in that only the former rules out a non-
specific interpretation of the indefinite noun phrase (e.g., Chiriacescu and von Heusinger 2010).
16
A referent’s discourse structuring potential (e.g., Chiriacescu and von Heusinger 2010) involves both measures of
persistence and topic-shift potential; however, in this work, we will focus on the more general notion of persistence.
61
accessibility is guided by comprehenders’ expectations about referent persistence, i.e., the
likelihood of referent re-mention.
However, there is also a body of evidence counter this direct link between persistence and
accessibility. Looking at other English constructions, Fukumura & van Gompel (2009), Kaiser
(2010) and Brocher et al. (2016) show that it not always the case that referents that are more likely
to be re-mentioned are also more likely to be referred back to with a pronoun. Likewise,
Chiriacescu (2014) found that re-mentions of indefinite this in English and pe-marked indefinites
in Romanian display low rates of pronominalization. Thus, accessibility and persistence do not
seem to always go hand in hand.
In bringing together backwards-looking and forward-looking approaches, the current work
aims to discover more not only about existential implicit arguments and how they behave in
discourse, but also how these different facets of discourse-level phenomena interact.
2.5.1. Discourse Persistence and Implicit Arguments

Experiment 3 extends the research on discourse persistence to the domain of arguments
that were introduced implicitly into the discourse. Are they any less likely than overt objects to be
mentioned again? To the extent of my knowledge, no prior work has looked at the persistence of
implicit arguments.
If, as suggested by Koenig and Mauner (1999), implicit arguments simply do not introduce
a discourse referent into the discourse model, one may assume that implicit arguments do not have
any discourse structuring properties. If so, implicit objects should display very low rates of re-
mention when compared to overt indefinite objects.
62
Indirect evidence suggesting that implicit objects may indeed be less discourse persistent
comes from prior work on the relationship between discourse structure, clause type and transitivity.
Besserman and Kaiser (2017) conducted a sentence continuation task that shows that optionally
transitive verbs in main clauses are most often accompanied by overt objects, whereas those in
subordinate clauses appear most often with implicit objects. These patterns confirmed
experimentally prior theoretical claims based on the discourse properties associated with these
clause types: main clauses generally encode foregrounded information, i.e. information that is
crucial in the unfolding of the discourse, while subordinate clauses often encode backgrounded
information, i.e. information that is not central to the development of the discourse (e.g., Hopper
and Thompson 1980, Tomlin 1985). This more frequent association of implicit objects with clauses
that present backgrounded information in the discourse leads me to suggest that implicit arguments
could generally be associated with a reduced discourse structuring potential, i.e., a lower discourse
persistence.
In sum, if comprehenders see the implicitness of existential implicit objects as a signal
indicating low persistence/low discourse structuring potential, we may observe low rates of re-
mention for implicit objects relative to overt ones.
However, implicitness may not be the sole factor determining existential referents’
persistence. Other properties may play a role in modulating how persistent a referent is: for
example, Brocher et al. (2016) found that for indefinite noun phrases that have been inferred from
prior discourse context
17
, uniqueness affected referents’ persistence. Their work reveals that non-
unique referents were re-mentioned more often than unique ones. In addition to uniqueness, and
especially relevant to the current study, specificity may also be connected to referents’ persistence.

17
“Inferrables”, in the sense of Prince (1981), which were contrasted with brand-new indefinites.
63
Both English indefinite this and Romanian pe-marking are associated to specificity, i.e. the idea
that the speaker has a specific (but not familiar nor uniquely identifiable, which would be
compatible with the use of a definite noun phrase) referent in mind. In these cases, this additional
specificity marker increased referents’ persistence relative to an unmarked indefinite. In the case
of implicit objects, specificity can vary: without any context, implicit objects may be interpreted
as completely underspecified. However, in the discourse-linked constructions used in this study,
and as confirmed by the secondary task in Experiment 2, implicit objects were interpreted in the
same fashion as overt ones: as a member of the context set. If specificity plays a crucial role in
determining referents’ discourse persistence, we may see small or no differences in the re-mention
patterns of implicit and overt objects in these d-linked constructions.
2.5.2. Discourse Persistence and the Question under Discussion

Experiment 3 also explores a phenomenon closely related to how discourse is structured:
the Question under Discussion (QUD; e.g., Carlson 1983, Ginzberg 1996, Roberts 1998). The
Question under Discussion is the often implicit (but sometimes explicit) question that is motivated
by speakers’ conversational goals, which cooperative comprehenders are attuned to and possibly
attempt to satisfy. That is, in a conversation, participants are assumed to be cooperatively trying
to answer questions that arise from their shared, intended conversational goals. As a consequence,
comprehension involves not only understanding sentences but also how they contribute to the
current conversational goals (e.g., are they introducing a question? Elaborating on a question?
Answering an already introduced question?).
64
There is an intuitive relationship between the QUD and the notion of discourse persistence:
the QUD is considered a driving force determining where the discourse is going; discourse
persistence is a measure (in proportion of re-mentions) of where the discourse is going.
Prior work looking into how QUDs affect comprehenders’ expectations about subsequent
discourse looked into how overt questions influence comprehenders’ re-mention biases in
production (e.g., Rohde et al. 2007). In a story continuation paradigm, Rohde et al. (2007) found
that following transfer of possession events (e.g., “Lisa[source] gave the book to Maggie[goal]”), an
overt question “Why?” led to more continuations about the source, whereas a question “What
happened next?” led to more continuations about the goal. These patterns indicate that
comprehenders are attuned to the operative QUD and speakers’ conversational goals when
anticipating what subsequent discourse will be about.
In light of these findings, I hypothesize that an overt question introducing a QUD that
explicitly refers to the object referent may be able to increase its relevance in the discourse model
and affect people’s expectations about the likelihood that said object referent will be mentioned
again (i.e., its discourse persistence).
2.5.3. Investigating the discourse persistence of implicit objects using sluicing

Experiments 1 and 2 showed that for uninformative referring expressions, implicit objects
are not as acceptable antecedents as overt objects. These findings reveal that implicit objects have
an overall lower degree of accessibility. However, Experiment 1 also revealed that implicit objects
are as acceptable as overt objects as antecedents of informative expressions. Experiment 3 further
explores implicit objects’ discourse status by switching the focus to a forward-looking question:
do implicit and overt referents differ in how likely they are to be mentioned again?
65
2.6. Experiment 3: Are implicit objects less discourse persistent than overt objects?

To investigate this question, a forced-choice sentence continuation task was designed. In
order to keep the experiments in this chapter maximally parallel to one another so that accessibility
and persistence patterns can be reasonably compared, items from Experiments 1 and 2 were
minimally modified. In those experiments, a single referent (i.e., the overt or implicit object, e.g.
“Ø/a book”) was available to be the antecedent of the sluice remnant (e.g., “which one/which
book”). In Experiment 3, the subject of the critical verb (e.g., “read”) was exchanged for an
indefinite, available antecedent (e.g., “a student”) for the sluice remnant (ex.2.18a-b)
18
:

(2.18) a. I gave some students several books to read. Later, I saw a student reading, but
I’m not sure _____

b. I gave some students several books to read. Later, I saw a student reading a book,
but I’m not sure _____

[ ] which student.
[ ] which book.

In addition, the actual sluice was manipulated to be incomplete – i.e., missing its remnant
– as in (ex.2.18a-b). Participants were asked to choose a continuation for the sluice in a forced-

18
In the previous experiments the subject was a pronoun with a specific referent, and as such it was not an available
antecedent to the sluice remnant (i.e., the remnant could not possibly refer back to the subject pronoun).
66
choice task, by picking either a subject continuation (e.g., “which student”) or an object
continuation (e.g., “which book”).
As prior work has found a locality effect in sluices with overt antecedents (e.g., Frazier and
Clifton 1998, Carlson et al. 2009, Harris et al. 2015), overt objects are expected to yield more
object continuations than subject continuations. If implicit objects are less persistent than overt
ones, i.e., if implicitness is a driving force determining referents’ persistence, I expect that implicit
objects will yield less object continuations than overt ones. On the other hand, if specificity is a
significant property influencing referents’ persistence in the discourse, I expect that these implicit
objects will be re-mentioned as much as the overt ones – because both are interpreted in relation
to the context set.
A similar task involving competition of subject and object referents as sluice antecedents
has been used in Harris (2019), which presented sentences like (ex.2.19a-b), in which
uninformative wh-remnants which one were ambiguous between referring back to a subject or an
object antecedent:

(2.19) a. A guest talked to Bill or Sue, but I don’t remember which one.

b. Bill or Sue talked to a guest, but I don’t remember which one.

Harris (2019) investigated how disjuncts (e.g., “Bill or Sue”) and indefinites (e.g. “a
guest”) compete in the sluice disambiguation. Participants read sentences like (ex.2.19a-b) and
then answered questions such as “What can’t I remember?”. I take advantage of this clever design
to investigate the discourse persistence of implicit arguments by looking at how an overt, indefinite
67
subject competes with (i) an overt indefinite object (ex.2.18b) and with (ii) an implicit object
(ex.2.18a). In order to avoid using uninformative which one phrases (as Experiments 1 and 2
indicate implicit and overt antecedents differ in acceptability for these expressions), a forced-
choice sentence continuation task was designed so that participants were asked to pick between
two informative wh-expressions to complete the sluice (e.g., either “which subject” or “which
object”).

Participants

42 native speakers of English participated in this experiment. 1 participant was excluded
due to not being a native English speaker, 4 participants were excluded due to poor performance
on comprehension questions (below 80% accuracy) and 5 participants were excluded due to having
already participated in related experiments (Experiments 1 or 2). Responses from the remaining
32 participants were analyzed.
2.6.1. Methods

Experiment 3 was programmed on Qualtrics and conducted online through Amazon
MTurk. Participants were asked to read sentence fragments and choose a completion answer.

Materials and Design

In parallel with Experiments 1 and 2, 30 target items and 20 fillers were designed. In
addition to the object manipulation (overt or implicit), a QUD manipulation was included by
68
having a No-QUD condition with only one speaker and no overt question, and another QUD
condition with two speakers (‘me’ and ‘a friend’) plus an overt question “Did they read them?”
(see Table 2.12 for example items). Fillers were manipulated so that half had one-person dialogues
with no question, while the other half mirrored the two-person dialogue with a question.
The overt question “Did they read them?” in the QUD condition was intended to
emphasize (by overtly expressing) the speakers’ discourse goals: the speaker wants to know if the
students who received books did in fact read the books that were given to them. This particular
phrasing of this question obviates the object (e.g., the books), especially in comparison to other
ways to phrase the same inquiry that mention the subject but do not overtly reference the object
(e.g., “Did they?”, “Did they do it?”). I predict that, if comprehenders are attuned to the QUDs put
forth by interlocutors’ discourse goals, conditions with overt questions emphasizing the role of the
object in the current discourse may increase the proportion to which the object is mentioned (in
both overt and implicit conditions).
19

Table 2.12 Experiment 3: Conditions and target item examples
Context
sentence:
Me: I gave some students several books to read.

No Question

With explicit QUD

Overt
object

Later, I saw a student reading a book,
but I’m not sure_____

Friend: Did they read them?

Me: Later, I saw a student reading a book,
but I’m not sure_____

19
The explicit question condition (QUD) is a dialogue between two parties. As this dialogue was constructed to differ
minimally from the no question condition, I did not include discourse markers (e.g., “well”, “so”) or other features
that could have contributed to the naturalness of the dialogue. Nonetheless, constructions were kept constant across
all conditions.

69

Implicit
object

Later, I saw a student reading, but I’m
not sure_____
Friend: Did they read them?

Me: Later, I saw a student reading a book,
but I’m not sure_____

2.6.2. Results

Participants’ responses were coded in relation to the proportion of object responses, i.e., 0
for subject and 1 for object responses. Logistic mixed effects analyses of responses were performed
with glmer (family binomial) using R. Object type and QUD type (and an interaction term) were
included as fixed effects. Object type was coded -1 (implicit), +1(overt) and QUD type was coded
-1 (no question), +1 (overt question), with random slopes for Subjects and Items when supported
by the data.
Surprisingly, there was no difference between the discourse persistence of implicit
and overt objects: both were chosen to be re-mentioned as the sluice remnant at approximately
the same rate (Table 2.13, Figure 2.4).
Objects were chosen as the sluice continuation more often than subjects across conditions
(approximately 66% of the time; an additional analysis confirms that the proportion of object
responses in each condition was significantly higher than chance), confirming the previously
established preference for local (object) correlates in sluicing (e.g., Frazier and Clifton 1998,
Carlson et al. 2009, Harris et al. 2015), even when the object is implicit. That is, implicitness did
not make the object referent less appealing: people chose to re-mention the implicit object as much
as the overt object. Crucially, there was no difference between implicit and overt objects,
suggesting that the discourse persistence of implicit objects is no different than overt objects, i.e.,
70
both are equally likely to be mentioned again in subsequent discourse – at least when the referring
expression used in re-mention is an informative one (e.g., “which book”).
Moreover, the Question Under Discussion did not seem to influence the discourse
persistence of implicit objects nor overt ones. The QUD manipulation yielded no effects: even
when an explicit question emphasizing the role of the object for the current interlocutor’s goals
was present, people were not more likely to re-mention the object.
Statistical analyses reveal no significant main effects nor interactions (Table 2.14).

Table 2.13 Experiment 3: Proportion of object responses and SE by object type and QUD

No question With overt QUD
% object SE % object SE
Implicit object 0.656 0.037 0.637 0.038
Overt object 0.643 0.037 0.700 0.036

Table 2.14 Experiment 3 Results: Main effects and interactions (“*”: p < 0.05)
Object Type QUD Object Type*QUD
b SE t b SE t b SE t
-0.075 0.096 -0.776 0.073 0.097 0.075 -0.100 0.097 -1.032

71
Figure 2.4 Experiment 3: Proportion of object responses +-1 SE by object type and QUD type

2.6.3. Discussion

Experiment 3 investigated the discourse persistence of implicit objects, i.e., how likely they
are to be mentioned again in the discourse. This forward-looking line of inquiry contributes to
further understanding how implicit arguments participate in the discourse by focusing on their
persistence.
Findings from this forced-choice sentence completion task revealed that implicit objects
were as likely as overt objects to be mentioned again. That is, people did not seem to take
implicitness to be a signal that a referent is less likely to participate in subsequent discourse. On
the contrary, implicit and overt objects exhibited similar persistence patterns, i.e., similar discourse
structuring potential.
72
Moreover, the Question under Discussion manipulation meant to obviate the interlocutor’s
conversational goals and highlight object relevance in the discourse did not yield significant
effects. I speculate a few reasons as to why: first, participants may have inferred a similar QUD in
the no QUD condition simply via the principle of relevance (in Gricean terms); that is, even
without the explicit question, comprehenders could have inferred that “did they read them?” (or
something similar) was most likely the operative QUD given the information provided. Second,
the actual explicit question in the QUD condition may not have done a good enough job at
emphasizing the relevance of the object versus the subject; since the question refers to both
subjects and objects, it could have had the undesired effect of emphasizing the relevance of both
the subject and the object in equal parts. Finally, it is possible the there are other, higher-ranking
heuristics influencing sluicing resolution that could have washed away any differences between
these conditions, such as a stronger structural bias that would dilute discourse considerations such
as the QUD.
A last note in regard to sluicing resolution is that Experiment 3 confirms a locality effect
even when the local antecedent is implicit (i.e., sprouting). That is particularly surprising
considering that implicit objects are thought to have no syntactic representation and also cannot
bear focus, which are the two main candidate explanations for the locality effect in sluicing
antecedents (e.g., Frazier and Clifton 1998, Carlson et al. 2009, Harris et al. 2015).
2.7. General Discussion

Prior work on implicit agents (e.g., Koenig and Mauner 1999) suggests that implicit
arguments may lack the ability to introduce a discourse referent and are thus infelicitous as
pronoun antecedents. This may be related to the fact that pronouns are reduced, less informative
73
expressions that typically require highly accessible antecedents. In this work, we extend the
investigation on the discourse status of implicit arguments to the realm of implicit objects, which
differ from implicit agents in a number of ways including their syntactic behavior and descriptive
content. In order to get a fuller picture of implicit objects’ discourse status, the current work
adopted not only a backwards-looking line of inquiry examining how varying referring expressions
are better able to refer back to implicit objects (i.e., their accessibility; Experiments 1 and 2) but
also a forward-looking approach investigating how implicit objects contribute to discourse
structuring by assessing likelihood of re-mention (Experiment 3). These experiments also utilized
a novel ‘tool’ to tap into referents’ accessibility and persistence: discourse-linked sluicing
constructions.
Experiments 1 and 2 revealed that implicit objects are not as acceptable as overt objects as
antecedents of uninformative which phrases in sluicing. These findings go hand in hand with prior
observations that implicit arguments are typically not felicitous antecedents for pronouns and may
lack the ability to introduce a discourse referent. Crucially, Experiment 1 also shows that implicit
objects seem just as accessible as overt ones to informative which phrases, i.e., there was no
accommodation cost. This finding, coupled with the fact that the overall raw mean rating of
implicit objects with uninformative expressions was not drastically lower (and not nearly as low
as the ungrammatical controls), suggests that despite their resistance to uninformative referring
expressions, implicit objects are not completely discourse inert.
In fact, results from Experiment 3 show that implicit objects can be as discourse persistent
as overt ones. These patterns reveal that the discourse structuring potential of implicit objects can
be similar to that of overt objects. Despite their (slightly) lower accessibility to pronouns, implicit
objects were re-mentioned just as often as overt objects, indicating that their accessibility may not
74
be a direct match to their persistence. However, the current study looked only at re-mention made
with informative phrases; further work is needed to assess implicit objects’ persistence in discourse
beyond sluices and informative wh-expressions.
Notably, the constructions used in these experiments were all discourse-linked
constructions, that is, they restricted the interpretation of object referents to a previously-
mentioned salient context set. This discourse-linking property has allowed for a more careful
control of how the implicit objects were interpreted, i.e., their specificity (which was confirmed
by the secondary task in Experiment 2). The interpretation of implicit objects can vary drastically,
from almost completely underspecified when no context is provided, to somewhat specified when
there is supporting context for that inference – as is the case of these discourse-linked sluices. It
may be precisely because of their context-set interpretation (which matched those of overt objects)
that the discourse status of implicit objects was close or parallel to that of overt objects in various
respects. Further studies are required to untangle the contributions of specificity from that of
implicitness to these referents’ accessibility and persistence in the discourse.

75
Chapter 3 : Existential implicit objects’ accessibility and persistence in
discourse: a story continuation task investigation

3.1. Introduction

Chapter 2 investigated the accessibility and persistence of existential implicit objects in discourse
using sluices. Findings from the set of studies presented in Chapter 2 reveal that existential IOs
are less accessible than overt objects to a reduced anaphoric expression such as “which one”, but
as accessible as overt objects to a referring expression such as “which book”. Moreover, the third
experiment indicated that existential IOs can be as persistent in discourse as overt objects when
the referring expression is informative, as “which book”.
In this chapter, we expand Chapter 1’s inquiry on existential IO’s discourse behavior
beyond the domain of sluices and which-phrases by using a story continuation task. Participants
are asked to read story prompts and write continuations. Crucially, participants may freely (i)
choose to refer to the implicit/overt object in the prompt or not, and (ii) choose any referring
expression when referring to the prompt object. Analyzing the story continuations will allow for a
measure of Existential IO’s accessibility and persistence in discourse. This approach differs from
the sluicing investigation in Chapter 1 by promoting more naturalistic language production that
provides rich empirical data for analysis.
Moreover, in this study we further examine whether Existential IOs’ discourse accessibility
and persistence patterns are due solely to the way they are introduced in discourse (i.e. their
implicitness), or whether more general discourse factors such as a prior context related or
unrelated to the object may affect IO’s behavior in discourse.
76
Lastly, an additional exploratory analysis of how discourse coherence relates to IO’s
accessibility and persistence was performed to further our understanding about how the coherence
relation expressed between sentences – e.g., whether a sentence explains an event from the prior
sentence, whether it talks about an event that succeeded an event from the prior sentence, etc. –
influences implicit referents’ behavior in discourse.
This chapter is structured as follows: Section 3.2 goes over some relevant notions on
accessibility, persistence and the association between the two (this section partly recapitulates the
content introduced in Chapters 1 and 2). Section 3.3 introduces other factors that may influence
implicit objects’ discourse behavior, such as prior context and discourse coherence. Section 3.4
describes a story continuation task designed to investigate Existential implicit objects’ discourse
accessibility and persistence, as well as how prior context may affect these measures. Section 3.5
concludes with a general discussion.
3.2. Discourse Behavior: Accessibility and Persistence

As in Chapter 1, in which I investigated the discourse status of implicit objects using
sluicing as a tool, the current chapter focuses on two main measures of discourse behavior:
referents’ accessibility and persistence in discourse. Accessibility represents a ‘backwards-
looking’ approach: how and when are various referring expressions, such as pronouns and definite
noun phrases, able to refer back to different kinds of discourse referents? Prior claims on existential
implicit arguments’ discourse status and their apparent ‘inertness’ are based on evidence related
to their degree of accessibility (e.g., Koenig and Mauner 1999, Williams 2015).
In order to access a fuller picture of existential implicit objects’ discourse behavior, the
current work also investigates their discourse persistence – a ‘forward-looking’ approach to
77
referent’s life in discourse. In this line of inquiry, one can ask: how likely are people to mention
implicit objects again? This investigation provides a better understanding of how existential
implicit objects contribute to the structuring of discourse.
The subsections below further discuss the notions of accessibility, persistence and how
they relate to one another.
3.2.1. Accessibility

Discourse accessibility is a notion used for a referent’s degree of activation in the language
user’s mental model of the discourse. Referents that are activated and easy to retrieve, such as
recently-mentioned referents, are considered highly accessible. In contrast, referents that were
mentioned much earlier on may be harder to retrieve from memory and are thus considered low in
accessibility (e.g., Prince 1981, Givón 1983, Ariel 1990).
Notably, there is a general pattern between referents’ degree of activation and the
expressions used to refer back to them anaphorically: more informative expressions such as noun
phrases (e.g., “the book”, “the girl”) are often used for referents low in accessibility. In contrast,
less informative, ‘reduced’ expressions such as pronouns (e.g., “it”, “she”) are generally used to
refer back to referents high in accessibility (e.g., Ariel 1990, Gundel et al. 1993).
Existential implicit arguments are generally viewed as ‘discourse inert’ (e.g., Williams
2015), a view supported by research on implicit agents of short passives (e.g., in (ex.3.1a): there
is an implicit agent for the event of sinking the ship, i.e., it means “the ship was sunk by someone”).
Koenig & Mauner (1999) show that although implicit agents are able to control the PRO subject
of a non-finite clause (e.g., (ex.3.1b): the someone responsible for sinking the ship is the one who
did it to collect the insurance), they are generally not felicitous antecedents for reduced referring
78
expressions such as the pronoun “he” (compare (ex.3.1c-b); examples from Koenig & Mauner
1999):

(3.1) a. The ship was sunk.

b. The ship was sunk to collect the insurance.

c. The ship was sunk. #He wanted to collect the insurance.

d. The ship was sunk by someonei. Hei wanted to collect the insurance.

Whereas the pronoun “he” can easily refer back to the overt agent “someone” in (ex.3.1d),
this is not the case when the agent is introduced implicitly (ex.3.1c), suggesting implicit arguments
are simply not as accessible in the discourse model. On the other hand, more informative referring
expressions such as definite noun phrases can more easily refer back to these implicit arguments
(ex.3.1e):

(3.1) e. The ship was sunk. The perpetrator/The criminal/The man wanted to collect the
insurance.

As seen in Chapter 2, a similar pattern seems to arise with Existential implicit objects: they
are less felicitous than overt objects as the antecedents of “which one” phrases in sluices. The
79
current chapter takes a closer look at Existential IOs’ discourse behavior to better understand their
accessibility in discourse beyond sluices and “which” phrases.
3.2.2. Persistence

In this chapter I take a closer look at another measure of discourse behavior: Existential
IOs’ discourse persistence, i.e., how likely it is that a referent (in this case, implicit objects) will
be mentioned again in subsequent discourse – regardless of the form/referential expression used.
Prior work has observed that not all referents are equally likely to be mentioned again in
discourse, i.e., they are not all equally persistent (e.g. Givón 1983, Chiriacescu & von Heusinger
2010). Subjects, for example, are generally more likely to be referred back to again in subsequent
discourse relative to objects (e.g., Givón 1983). Sometimes a noun phrase’s form can affect its
persistence: in English, indefinite noun phrases headed by ‘this’ (e.g., “So, Ana was reading this
book yesterday…”) are more likely to be referred back to than indefinites headed by ‘a/an’ (e.g.,
“So, Ana was reading a book yesterday…”; e.g., Chiriacescu 2014).
An underexplored domain concerns the notion of implicit arguments’ persistence in
discourse: how likely are they to be mentioned again in discourse?
Work on patients of constructions with causative verbs (e.g., “chop”, “kill”, etc.) such as
in ((ex.3.2a-b), from Goldberg 2001) suggests that patient arguments are omitted when they are
indefinite, non-specific and generally low in discourse prominence.

(3.2) a. The chef-in-training chopped Ø and diced Ø all afternoon.

b. Tigers only kill Ø at night.
80

For example, if the action expressed by the verb is being emphasized (as in (ex.3.2a)), or
if the timing of the action is emphasized (as in (ex.3.2b)), discourse prominence shifts away from
the patient; thus the patient is more likely to be introduced implicitly.
In such cases, we may expect these low-prominence patients to be less likely to be
mentioned again in subsequent discourse, because other aspects of the event are more relevant to
the discourse (e.g., the action, the timing). The same idea can be applied to implicit objects: they
are typically interpreted as indefinite and non-specific, and if they are omitted due to low discourse
prominence, they may be inferred to be less likely to be referred back to in subsequent discourse.
Although Experiment 3 in Chapter 2 indicated that IOs are as persistent as overt objects, there are
three factors that may have helped boost IO’s persistence in those sentences: first, all sentences
contained a previous context with a salient set that helped narrow down the identity of the object.
Are unspecified IOs as persistent? Secondly, the only referring expressions available in that
experiment were informative which-phrases, which are compatible with IO’s accessibility degree.
However, if people are free to choose any expression, will they re-mention the IO as often?
Thirdly, in Chapter 2, all sentences were sluices, which display a locality effect in relation to the
wh-remnant interpretation – could the locality effect cold have boosted IO’s prominence in
discourse? To address these questions, the current chapter implements a task that promotes more
naturalistic language production.
3.2.3. The relationship between Accessibility and Persistence

One could expect that, due to their relative lower accessibility (lower activation in the
language user’s mental model of the discourse; as seen in Chapter 2), implicit objects would also
81
display lower persistence. Prior work has indeed suggested that accessibility and persistence are
closely linked: in a sentence continuation task, Arnold 2001 found that after reading transfer of
possession events (e.g., “Lisa[source] gave the book to Maggie[goal]”) participants were more likely
to write a continuation mentioning the goal (“Maggie”) than the source (“Lisa”). Moreover,
participants also used pronouns more often when referring back to the goal relative to the source.
Thus, Arnold 2001 proposed that the accessibility of a referent in discourse is guided by our
expectations that the referent will be mentioned again. However, there is also a body of evidence
counter this direct link between persistence and accessibility: Looking at other English
constructions, Fukumura & van Gompel (2009), Kaiser (2010), Brocher et al. (2018) and
Chiriacescu (2014) show that it not always the case that referents that are more likely to be re-
mentioned are also most likely to be referred back to with a pronoun (see also Chapter 1 for a more
in-depth discussion).
This asymmetry between referents’ accessibility and persistence in discourse highlights the
need for further research on different kinds of referents’ discourse behavior and how other
discourse factors may modulate that behavior. Thus, in the current Chapter I explore how prior
context and discourse coherence may interact with implicit objects’ accessibility and persistence
in discourse.
3.3. Broader discourse factors: prior context, discourse coherence

This work investigates how the implicitness of direct objects affects their accessibility and
persistence in discourse. However, whether a referent was introduced overtly or implicitly may
not be the single determining factor in its discourse behavior: other discourse properties may play
a role in whether a referent may be mentioned again and which rereferring expression may be used.
82
3.3.1. Prior context

Amongst various properties, specificity seems to be connected to referents’ persistence.
Both English indefinite this and Romanian pe-marking on indefinites are associated with
specificity, i.e. the idea that the speaker has a specific referent in mind (but not a familiar or a
uniquely identifiable referent, which would be compatible with the use of a definite noun phrase).
In these cases, indefinites that indicate specificity displayed higher discourse persistence relative
to unmarked ones (Chiriacescu and von Heusinger 2010, Chiriacescu 2014).
Existential implicit objects are generally non-specific: although their interpretation is
constrained by verb semantics (e.g., the existential IO of “read” is something that one typically
reads, such as books/magazines/letters), the identity of the object is unspecified.
However, the interpretation of existential IOs is possibly modulated by inferences when
supported by context. For example, in “Ana read.”, one can only assume Ana read something
typically read; but in “Ana went to the library and read.”, language users may additionally infer
that she read something that she got at the library. This extra piece of information about what Ana
read may help increase this implicit objects’ prominence in discourse, and consequently affect
their discourse persistence, similarly to how English and Romanians indefinites that were marked
for specificity displayed higher persistence.
In the current work, the role of context on implicit objects’ discourse behavior is thus tested
by manipulating whether or not prior context would provide information that could potentially
help people generate inferences about the implicit object (e.g., “Ana received several books. Later,
she was reading…”) or not provide any information related to the identity of the implicit object at
all (e.g., “Ana went to the park. Later, she was reading…). As the related context may promote the
83
inference that Ana was reading one of the books she received, implicit objects may display
specific-like behavior such as increased prominence and persistence in discourse.
3.3.2. Discourse Coherence

Discourse coherence pertains to how connected sentences in discourse relate to one
another. For instance, in (ex.3.3a) the second sentence expands on the event expressed by the first
sentence by identifying what Moira was baking. In (ex.3.3b), the second sentence offers an
explanation as to why the event described in the first sentence occurred: Moira baked because she
wanted surprise Helen. In (ex.3.3c), the second sentence describes an event that succeeded the
event from the first sentence. In each of these cases, the sentences are connected by different kinds
of coherence relations (e.g., “Elaboration”, “Explanation” and “Occasion” coherence relations,
respectively; see Kehler 2002, Asher & Lascarides 2003).

(3.3) a. Moira baked last night. She was making muffins.

b. Moira baked last night. She wanted to surprise Helen.

c. Moira baked last night. Afterwards, she went to a movie.

Prior work has shown that the coherence relation between sentences can influence
referents’ behavior in discourse. For example, Arnold 2001 showed that readers tend to refer back
to the goal referent more often than the source referent following transfer of possession events,
and that this effect is modulated by whether their continuations related more to the events’ cause
84
or to a subsequent event. Rohde 2006 further confirms that choice of who to mention next after
transfer-of-possession events is be modulated by the coherence relation between sentences:
Participants wrote continuations after reading prompts such as “John[source] handed a book to
Bob[goal]. He…”. Their continuations were annotated for the coherence relation expressed between
the prompt and the continuation. The results revealed that when continuations were about an event
that followed or resulted from the prompt event (Occasion and Result coherence relations),
continuations were more likely to refer back to the goal referent (“Bob”), as found in Arnold 2001.
However, when continuations expressed other types of coherence relations such as further
expanding on aspects of the prompt event or explaining the prompt event (Elaboration and
Explanation coherence relations), continuations were actually more likely to refer to the source
referent (“John”). Rohde 2006 shows that likelihood of being referred to in subsequent discourse
(persistent) is modulated by the coherence relation expressed between sentences.
It is not entirely clear how discourse coherence may affect implicit objects’ accessibility
and persistence in discourse, therefore coherence was not directly manipulated or tested in the
current work. Nonetheless, as coherence relations have been shown to play a significant role in
other types of referents’ discourse behavior, an additional exploratory analysis (reported in Section
3.4.2) on how coherence modulates IOs’ accessibility and persistence was performed to further
shed light on how these factors interact in discourse.
3.4. Investigating the discourse behavior of implicit objects in a story continuation task

A story continuation task was designed to investigate how implicit objects differ from overt
ones in their discourse behavior, i.e. their persistence and accessibility. The current study also
examines whether a prior context that helps narrow down the identity of the object via inference
85
(e.g., someone receives books and then reads: they likely read one of the books received) may
increase IO’s prominence in discourse, consequently affecting its persistence and accessibility.
Lastly, this work includes an exploratory analysis on how coherence relations modulates IO’s
discourse behavior.

Participants

60 native speakers of English participated online through Amazon Mechanical Turk. 5
participants were excluded due to writing meaningless, ungrammatical story continuations, and 7
participants were excluded due to having already participated in related experiments (from Chapter
2). Responses from the remaining 48 participants were analyzed.
3.4.1 Methods

Sentence or story continuations tasks (e.g., Gernsbacher & Shroyer 1989, Stevenson et al.
1994, Arnold 2001, Rohde 2006) combine both language comprehension (because participants
need to read the story prompts) and production (because they need to write continuations based on
the prompt they have read). While comprehending the prompts, participants are presumably
generating a mental model of the discourse. Crucially, participants write continuations based on
how they comprehended the prompt, i.e., based on their mental discourse model of the prompt.
Consequently, assessing participants’ continuations and whether/how they choose to refer back to
the prompt object provides a window into participants’ mental representation of the prompt object
referent, its prominence and accessibility in the mental discourse model. More specifically, when
participants choose to refer back to the prompt object, it reveals that participants deemed the object
86
prominent in discourse and likely to occur again. When participants choose to refer back to the
prompt object with a reduced expression such as a pronoun, it reveals how activated/accessible the
referent was in their mental model of the discourse.
In sum, participants were asked to read story prompts (see examples in Table 3.1) and
produce continuations at least three sentences long. The three sentence requirement was stipulated
to provide more opportunity for references to the prompt object (e.g., in case reference did not
occur in the first sentence) and specially so that a wider context in participants’ continuations
would facilitate data annotation (e.g., see for instance (ex.3.11) in Table 3.2: the pronoun ‘it’ in
the first sentence seems to be used to refer back to the implicit object of ‘writing’ in the prompt,
and subsequent instances of the pronoun ‘it’ in the second and third sentences further confirm this
intuition).
Continuations were annotated for (i) mention of the prompt object, (ii) form of prompt
object mention, (iii) grammatical position and (iv) coherence relation.

Design and Materials

The experiment contained 20 target and 30 filler items. On target trials, we manipulated
the story prompt’s (i) object type (either overt or implicit) and (ii) context sentence (either a
‘strong’ context containing a set related to the object or an ‘irrelevant’ context unrelated to the
object). The targets used 20 different OPT verbs (e.g. read, bake, study). As exemplified in Table
3.1 (ex.3.4a-d), targets either introduced a context set or not (‘strong’ context vs. ‘irrelevant’
context) and included an OPT verb with either an overt or an implicit object (e.g., “reading a book”
or “reading”). The current targets were minimally modified from the experiments in Chapter 2,
87
and as such utilized the same simple context sets (e.g., “several books”), as well as the same set of
OPT verbs and object nouns.

Table 3.1 Examples of target items and conditions
Condition Story Prompt
‘Strong’ context +
implicit object
3.4a. I gave Sarah several books to read.
Later, I saw her reading.
‘Strong’ context +
overt object
3.4b. I gave Sarah several books to read.
Later, I saw her reading a book.
‘Irrelevant’ context +
implicit object
3.4c. I went to the gym with Sarah.
Later, I saw her reading.
‘Irrelevant’ context +
overt object
3.4d. I went to the gym with Sarah.
Later, I saw her reading a book.

Data Annotation

Participants’ continuations were annotated for (i) whether the object referent is mentioned,
(ii) if so, what form/expression is used to mention it, (iii) in what grammatical position the mention
appeared, and (iv) what coherence relation is expressed between the continuation sentence and the
prompt sentence.
For (i) mention of the object referent, each continuation was categorized as falling into
one of these groups:
- there was a clear mention of the prompt object (yes)
- there was no mention of the prompt object (no)
- there was an expression that could be referring to the prompt object, but it wasn’t clear whether
or not it was about the prompt object (unclear) or
- the continuation was incomprehensible or ungrammatical (other).
88
Only the first possible mention of the prompt object (regardless of whether it occurred in
sentence 1, 2 or 3 of the continuation) was considered for analysis, because any subsequent
reference could be referring back to this first mention (i.e., it would not be possible to determine
whether this subsequent reference was referring back to the prompt object or not).
Decisions on whether an expression was clearly mentioning the prompt object were made
on a conservative basis: if there seemed to be any degree of ambiguity regarding the antecedent of
a referring expression (e.g., if the expression could possibly also be referring to the event or
something else in the context sentence), the expression was considered ‘unclear’ regarding whether
or not it mentioned the prompt object (see (ex.3.6) on Table 3.2).
For (ii) the form/expression used in mentions of the object referent, continuations that
were annotated as mentioning the prompt object were also annotated for the form of the expression
used to mention the prompt object, i.e., whether it was an indefinite noun phrase, a definite noun
phrase, a pronoun, a wh-phrase, a bare nominal, a possessive, etc.
For (iii) the grammatical position the mention appeared in, continuations that were
annotated as mentioning the prompt object were also annotated for the grammatical position said
mention appeared in, i.e., whether they were a subject, a direct object, an indirect object, etc.
Finally, for (iv) the coherence relation expressed between the continuation sentence
and the prompt sentence, continuations were annotated for which coherence relation connected
the continuation sentence and the prompt sentence
20
. Any pair of sentences in discourse conveys
a type of coherence relation between them; for instance, the sentence following the prompt may

20
For the coherence relation analysis, only continuations that mentioned the object in the first sentence of the
continuation were considered. That is because only the first sentence has a direct coherence relation to the prompt
sentence, which were controlled for the experiment. Sentences 2 and 3 of continuations would have to have been
annotated for their coherence relation to their preceding sentence (1 and 2 respectively), which varied greatly across
continuations.
89
be about an event that occurred subsequently to the event described in the prompt (Occasion), it
may be a sentence explaining why the event in the prompt sentence occurred (Explanation), it may
be a sentence expanding on some aspect of the prompt sentence (Elaboration), etc. Thus,
continuations were annotated for which coherence related could be inferred to hold between the
prompt and the first sentence the continuation, using the set of coherence relations from Kehler
(2002): Elaboration, Occasion, Explanation, Result, Violated Expectations or Parallel coherence
relations (see Rohde 2008 for a detailed description of these coherence relations and how they can
be identified).
Table 3.2 includes examples of continuations produced by participants and how they were
annotated for (i) Mention of the object referent, (ii) Form/expression used, (iii) Grammatical
position and (iv) Coherence relation.

Table 3.2 Examples of continuations produced by participants and how they were annotated (the first mention
of the prompt object appears underlined in bold).

Ex. n
o

Prompt and Continuation (S1, S2 and S3)

Annotation

(3.5)

I had dinner with Frank. That evening, I saw him sculpting.

S1: He had his headphones in so he didn't hear me approaching.
S2: I watched him work for a few minutes and then left him alone.
S3: He is a very talented artist.

(i) No
(ii) -
(iii) -
(iv) -

(3.6)

I sent William several melodies to play. That night, I saw him
playing a melody.

S1: He did a beautiful job.
S2: I told him that it sounded lovely.
S3: He thanked me and continued playing.

(i) Unclear
(could be
reference to
object or event)
(ii) -
(iii) -
(iv) -

90

(3.7)

I went to the park with Albert. That night, I saw him writing an
essay.

S1: It was the park experience and how much fun he had.
S2: He also made a drawing.
S3: He drew himself, the dog, and I.

(i) Other
(ungrammatical
S1)
(ii) -
(iii) -
(iv) -

(3.8)

I handed Victoria several books to read. Later, I saw her reading.

S1: She was intently reading a copy of The Cat Who Walked
Through Walls.
S2: She blushed when she noticed me.
S3: I could guess which parts she was reading.

(i) Yes
(ii) Indefinite NP
(iii) Direct obj.
(iv) Elaboration

(3.9)

I had drinks with Betty. The next day, I saw her knitting.

S1: I asked what she was knitting.
S2: She told me she was pregnant and it was for her baby.
S3: I was shocked she would have drinks with me while she was
pregnant.

(i) Yes
(ii) Wh-phrase
(iii) Embedded
direct obj.
(iv) Occasion

(3.10)

I had brunch with Caitlin. The next day, I saw her sketching.

S1: She had told me that she was taking up sketching again.
S2: I was intrigued and looked at the sketch.
S3: It’s a mystery why she ever stopped because she is extremely
artistic.

(i) Yes
(ii) Definite NP
(iii) Direct obj.
(iv) – (not S1)

(3.11)

I went to the park with Albert. That night, I saw him writing.

S1: I asked if it was a novel.
S2: He said it was a grocery list.
S3: I read it to be sure.

(i) Yes
(ii) Pronoun
(iii) Embedded
subject
(iv) Occasion

(3.12)

I went to the grocery store with Marge. That evening, I saw her
driving a car.

S1: It was silver and I hadn't seen it before.
S2: I wondered if it was really her.
S3: Then she waved at me.

(i) Yes
(ii) Pronoun
(iii) Subject
(iv) Elaboration
91
The annotation allowed to determine whether any given mention of the object in the
continuations constituted a case of referential persistence. An expression was only considered to
be referring back to the prompt object (i.e., an instance of discourse persistence) when it was
annotated for (i) clearly mentioning the object from the prompt and when (ii) the form of the
referring expression indicated givenness, i.e., that its antecedent (the prompt object) had already
been evoked in discourse. Hence, re-mentions of the prompt object made with pronouns and
definite noun phrases were considered instances of persistence (ex.3.10, ex.3.11). In contrast,
continuations that mentioned the prompt object using expressions that do not suggest givenness,
but rather suggest newness in the discourse (e.g., indefinite noun phrases, wh-phrases, bare
nominals) were considered as introducing (or re-introducing) a referent in the discourse, and
therefore were not considered instances of persistence (ex.3.8, ex.3.9).
To further illustrate this point, continuations such as (ex.3.13a-b) mention the object from
the prompt (the implicit object of ‘reading’), and do so with expressions that indicate givenness in
discourse (the definite noun phrase “the book”, the pronoun “it”). That is, these expressions are
referring back to the referent introduced by the implicit object of ‘reading’. In contrast,
continuations such as (ex.3.13c-d) also mention the object from the prompt but do so with
expressions that indicate newness in discourse (the indefinite noun phrase “a book about cats”, the
wh-phrase “what”). These expressions are not referring back to a previously-introduced referent,
and consequently they do not constitute evidence that the referent introduced by the implicit object
of ‘reading’ is ‘persisting’ in discourse.
It is interesting to note that the majority of cases in which an expression that indicates
newness was used to talk about the object prompt occurred in sentences that repeated the prompt
verb, as (ex.3.8, ex.3.9). Once the verb is repeated, it becomes impossible to determine whether
92
the referring expression in object position is referring back to the prompt object or has been
licensed by this new transitive occurrence of the verb. The fact that the continuations repeat the
verb to talk about the object referent, in addition to the fact that the referent is realized with an
expression that indicates newness, suggests that there was a need to re-introduce the referent in
discourse – presumably because in these instances the implicit object was not considered discourse
prominent enough to be referred back with an expression that indicates givenness.

(3.13) I gave Sarah several books to read. Later, I saw her reading.

a. The book was about cats.

b. She was so engaged, she couldn’t put it down.

c. She was reading a book about cats.

d. I asked her what she was reading.

The proportion of cases in which the continuation referred back to the prompt object with
an expression that indicates givenness (definite noun phrases, pronouns) constituted the primary
measure of persistence.
As the principal measure of accessibility, the proportion of references to the prompt object
that were made using expressions that are indicators of high accessibility, i.e. pronouns (ex.311,
93
3.12, 3.13b) was considered. That is, pronominalization rate was considered a proxy for indicating
a referent high in accessibility.
A secondary measure of accessibility was the proportion of references to the prompt object
that were made in subject position. Subjecthood is generally considered an indication of higher
accessibility (e.g., Chafe 1976, Givón 1983, Brennan et al. 1987). Moreover, work on language
production reveal that language users tend to place more accessible referents in the beginning of
sentences (which in English tends to be the subject; e.g., Bock and Warren 1985, Bock 1987).
Therefore, references to the object prompt that were realized in subject position were considered
a proxy for indicating a referent high in accessibility.
Crucially, only references made in subject position of a main clause were considered for
this analysis (see (ex.3.12)). That is because main clauses typically encode foregrounded discourse
information, i.e., “the material which supplies the main points of the discourse” (e.g., Hopper &
Thompson 1980). References made in subject position of an embedded clause (e.g., (ex.3.11))
were not considered an indicator of high accessibility because embedded clauses typically encode
information that is backgrounded in discourse, that is, information that is less crucial for the
structure and development of the discourse (e.g., Hopper & Thompson 1980, Tomlin 1985). Thus,
I assume that subjects of embedded/foregrounded clauses are relatively less accessible than those
of main/foregrounded clauses, and therefore should not be taken as straightforward evidence that
a referent is high in accessibility.
Finally, an additional exploratory analysis regarding how references to the prompt object
were distributed amongst different kinds of coherence relations is presented. This analysis is meant
to expand our understanding of which discourse factors contribute to implicit objects’ behavior in
discourse.
94
A second annotator worked on a subset of the data (33.3% of all continuations) to allow
for an assessment of annotator agreement. The two annotators displayed 97.64% agreement
(percent of instances in which the same annotation was used for each story continuation). This
annotator agreement assessment provides additional insurance that continuations were interpreted
and annotated appropriately in relation to the prompt object.
3.4.2. Results and Discussions

Results from the story continuation task are described below: (i) first, the findings
pertaining to existential implicit objects’ discourse persistence (likelihood of referring back to
prompt object) are presented, followed by a discussion of the findings. (ii) Secondly, findings
related to existential IO’s discourse accessibility (rate of pronominalization) are presented,
followed by a discussion. (iii) Thirdly, findings on another measure of accessibility – promotion
to subject of the main clause – are discussed alongside a discussion on the findings. (iv) Finally,
an exploratory analysis on the effects of coherence relations pertaining to IO’s persistence and
accessibility in discourse is discussed.

Results on Discourse Persistence: Likelihood of referring back to prompt object

Following the annotation criteria described above, continuations were coded for whether
they contained a clear reference to the object prompt or not (+1 and 0 respectively). Generalized
Binominal linear mixed effects analyses were performed with glmer using R (R Core Team, 2017).
Prompt object type (implicit or overt), Context type (strong or irrelevant) and an interaction term
95
were included as fixed effects. Maximal random effect structure were included when supported by
the data (see Barr 2013).
As previously described, only the first clear reference to the prompt object was considered
for analysis, regardless of whether it occurred in the first, second or third sentences of the
continuation (S1, S2 and S3). The great majority of references to the prompt object occurred in the
first sentence of continuation: for both implicit and overt objects in both ‘strong’ and ‘irrelevant’
context conditions, the likelihood of referencing to the prompt object drops drastically the further
away from the prompt, i.e., in sentences 2 and 3 (Table 3.3, Figure 3.1). For the statistical analysis,
all first clear references to the prompt object across the three sentences were aggregated (Table
3.4, Figure 3.2).

Table 3.3 Proportion and SE of references to prompt object by condition and sentences (S1/S2/S3)

Conditions:
S1 S2 S3
Prop. SE Prop. SE Prop. SE
‘Strong’ context +
implicit object
0.154 0.023 0.012 0.007 0.033 0.011
‘Strong’ context +
overt object
0.288 0.029 0.045 0.013 0.045 0.013
‘Irrelevant’ context +
implicit object
0.070 0.016 0.025 0.010 0.020 0.009
‘Irrelevant’ context +
overt object
0.262 0 .028 0.070 0.016 0.029 0.010

96
Figure 3.1 Proportion of references to prompt object +-1 SE by conditions and sentences (S1/S2/S3)

Findings reveal that overt objects are overall more persistent, i.e., they were more likely
to be referred back to in subsequent discourse relative to implicit objects: Conditions with
overt objects in the story prompt yielded a significantly higher proportion of continuations
containing reference to the prompt object relative to conditions with implicit objects in the prompt
(Tables 3.4, 3.5 and Figure 3.2).
In addition, a context related to the prompt object (‘strong’ context) yielded a higher
proportion of references to the prompt object relative to an unrelated context (‘irrelevant’
context) – but only when the prompt object was implicit (Tables 3.4, 3.5, 3.6 and Figure 3.2).
That is: the strong context, which mentions a set related to the prompt object (e.g., “several
books”), increases the persistence/likelihood of referring to the prompt object when it was implicit
(e.g., the implicit object of “reading”) but not when it was overt (e.g., “a book”).
97

Table 3.4 Proportion and SE of references to prompt object by condition
Proportion SE
‘Strong’ context + implicit object 0.200 0.025
‘Strong’ context + overt object
0.379 0.031
‘Irrelevant’ context + implicit object 0.116 0.020
‘Irrelevant’ context + overt object
0.362 0.031

Figure 3.2 Proportion of references to prompt object +-1 SE by context type and prompt object type

Table 3.5 Results (Persistence): Main effects and interactions (“*”: p < 0.05)
Main Effects Interaction
Object Type Context Type Object Type*Context Type
b SE z p b SE z p b SE z p
0.625 0.082 7.564 <0.001* 0.185 0.081 2.268 0.023* -0.149 0.081 -1.828 0.067

Table 3.6 Results (Persistence): Planned pair comparisons (“*”: p < 0.05)
Contrasts:
Context Type
b SE z p
Implicit objects 0.325 0.130 2.507 0.012*
Overt objects 0.036 0.100 0.363 0.716

98
Discussion (Discourse Persistence: Likelihood of referring back to prompt object)

Findings related to objects’ likelihood of referring back to the prompt object revealed that
implicit objects are less discourse persistent than overt objects. This is compatible with prior claims
suggesting that implicit arguments are generally less prominent in discourse (e.g., Koenig and
Mauner 1999, Goldberg 2001, Williams 2015).
Notably, context influenced the likelihood of referring back to implicit objects (but not
overt ones). This effect may have happened because a prior context related to the implicit object
increases its prominence in discourse, despite its implicitness. This prior context could have also
helped to ‘narrow down’ the identity/specificity of the implicit object (e.g., the implicit object may
be interpreted as a member of the previously mentioned set), making the implicit objects’
interpretation more up to par with the overt object.

Results on Discourse Persistence: Context effects on overall mentions of prompt object

The analysis described in the previous section considered the proportion of references to
the object prompt (made with referring expressions that indicate givenness, as described in section
3.3) relative to all continuations produced by participants. As seen above, a context related to the
prompt object increased the likelihood of referring back to the implicit object.
However, one may wonder whether the context affected implicit objects’ referential
persistence in discourse, or whether the context simply increased the overall amount of times
participants mention the object with any kind of referring expression, like indefinites –
consequently also increasing the proportion of references to the object with definite noun phrases
99
and pronouns. That is, context could have simply raised participants’ interest in further addressing
the prompt object in their story continuations, and not necessarily increased the prompt objects’
referential persistence in discourse.
As previously mentioned, referring expressions that signal newness in the discourse (such
as indefinites) do not indicate that people are referring back to the prompt object referent, that is,
they are not indicators of referential persistence. Rather, they suggest people are talking about the
prompt object by (re-)introducing it in the discourse (e.g., (ex.3.8-3.9) are instances of such
continuations).
In (ex.3.8), the continuation is clearly mentioning what I saw Victoria reading, i.e., the
implicit object of “reading” from the prompt: she was reading a copy of The Cat Who Walked
Through Walls. Not only is the referring expression here an indefinite, but the verb has been
repeated, further suggesting this is a mention of the prompt object that is re-introducing that
referent in discourse. This re-introduction presumably occurs because the prompt object referent
was not deemed accessible enough to be picked up with a definite noun phrase or pronoun, e.g.,
“The book/it was a copy of The Cat Who Walked Through Walls”.
Similarly, in (ex.3.9), the continuation is again mentioning the prompt object, i.e., the
question in “I asked what she was knitting.” is inquiring about the identity of what I saw Betty
knitting – the implicit object of knitting in the prompt. However, the wh-phrase “what” does not
signal givenness in the discourse and is not referring back to the previous implicit object, but rather
re-introducing it in discourse.
For clarification, I henceforth use the terms “talk about” and “mention” when continuations
include any expression (e.g., indefinite and definite noun phrases, wh- phrases, pronouns) that is
being used to identify the same referent as the one introduced by the object prompt. I use the term
100
“reference” to the prompt object exclusively when the expression used indicates givenness in the
discourse (e.g., definite noun phrases and pronouns). Consequently, the proportion of references
to the prompt object is a subset of all the mentions related to the prompt object.
The previous analysis of referential persistence looked only at references to the prompt
object. Below, an analysis was performed to assess whether the context manipulation had an effect
on the proportion of times continuations talked about the prompt object using any kind of referring
expression (see Table 3.7 for the distinction between referring vs. talking about/mentioning).

Table 3.7 'Persistence' vs. 'Talking about/Mentioning the object' proportions
Referring to the object (Persistence) Talking about/Mentioning the object

𝑟𝑒𝑓𝑒𝑛𝑐𝑒𝑠 𝑡𝑜 𝑡 ℎ𝑒 𝑝𝑟𝑜𝑚𝑝𝑡 𝑜𝑏𝑗𝑒𝑐𝑡 𝑚𝑎𝑑𝑒 𝑤𝑖𝑡 ℎ 𝑑𝑒𝑓𝑖𝑛𝑖𝑡𝑒𝑠 , 𝑝𝑟𝑜𝑛𝑜𝑢𝑛𝑠 𝑎𝑙𝑙 𝑐𝑜𝑛𝑡𝑖𝑛𝑢𝑎𝑡𝑖𝑜𝑛𝑠

𝑐𝑜𝑛𝑡𝑖𝑛𝑢𝑎𝑡𝑖𝑜𝑛𝑠 𝑡 ℎ𝑎𝑡 𝑡𝑎𝑙𝑘𝑒𝑑 𝑎𝑏𝑜𝑢𝑡 𝑜𝑏𝑗𝑒𝑐𝑡 𝑚𝑎𝑑𝑒 𝑤𝑖𝑡 ℎ 𝑎𝑛𝑦 𝑘𝑖𝑛𝑑 𝑜𝑓 𝑟𝑒𝑓𝑒𝑟𝑟𝑖𝑛𝑔 𝑒𝑥𝑝𝑟𝑒𝑠𝑠𝑖𝑜𝑛 𝑎𝑙𝑙 𝑐𝑜𝑛𝑡𝑖𝑛𝑢𝑎𝑡𝑖𝑜𝑛𝑠

The proportion of all continuations that talk about the prompt object reveals that overt
objects are mentioned more often than implicit ones (Tables 3.8, 3.9 and Figure 3.3). That means
that, in general, overt objects are more likely to be mentioned than implicit ones; and as seen
previously, overt objects are also more likely to be referred to with an expression that indicates
givenness in discourse.
Notably, context had no effect on the proportion of times the prompt object was mentioned
(Tables 3.8, 3.9 and Figure 3.3): prior context did not influence the rate at which implicit and overt
objects were talked about in participants’ continuations.
A direct comparison between references to the prompt object and all prompt object
mentions is illustrated in Figure 3.4.
101

Table 3.8 Proportion and SE of continuations that talked about prompt object by condition
Proportion SE
‘Strong’ context + implicit object 0.458 0.032
‘Strong’ context + overt object
0.550 0.032
‘Irrelevant’ context + implicit object 0.458 0.032
‘Irrelevant’ context + overt object
0.533 0.032

Table 3.9 Results (Overall mentions): Main effects and interactions (“*”: p < 0.05)
Main Effects Interaction
Object Type Context Type Object Type*Context Type
b SE z p b SE z p b SE z p
0.187 0.068 2.720 0.006* 0.019 0.068 0.281 0.778 0.018 0.068 0.270 0.787

Figure 3.3 Proportion of continuations that TALK ABOUT/MENTION the prompt object (with any referring
expression) +-1 SE by context type and prompt object type

102
Figure 3.4 Proportion of REFERENCES relative to ALL MENTIONS by context type and prompt object type
(Mention includes all instances of talking about the prompt object with any referring expression; Reference
includes only the instances of talking about the prompt object with given expressions)

Discussion (Discourse Persistence: Context effects on overall mentions of prompt object)

The analysis of the overall mentions of the prompt object shows that not only overt objects
are more likely to be referred back to with expressions that indicate givenness, they are also overall
more likely to be mentioned in subsequent discourse with any kind of referring expression.
Most of the instances in which the implicit object was talked about were realized with
referring expressions that indicate newness, like indefinite noun phrases and wh-phrases (Figure
3.4). These patterns suggest that the implicit object is most often not being referred back to (an
instance of persistence), but rather they are being re-introduced in discourse. Overt objects, on the
other hand, were most often referred back to with an expression that indicates givenness, like
definite noun phrases and pronouns. These findings again confirm that implicit objects are lower
103
in persistence relative to overt objects.
Nonetheless, these findings also indicate that implicit objects are not completely irrelevant
to subsequent discourse: implicit objects were talked about a considerable amount, almost to the
same extent as overt objects (e.g., approximately 45% and 54% of all continuations respectively).
That is, implicit objects’ contribution to the structuring of subsequent discourse is not entirely
void. Most of the time, however, this contribution seems to require that the IO be re-introduced to
the discourse model.
Crucially, prior context did not influence how often implicit or overt objects were talked
about in discourse. This finding assuages possible concerns that prior context could have simply
increased the overall likelihood of the implicit object being mentioned. Prior context did not
affect overall mentions; it only affected implicit objects’ likelihood of being referred back to
with an expression indicating givenness, i.e., their discourse persistence.

Results on Discourse Accessibility: Pronominalization rate

Continuations with a clear reference to the prompt object were also annotated and coded
for whether the reference was made with a pronoun or not (+1 and 0 respectively). Continuations
that did not contain reference to the prompt object were excluded from analysis. Generalized
Binominal linear mixed effects analyses were performed with glmer using R (R Core Team, 2017).
Prompt object type (implicit or overt), context type (‘strong’ or ‘irrelevant’) and an interaction
term were included as fixed effects. The maximal random effect structure supported by the data
was used.
104
References to the prompt object were made with pronouns more often when the prompt
object was overt relative to when it was implicit (Tables 3.10, 3.11 and Figure 3.5). Taking
pronominalization rate to be a proxy for degree of accessibility, the fact that implicit objects are
pronominalized less often indicates that they are indeed less accessible in discourse than overt
objects.
In contrast to what was seen in regard to discourse persistence, context had no effect on
the proportion of pronominalizations for either implicit or overt objects.

Table 3.10 Proportion and SE of prompt object pronominalization by condition
Proportion SE
‘Strong’ context + implicit object 0.396 0.071
‘Strong’ context + overt object
0.626 0.051
‘Irrelevant’ context + implicit object 0.393 0.094
‘Irrelevant’ context + overt object
0.621 0.052

Figure 3.5 Proportion of pronominalizations +-1 SE by context type and prompt object type

105

Table 3.11 Results (Accessibility: pronominalization rate): Main effects and interactions (“*”: p < 0.05)
Main Effects Interaction
Object Type Context Type Object Type*Context Type
b SE z p b SE z p b SE z p
0.759 0.282 2.685 0.007* 0.025 0.207 0.124 0.901 -0.045 0.209 -0.218 0.827

Discussion (Discourse Accessibility: Pronominalization rate)

Findings from the story continuation task confirm prior claims that implicit arguments,
including implicit objects, are lower in accessibility relative to overt arguments (e.g., Koenig and
Mauner 1999, Williams 2015). While references to overt objects were pronominalized about 62%
of the time, refences to implicit objects were realized with pronouns about 39% of the time.
Although implicit objects were indeed less accessible, their pronominalization rate was not
negligible. Again, these findings suggest that implicit objects do have a presence in discourse,
albeit a relatively timid one.
Notably, and in contrast to findings pertaining to their persistence, prior context did not
affect the likelihood that implicit or overt objects were referred back to with a pronoun. That is, a
context related to the implicit object increased its likelihood of being referred back to in subsequent
discourse, but it did not influence its accessibility to reduced referring expressions. Although it has
been proposed that these two discourse measures are directly connected (e.g., Arnold 2001), prior
work has shown that some referents can be highly persistent but not highly accessible (e.g., Kehler
et al. 2008, Fukumura & van Gompel 2010, Kaiser 2010, Brocher et al. 2018). The current work
contributes to this body of work and exposes an additional layer in the asymmetry between
persistence and accessibility: for implicit objects, prior context affected the former but not the
106
latter. This finding indicated that at least in the realm of existential implicit arguments,
accessibility and persistence may display different sensitivities to discourse factors such as prior
context.

Results on Discourse Accessibility: Promotion to Subject

Continuations with a clear reference to the prompt object were also annotated and coded
for whether the reference was made in subject position or not. Grammatical subjects are typically
more accessible referents than grammatical objects (e.g., Chafe 1976, Givón 1983, Brennan et al.
1987). Thus, instances when the prompt object is ‘promoted’ to subject position in participants’
continuations can be considered a proxy for referent accessibility.
References to the prompt object were realized in subject position as often for implicit as
for overt prompt objects (Tables 3.12, 3.13 and Figure 3.6). Based on promotions to subject
position, there is no indication that overt objects are more accessible than implicit ones.
Moreover, there was no effect of context on how often implicit and overt objects were
promoted to subject position. This finding once again indicates that prior context does not
influence the accessibility of implicit objects in discourse.

Table 3.12 Proportion and SE of prompt object references realized in Subject position by condition
Proportion SE
‘Strong’ context + implicit object 0.312 0.067
‘Strong’ context + overt object
0.451 0.052
‘Irrelevant’ context + implicit object 0.321 0.089
‘Irrelevant’ context + overt object
0.368 0.052

107

Table 3.13 Results (Accessibility: Promotion to subject): Main effects and interactions (“*”: p < 0.05)
Main Effects Interaction
Object Type Context Type Object Type*Context Type
b SE z p b SE z p b SE z p
0.347 0.222 1.564 0.117 0.110 0.187 0.593 0.553 0.110 0.189 0.580 0.561

Figure 3.6 Proportion of references to the prompt object realized in Subject position +-1 SE by context type
and prompt object type

Discussion (Discourse Accessibility: Promotion to Subject)

References to implicit objects were realized in subject position as often as references to
overt objects. These findings do not provide evidence that overt and implicit objects differ on their
degree of accessibility. This measure – the likelihood that a referent gets referred back to by an
expression in subject position – might not be as sensitive to referents’ accessibility as
pronominalization rate, at least within the context of the current experiment.
108
Note, however, that prompt objects were not promoted to subject position very often: only
between 31% and 45% of all references were realized as the subject of a main clause. This is
compatible with the observation that direct objects in general display relatively lower accessibility
in discourse (e.g., Givón 1983). It is possible that both implicit and overt objects might be at ceiling
in regard to promotion to subject position because there are only two referents in the prompts that
can be referred back to in continuations, and prompt objects are in competition with prompt
subjects for the honor.
Context had no significant effect on implicit and overt objects’ likelihood to be referred
back to with an expression in subject position. The lack of context effects on prompt objects’ rate
of promotion to subject parallels the lack of context effects on the pronominalization rate of prompt
objects: prior context does not seem to play a role modulating referents’ accessibility in discourse.

Exploratory Analysis: Results related to coherence relations effects

Continuations were also annotated for the coherence relation expressed between the
continuation sentence and the previous sentence (following the annotation criteria for coherence
relations from Rohde 2008). For this exploratory analysis, only the first sentence of participants’
continuations was considered because they were directly related to the prompt sentence.
Continuations were annotated for whether they could be inferred to express a coherence relation
such as Elaboration, Explanation, Occasion, Parallel, Result, Violated Expectation or “other” when
the type of coherence relation was unclear/ambiguous or did not fit any of the relations in this set.
First, to understand how context and object type (implicit or overt) affect the likelihood of
any given coherence relation occurring, we examine the distribution of all continuations
109
(regardless of whether they refer to the prompt object or not) by coherence relation, context and
object type (Table 3.14, Figure 3.7).
The vast majority of continuations either expanded on an aspect of the prompt sentence
event (Elaboration) or talked about an event that followed the event expressed in the prompt
sentence (Occasion); other coherence relations appeared in less than 4.1% of all continuations
across conditions.
Both context type and object type seem to influence the proportion of Elaboration and
Occasion coherence relations occurring in continuations: the combination of a context related to
the object (‘strong’) with an implicit object yielded the highest proportion of Elaborations (0.65)
and lower proportion of Occasion (0.26) relations. In contrast, the combination of an unrelated
context (‘irrelevant’) with an overt object yielded the lowest proportion of Elaborations (0.46) and
highest proportion of Occasion relations (0.45). The distribution of Elaboration and Occasion
relations in other conditions fall somewhere in between (see Table 3.14).

Table 3.14 Distribution of all continuations by condition and coherence relation +-1 SE

Elaboration Explanation Occasion Other Result
Violated
Exp.

Mean SE Mean SE Mean SE Mean SE Mean SE Mean SE
‘Strong’ ctx
+ imp.obj
0.654 0.030 0.029 0.010 0.267 0.028 0.016 0.008 0.033 0.011 0 0
‘Strong’ ctx
+ overt obj
0.583 0.031 0.016 0.008 0.333 0.030 0.016 0.008 0.041 0.012 0.004 0.004
‘Irrel.’ ctx +
imp. obj
0.558 0.032 0.041 0.012 0.358 0.031 0.033 0.011 0.004 0.004 0.004 0.004
‘Irrel.’ ctx +
overt obj
0.462 0.032 0.041 0.012 0.450 0.032 0.025 0.010 0.008 0.005 0.008 0.005

110
Figure 3.7 Distribution of all continuations by condition and coherence relation +-1 SE

In general, contexts related to the object lead to more Elaborations, and unrelated contexts
seem to increase the amount of Occasion relations. Additionally, implicit objects lead to more
Elaborations, whereas overt objects relatively increase the amount of Occasion relations.
To analyze the relation between discourse coherence and referents’ persistence, the
distribution of references to the prompt object by coherence relation, context and object type was
examined. In other words, when people wrote continuations that referred back to the prompt
object, what types of coherence relations were used?
Continuations with references to the prompt object were made almost exclusively with
sentences that expressed Elaboration and Occasion coherence relations (see Table 3.15, Figure
3.8), which is expected given that these were the most prevalent coherence relations used overall.

111
Table 3.15 Distribution of references to the prompt object by condition and coherence relation +-1 SE

Elaboration Explanation Occasion Other Result
Violated
Exp.

Mean SE Mean SE Mean SE Mean SE Mean SE Mean SE
‘Strong’ ctx
+ imp.obj
0.654 0.030 0 0 0.324 0.078 0 0 0 0 0 0
‘Strong’ ctx
+ overt obj
0.696 0.055 0 0 0.275 0.054 0 0 0.014 0.014 0 0
‘Irrel.’ ctx +
imp. obj
0.353 0.119 0 0 0.647 0.119 0 0 0 0 0 0
‘Irrel.’ ctx +
overt obj
0.571 0.062 0 0 0.413 0.062 0 0 0 0 0 0

Figure 3.8 Distribution of references to the prompt object by condition and coherence relation +-1 SE

Surprisingly, the distribution of continuations with references to the prompt object by
coherence relation does not exactly mirror the distribution of all continuations by coherence
relation. Specifically, the combination of an ‘irrelevant’ context and an implicit object ‘flipped’
the proportion of Elaboration and Occasion relations: whereas most continuations in this condition
were made with Elaborations (0.55) relative to Occasions (0.35), most references to the prompt
object in this condition were made with Occasions (0.64) relative to Elaborations (0.35). This was
112
also the condition with fewer references to the prompt object overall: 0.11 of all continuations with
references to the prompt object (see Table 3.4, Figure 3.2). In all other conditions, references to
the prompt object were made more frequently with Elaborations relative to Occasion relations.
These patterns suggest that overtness of the object – as well as a ‘strong’ context related to
the object – may both be factors that increase the proportion of references to the prompt
object, mostly with elaboration continuations.
To further investigate the different effects of object and context type by coherence relation,
Elaboration and Occasion continuations were further examined separately.
21

In continuations that expressed Elaboration coherence relations, the same general effects
seen in the data as a whole appear again: people referred back to the prompt object more often
when it was overt relative to when it was implicit, and a context related to the prompt object
(‘strong’ context) increased the proportion of references to the prompt object for implicit objects
but not for overt objects (Tables 3.16, 3.17, 3.18 and Figure 3.9).
In contrast, when looking only at continuations that express Occasion coherence relations
(ex.3.11), context type has no effect on the proportion of references to the prompt object. In
addition, there are more references to the prompt object when it is overt – but only when the context
is not related to the prompt object (Tables 3.16, 3.19, 3.20 and Figure 3.10).

Table 3.16 Proportion of references to the prompt object in Elaboration and Occasion relations +-1 SE

Elaboration Occasion
Mean SE Mean SE
‘Strong’ ctx + imp.obj 0.159 0.029 0.188 0.049
‘Strong’ ctx + overt obj 0.348 0.040 0.238 0.047
‘Irrelevant’ ctx + imp. obj 0.044 0.017 0.128 0.036
‘Irrelevant’ ctx + overt obj 0.330 0.044 0.241 0.041

21
Other coherence relations were excluded from this analysis because they did not occur frequently, and they did
not contain a significant proportion of references to the prompt object.
113
Figure 3.9 References in Elaborations

Figure 3.10 References in Occasions

114
Table 3.17 Results (Exploratory Analysis: Persistence in Elaboration coherence relations): Main effects and
interactions (“*”: p < 0.05)
Main Effects Interaction
Object Type Context Type Object Type*Context Type
b SE z p b SE z p b SE z p
0.769 0.112 6.846 < 0.001* 0.239 0.110 2.116 0.030* -0.177 0.110 -1.607 0.108

Table 3.18 Results (Exploratory Analysis: Persistence in Elaboration coherence relations): Planned pair
comparisons (“*”: p < 0.05)
Contrasts:
Context Type
b SE z p
Implicit objects 0.412 0.171 2.399 0.016*
Overt objects 0.088 0.164 0.535 0.593

Table 3.19 Results (Exploratory Analysis: Persistence in Occasion coherence relations): Main effects and
interactions (“*”: p < 0.05)
Main Effects Interaction
Object Type Context Type Object Type*Context Type
b SE z p b SE z p b SE z p
0.389 0.134 2.888 0.009* 0.168 0.135 1.242 0.214 -0.162 0.134 -1.200 0.229

Table 3.20 Results (Exploratory Analysis: Persistence in Occasion coherence relations): Planned pair
comparisons (“*”: p < 0.05)
Contrasts:
Context Type
b SE z p
‘Irrelevant’ context 0.588 0.003 161.600 p < 0.001*
‘Strong’ context 0.337 0.289 1.164 0.244

Why do we see differences between Elaboration and Occasion coherence relations? One
possibility is that when people choose to talk about what happens next (Occasion), the prompt
event is being conceptualized as a completed event (and therefore more likely to be followed by a
description of a subsequent event). The perceived completeness of an event influences which
components of the event are perceived as more salient: in completed events, the end-state is
115
focused (e.g., Moens & Steedman 1988)
22
. For instance, a completed event of “writing a letter”
emphasizes the letter being finished. Similarly, in cases where “writing” (with no overt object) is
perceived as a completed event, the implicit object (“the something that has been written”) may
be considered more prominent in the discourse (e.g., more overt-like), reducing differences
between implicit and overt objects’ discourse behavior and washing away the effects seen in the
data as a whole/in Elaborations. Another possibility is that the effects do not appear as strongly in
Occasion continuations simply because there are too few of them in the dataset (the majority of
continuations were Elaborations). In any case, the interaction between coherence relations and the
persistence of implicit objects in discourse remains an open question in need of future systematic
empirical investigation.
Unfortunately, examining accessibility measures (i.e., pronominalization and promotion to
subject) in regard to coherence relations is not informative in this dataset, because it requires
further sub-setting the data
23
leaving insufficient instances of pronominalization and promotion to
subject to be meaningfully analyzed.

22
This may be specially the case for creation verbs (11 out of 20 verbs in target items), in which the end-state
culminates with the creation of the implicit object (e.g., “Lisa wrote.” => the something she wrote did not exist before
the event, and it’s created as the event reaches its end-point).
23
The coherence relations analysis splits the data (into Elaboration and Occasion groups); the persistence analysis
above looks at references to the prompt object, and the accessibility analysis would require further zooming into the
proportion of these references made with pronouns/in subject position, which yields small numbers per condition.

116
3.5 General Discussion

Discourse Persistence & Accessibility

Prior work on implicit arguments suggests that they are less prominent in discourse (e.g.,
Goldberg 2001 on implicit patients), that they are less accessible in the discourse model (e.g.,
Koenig and Mauner 1999 on implicit agents) and that they are generally “discourse inert” (e.g.,
Williams 2005). The current chapter expands the investigation on implicit arguments’ discourse
behavior with a story continuation task, to further our understanding of implicit objects’ discourse
properties, such as its accessibility and persistence, as well as how other discourse factors such as
context and discourse coherence contribute to these properties.
The results from the story continuation task revealed that objects that were introduced into
the discourse implicitly are less likely to be referred back to in subsequent discourse (i.e., they are
less persistent) relative to overt objects. That is, participants’ continuations referred back to the
prompt object more often when it had been introduced overtly (e.g., “Later, I saw her reading a
book.”) relative to when it had been introduced implicitly (e.g., “Later, I saw her reading.”). This
finding goes in the expected direction based on prior work on other types of implicit arguments:
Implicit objects are relatively lower in discourse prominence and thus less likely to persist in
discourse.
Note that even when objects were introduced overtly, they were not referred back to very
often in subsequent discourse (less than 40% of continuations). This is not surprising: grammatical
objects are generally seen as less prominent and less accessible (e.g., Givón 1983). The current
study shows that when objects are introduced implicitly they become even less prominent in
discourse.
117
How do the current findings on existential IO’s persistence compare to those from the
previous chapter? The third experiment in Chapter 2 found existential IOs to be as persistent as
overt ones; in contrast, the current chapter found IOs to be less persistent than overt objects. Note,
however, that target sentences in Chapter 2 always had a prior context related to the implicit object,
and only “which noun” expressions could be used to refer back to it. When it comes to the current
chapter, if we only consider continuations that had a context related to the prompt object and which
referred to the prompt object with definite noun phrases (i.e., a parallel informativity level to
“which noun”), we see no significant difference between the persistence of implicit and overt
objects: that is, findings related to persistence are not contradictory across Chapters 2 and 3 – they
are compatible with one another. Moreover, this comparison highlights the observation that despite
their generally lower persistence, there are indeed contexts in which implicit objects are as
persistent as overt ones.
Results from the story continuation task also reveal that implicit objects are less likely to
be referred to with a reduced expression such as a pronoun (i.e., they are less accessible) relative
to overt objects. That is, prompts with overt objects yielded higher rates of pronominalization, i.e.,
of references to the prompt object made with a pronoun, relative to prompts with implicit objects.
This finding matches the accessibility findings from the sluicing studies in Chapter 2, and is
compatible with prior observations that other implicit arguments, such as implicit agents, are
generally less accessible in discourse: implicit objects are less salient/activated in people’s mental
model of the discourse. Consequently, implicit objects are less likely to be referred to with an
expression that indicates a high degree of accessibility in the discourse - like pronouns.
On the other hand, no differences between implicit and overt objects appeared regarding
how frequently they were promoted to subject position. This might be due to the fact that both
118
implicit and overt objects were competing with referents in more salient positions (the prompt
main clause subject “I” and possibly the prompt embedded clause subject, e.g., “Sarah”) for the
prime spot of subject in the continuations, and therefore both implicit and overt objects were not
promoted to subject very often – i.e., objects might have been ‘at ceiling’. Alternatively, the
measure of promotion to subject position may be less sensitive (relative to pronominalization) to
the different accessibility degrees of referents introduced implicitly and overtly.
Crucially, the investigation described in the current chapter shows that implicit objects may
be less persistent and accessible, but they are not entirely “discourse inert”. Although in relatively
low rates, objects introduced implicitly were still referred to with an expression indicating
givenness about 11% of the time when the context was unrelated to the object, and 20% of the
time when the context was related to the object. Of those references, 39% were realized with a
pronoun. These patterns suggest that existential implicit objects’ presence in discourse is not
negligible: albeit timidly and infrequently, implicit objects do participate in the structuring of
subsequent discourse and can sometimes even be referred back to with pronouns. Moreover, their
discourse status can be influenced by broader discourse factors, such as prior context.

Context effects and the relationship between Persistence and Accessibility

The findings from the story continuation task further reveal that broader discourse factors
such as prior context can affect implicit objects’ behavior in discourse. When prior context
introduced a set related to the object (e.g., “I gave Sarah several books to read…”), implicit objects
were referred back to significantly more often than when the previous context was completely
unrelated to the object (e.g., “I went to the gym with Sarah.”). This shows that implicit objects’
119
persistence in discourse is modulated by context: contexts emphasizing the object/providing a
possible identity for the object increase its likelihood of being referred back to in subsequent
discourse.
In contrast, there was no effect of context on overt objects’ likelihood of being referred
back to in continuations. It may be that for implicit objects, people make more use of contextual
cues in determining their discourse persistence; and in contrast, people do not need to rely on these
cues when dealing with overt objects. For example, the related context can promote inferences
about the identity of the implicit object (e.g., Sarah must have been reading one of the books I gave
her), but no inference is needed for overt objects because the identity of the object is
explicit/known.
Notably, context had no effect on the pronominalization rates of neither overt nor implicit
objects. That is, a context related to the prompt object does not affect its accessibility to subsequent
referring expressions. This suggests that accessibility is affected by the form of a referent – their
implicitness – but is not sensitive to broader discourse factors such as prior context. The current
findings also match findings from Chapter 2, which had also not found effects of prior context (set
complexity) on the accessibility of existential IOS. Context also did not affect the rate of
promotions to subject position, which again supports the idea that context modulates implicit
object’s persistence but it does not seem to play a role modulating their accessibility in
discourse.
The fact that prior context affected how often people referred back to the implicit object,
but not how often people used pronouns in doing so, highlights an asymmetry between the notions
of accessibility and persistence in discourse. Other studies have shown that accessibility is not
always tightly linked to persistence (e.g., Kehler et al. 2008, Fukumura & van Gompel 2010,
120
Kaiser 2010, Brocher et al. 2018), and the current findings extend this asymmetry to the domain
of existential referents introduced implicitly. Moreover, the current work presents a novel finding
regarding this asymmetry: that factors external to the referent itself (i.e., prior context) can affect
persistence but not accessibility. These patterns emphasize the need for a better understanding of
how various discourse properties (e.g., prior related/unrelated context) contribute to people’s
assessment of referents’ persistence and accessibility in their mental models of the discourse. In
particular, people may use different heuristics when evaluating the persistence and accessibility of
implicit and overt arguments. For implicit objects, at least, it seems that persistence in discourse
may be more influenced by broader discourse factors, whereas their accessibility is not.

Discourse Coherence

Although the study described in this chapter did not systematically manipulate discourse
coherence, an exploratory analysis was performed to look into whether the coherence relation
inferred to hold between two sentences is related to the discourse behavior of implicit objects. This
additional analysis was conducted based on findings from work on transfer of possession events,
which revealed that coherence relation influences how often each referent from a transfer of
possession event (the source or goal) is talked about in subsequent discourse (e.g., Rohde 2006).
The preliminary findings reported here reveal that coherence relation seems to be related
to implicit objects’ persistence in discourse. The difference between implicit and overt objects’
persistence, as well as the effect of context on implicit objects’ persistence, appeared most clearly
on continuations that expanded on the event described by the prompt (Elaboration), which were
the majority of continuations. In contrast, on continuations that described a subsequent event
121
(Occasion), implicit and overt objects patterned more closely together in terms of how often they
were referred back to. It is not clear why the effects seen in Elaborations seem to fade in Occasion
continuations; it is possible that differences between the effects seen in Elaborations and Occasion
relations may be due to the fact that there were much fewer Occasion continuations in the data.
Future systematic investigation is necessary to further our understanding of how discourse
coherence interacts with implicit arguments’ behavior in discourse.
In sum, the current work utilized a story continuation task to better understand how
existential implicit objects participate in people’s mental models of the discourse. This task reveals
that while implicit objects are indeed less accessible and persistent in discourse relative to their
overt counterparts, they do timidly participate in the structuring of discourse and can sometimes
be referred to with reduced expressions like pronouns. Crucially, a prior context related to the
object affects implicit objects’ discourse persistence but not their accessibility, highlighting the
asymmetry between these two measures and the importance of investigating both to get a more
thorough picture of referents’ life in discourse. Finally, an exploratory analysis on coherence
relations suggests that discourse coherence may be a crucial ingredient in how people build mental
models of the discourse, influencing the discourse status of referents like implicit objects.

122
Chapter 4 : Implicit objects in comprehension: Effects of context-dependency

4.1. Introduction

Chapters 1-2 present investigations on the discourse behavior of existential implicit objects
(existential IOs), which belong to the class of existential implicit arguments (existential IAs).
Existential IAs have been typically thought of as ‘discourse inert’ (e.g., Koenig & Mauner 1999,
Williams 2015). The findings reported in Chapters 1-2 reveal that existential IOs are indeed
generally less accessible and persistent in the discourse relative to overt objects, but they are not
entirely inert and participate (albeit timidly) in the structuring of subsequent discourse.
Chapter 3 expands on Chapters 1-2 by investigating the underexplored domain of implicit
objects in real-time language comprehension. Specifically, this chapter presents an inquiry into
how the comprehension of Existential IOs differs from the comprehension of another kind of
implicit object, anaphoric IOs.
Existential and anaphoric implicit arguments contrast in a crucial way: whereas the first
are considered context-independent because they do not require any pre-existing context for
interpretation, the latter are context-dependent and require an available referent in context for
successful interpretation (e.g., Fillmore 1969, Condoravdi & Gawron 1996, Anderbois 2012).
The current chapter investigates effects of context-dependency on the comprehension of
implicit objects in Brazilian Portuguese (BP), a language in which both context-independent
(existential) and context-dependent (anaphoric) implicit objects with similar semantic properties
are common. Although these implicit direct objects share an identical unpronounced/invisible
form, their discourse functions are quite different: one requires a referent in discourse to supply its
meaning, whereas the other does not.
123
A self-paced reading task was designed to test whether processing context-dependent,
anaphoric implicit objects without an antecedent/referent available in context leads to
comprehension difficulties (seen in a slowdown in reading times) relative to when an antecedent
is available. Based on their context-dependence properties, it is expected that when people
encounter an anaphoric, context-dependent IO without a supporting context (i.e., without an
available antecedent), they will face comprehension difficulties. On the other hand, when people
encounter an existential, context-independent IO without a referent available, no difficulties should
arise during comprehension because its interpretation does not require a referent.
This self-paced reading task also allowed for an additional exploratory inquiry on the
comprehension of anaphoric null objects: specifically, what happens when its antecedent is an
existential implicit object. As Existential IOs are typically less accessible in discourse (see
Chapters 1-2), referring back to an Existential IO with an anaphoric IO could yield comprehension
difficulties relative to referring back to overt object antecedents.
This chapter is structured as follows: section 4.2 describes a crucial notion that
distinguishes two major classes of implicit arguments: context-dependency. This section also goes
over cases of context-dependent and context-independent implicit objects in English and in
Brazilian Portuguese (note that this section partly recapitulates notions previously introduced in
Chapters 1-3). Section 4.3 presents a self-paced reading experiment designed to investigate effects
of context-dependency during the comprehension of implicit objects. Section 4.4 concludes with
a general discussion on the findings.

124
4.2. Context-dependency and Implicit Arguments

Linguistic expressions can depend on context for their interpretation or not; thus they can
be broadly classified as context-dependent or context-independent (e.g., see Condoravdi &
Gawron 1996, Fillmore 1969, Anderbois 2012).
In formal terms, “context-dependency is traditionally taken to involve the context of
utterance and meanings are specified as functions from contexts to content” (Condoravdi &
Gawron 1996; see also Cresswell 1973, Kaplan 1989, Lewis 1981, Stalnaker 1972). That is,
context-dependent expressions are those for which successful meaning interpretation relies on
values provided by the context of the utterance, including previous discourse. Some examples of
overt context-dependent expressions are deictic and demonstrative pronouns (e.g., “that”), as well
as pronominal anaphora (e.g., “she”). In contrast, expressions such as the indefinite noun phrase
“a woman” do not require context for interpretation and thus are considered context-independent.
The domain of implicit arguments also contains expressions that are context-dependent or
context-independent. For instance, in English, implicit agents of short passives (e.g. the ‘someone’
who did the sinking in (ex.4.1)) and existential implicit objects (as the ones from Chapters 1-2; the
‘something’ being eaten in (ex.4.2)) typically do not require any further context for successful
interpretation. That is, these implicit arguments are normally understood as an unspecified
‘someone’ or ‘something’ and their meaning does not need to rely on any value provided by
context.

(4.1) The ship was sunk.

(4.2) Ana was eating.
125

In contrast, some implicit arguments are context-dependent. For example, the implicit
arguments of verbs such as “win” and “apply” (see ex.4.3a, 4.4a; examples adapted from
Anderbois 2012) derive their interpretation from some value provided by context (e.g., Fillmore
1969, Partee 1989, Condoravdi & Gawron 1996, Anderbois 2012). In (ex.4.3a), it is understood
that Moira won the bet she placed; in (ex.4.4a), that Alexis applied to the job opening she had
found out about the day before.
When no referent is supplied by context (ex.4.3b, 4.4b), sentences with context-dependent
implicit arguments are not felicitous. That is because these implicit arguments have no
referent/antecedent to supply the value to derive their meaning.

(4.3) a. Moira placed a bet and won.

b. #Moira won. [when presented with no prior context]

(4.4) a. Alexis found out about a job opening and the very next day she applied.

b. #Alexis applied. [when presented with no prior context]

How does the perceived infelicity of such sentences (ex.4.3b, ex.4.4b) relates to how they
are comprehended by the language processor? In the current chapter, I explore the notion that the
root of this infelicity – i.e., the absence of a suitable antecedent for the interpretation of an
anaphoric IO – may also cause difficulties during comprehension. This idea is further explained
126
below in Section 4.2.1.

A note on inferencing

The contrast between context-dependent and context-independent implicit arguments may
sometimes seem hazy due to the fact that people can make context-based inferences about the
identity of a context-independent implicit argument (for more on inferences, see e.g. Trabasso &
Van Den Broek 1985, McKoon & Ratcliff 1992, Birner 1997). That is, although context-
independent expressions do not require context/a referent for interpretation (unlike context-
dependent expressions), their interpretation can be influenced by the context via inferencing.
24
For
example, in (ex.4.5a), although the ‘something’ implicit argument of “ate” can felicitously be
interpreted independent of context as in (ex.4.2), it can also be inferred to match or include the
previously-mentioned piece of bread on the table (e.g., “John ate something, and this something
John ate could have been/could have included the piece of bread on the table).

(4.5) a. There was a piece of bread on the table. John ate.

b. There was a piece of bread on the table. John ate it.

It is relevant to note that these inferences are not the same as an anaphoric interpretation
such as the one that occurs with context-dependent expressions, e.g. with the overt pronoun “it” in

24
As seen in Chapter 2, existential implicit objects’ discourse behavior was affected when prior context allowed
inferences about the identity of the implicit object (e.g. “I gave Sarah some books… later I saw her reading => she
might have been reading one of the books I gave her).

127
(ex.4.5b). Contrast (ex.4.5a) with (ex.4.5b): in the latter, the pronoun is interpreted as coreferential
with the previously-mentioned piece of bread. In this case, it is clear that John ate the piece of
bread, and unlike (ex.4.5b), the sentence is not compatible with John having eaten something else.
When it comes to implicit expressions, the contrast between how context-dependent and
context-independent implicit arguments are interpreted becomes clearer when they are inspected
under negation (ex.4.6a-b) from Condoravdi & Gawron 1986):

(4.6) a. There was a piece of bread on the table but John didn’t eat.

b. There was a good job available here but Fred didn’t apply.

(4.7) a. #There was a piece of bread on the table but John didn’t eat; he only ate chips.

b. There was a good job available here but Fred didn’t apply; he only applied to a
volunteer position in another country.

Example (ex.4.6a) shows that even when an appropriate referent is available (e.g., “piece
of bread”), the context-independent implicit argument of “eat” under negation is interpreted as
“anything” (e.g., “There was a piece of bread on the table but John didn’t eat anything). That is,
the negation does not allow for the inference that the ‘something’ being eaten is the previously-
mentioned piece of bread (like in (ex.4.5a)), because in this case there is no ‘something’ that has
been eaten. In fact, it would be infelicitous to follow such a sentence by declaring that John did, in
fact, eat something other than the piece of bread (ex.4.7a).
128
In contrast, (ex.4.6b) shows that when a referent is available, even under negation, the
context-dependent implicit argument of “apply” is still interpreted anaphorically (e.g., “There was
a good job available here but Fred didn’t apply to this good job available here). This interpretation
is perfectly compatible with a continuation that indicates that Fred applied to a different position
(ex.4.7b), and it parallels the also context-dependent interpretation with overt pronouns (e.g.,
“There was a good job available here but Fred didn’t apply for it”).
In sum, two points are most relevant here: (i) context-independent implicit arguments do
not require a referent for interpretation, but context-dependent implicit arguments do; and (ii)
although sometimes inferences like in (ex.4.5) are possible, context-independent implicit
arguments are not interpreted anaphorically, but context-dependent arguments are (e.g., Fillmore
1969, Anderbois 2012). The latter point is made evident by the examples in which the implicit
argument occurs under the scope of negation.
Based on the differences between the interpretation of context-dependent and context-
independent implicit arguments, the current work investigates context effects on the
comprehension of existential and anaphoric implicit objects.
In the realm of overt referring expressions, context-dependent expressions like the
pronouns “she” and “he” can generate comprehension difficulties in the absence of an available
antecedent (e.g., Van Berkum et al. 2004, Nieuwland and Van Berkum 2006, Sanford et al. 2008,
Filik et al. 2008). These pronouns are interpreted in relation to a value acquired from context, e.g.
a value provided by an antecedent expression. Experimental studies have shown that when no such
value is encountered in the comprehension of “she/he” pronouns, difficulties ensue. For instance,
event-related potential (ERP) studies investigated the comprehension of sentences with “she/he”
antecedentless pronouns such as “Anna shot at Linda as he jumped over the fence” (in which there
129
are only female referent antecedents, i.e. no suitable referents for the interpretation of the pronoun
“he”). They found that these cases lead to an event-related potential known as P600 (e.g., Van
Berkum et al. 2004, Nieuwland and Van Berkum 2006), an effect that typically arises in the
occurrence of comprehension difficulties related to syntactic phenomena (see Hagoort et al.
1999).
25

Other studies looked at sentences in which there was no explicit antecedent for the
pronouns “she/he” at all, only an antecedent that could be inferred from prior context: e.g., “The
in-flight meal I got was more impressive than usual. In fact, she courteously presented the food…”.
These sentences lead to slower reading times (e.g., Sanford et al. 2008), as well as other event-
related potential effects indicating difficulty (e.g., Filik et al. 2008).
The latter set of studies also compared the comprehension of “she/he” with the plural
pronoun “they”, which can be used both to refer to a specific referent (like “she/he”) and in more
generic senses with no referent specified. One such generic case is known as ‘Institutional they’,
such as in “The in-flight meal I got was more impressive than usual. In fact, they courteously
presented the food…” (example from Filik et al. 2008). The studies found that comprehension of
the pronoun “they” without explicit antecedents did not generate difficulties, in contrast with the
pronouns “she/he” (e.g., Sanford et al. 2008, Filik et al. 2008).
The current chapter expands on prior work by investigating effects of context-dependency
in the comprehension of implicit expressions. Anaphoric implicit objects are context-dependent,
i.e. they require an antecedent for interpretation. Thus, I hypothesize that they may parallel the

25
It is unclear whether the P600 effect occurs because there is no suitable referent for the pronoun “he” or due to an
attempt at interpreting the pronoun as referring back to a referent that does not match in gender (which could be a ‘last
resort’ approach in cases where no suitable referent has been found). In any case, these findings suggest that processing
these context-dependent pronouns in contexts that lack appropriate antecedents generates an additional burden during
comprehension.
130
context-dependent pronouns “she/he”: that is, that the comprehension of anaphoric implicit objects
without a suitable referent in context will generate difficulties. On the other hand, existential
implicit objects are context-independent and do not require antecedents for interpretation, thus
they may parallel the comprehension of the pronoun “they”. Thus, it is not expected that processing
existential implicit objects without antecedents will generate comprehension difficulties.
In order to investigate the context-dependency of implicit objects, an experiment was
designed in Brazilian Portuguese, a language in which both context-dependent and context-
independent direct objects of similar semantic properties are common.
4.2.1. Existential and Anaphoric implicit objects in English and Brazilian Portuguese

Although English has cases of both context-dependent and context-independent implicit
objects, the former is rarer and varies widely. For example, the implicit object of a verb like “won”
is better paraphrased with a noun phrase (ex.4.8a; implicit object is paraphrased between brackets),
whereas the IO of “apply” is better paraphrased with a prepositional phrase, that is, it’s actually an
indirect object (ex.4.8b).
English context-dependent IOs also vary in their thematic roles. Whereas the vast majority
(if not all) of context-independent IOs are themes of the verb, context-dependent IOs are more
difficult to classify. Anderbois (2012) proposes some generalizations about the thematic role
distribution of implicit arguments in English: he suggests that Existential (context-independent)
implicit arguments can be themes or agents, and that Anaphoric (context-dependent) implicit
arguments do not occur in those same roles. Anderbois classifies the implicit argument of “win”
in (ex.4.8a), for example, as ‘bet’ or ‘championship’; and the implicit argument of “found out”
(ex.4.8c) and “noticed” (ex.4.8d) as ‘state of affairs’ (examples adapted from Anderbois 2012):
131

(4.8) a. Moira placed a bet and won [the bet].

b. Alexis found out about a job opening and applied [to/for the job].

c. David had a secret and Stevie found out [that David had a secret/about David’s
secret/what David’s secret was].

d. Twyla got a haircut and Johnny noticed [that Twyla got a haircut/Twyla’s haircut].

In contrast to English, context-dependent IOs in Brazilian Portuguese are typically themes
of the verb; as are context-independent IOs. Thus, in order to systematically investigate context-
dependency effects on the comprehension of implicit objects, the current chapter turns to Brazilian
Portuguese (BP), a language that provides a more direct comparison between context-independent
existential implicit objects and context-dependent anaphoric implicit objects.
In BP, existential implicit objects are similar to those in English (e.g., as seen in Chapters
1-2): they are the implicit objects of optionally transitive verbs such as “eat”, “read”, “cook”, etc.
Just like English existential IOs, BP existential IOs can be interpreted independently of context,
are typically paraphrased with ‘something’, and perform the role of themes of the verb.
There is still debate on the literature regarding the syntactic nature of these anaphoric IOs
in BP: some suggest that they are covert pronouns (e.g., Farrell 1990, Kato 1993) whereas others
argue they are ellipsis of a determiner phrase (e.g., Cyrino 1996, Cyrino & Lopes 2016). Although
an interesting question, the syntactic status of anaphoric IOs in BP is not consequential for the
132
current investigation. The key property relevant to this dissertation is that these ‘null objects’ are
phonologically null (‘phonologically unrealized’), and thus share this surface property with
existential implicit objects.
Crucial to the purposes of this work, anaphoric implicit objects in BP are context-
dependent; that is, their meaning is interpreted through a value/referent provided by context
(ex.4.9a). So, when a transitive verb such as “devour” appears without an overt object and there’s
an available referent in context, the implicit object is interpreted anaphorically: (ex.4.9a) is akin to
“Julia received a cake and devoured it quickly”. Without a referent, anaphoric IOs in BP are not
felicitous (ex.4.9b).

(4.9) a. Julia recebeu um bolok e devorou ∅k rapidinho.
Julia received a cakek and devoured ∅k quickly.

b. #Julia devorou ∅ rapidinho.
Julia devoured ∅ quickly.

Optionally transitive verbs like “eat” can also have anaphoric IOs if a referent is available
in context (ex.4.10a); in this example, the sentence is interpreted similarly to “Julia received a cake
and ate it quickly”. This anaphoric interpretation survives even under negation: (ex.4.10b) is
interpreted as “Julia received a cake but didn’t eat it/the cake she received.”
However, when no referent is present, optionally transitive verbs with no overt objects can
have existential IOs and are thus felicitous; in this case, just like in the previously examined
English examples, the implicit object is interpreted as an unspecified ‘something’. That is,
133
(ex.4.10b) is interpreted as “Julia ate something quickly”.
For a summary of the kinds of implicit objects allowed by each verb type, see Table 4.1.
26

(4.10) a. Julia recebeu um bolok e comeu ∅k rapidinho.
Julia received a cakek and ate ∅k quickly.

b. Julia recebeu um bolok mas não comeu ∅k.
Julia received a cakek but didn’t eat ∅k.

c. Julia comeu ∅ rapidinho.
Julia ate ∅ quickly.

Table 4.1 Summary of the types of implicit objects allowed per verb type.
Optionally Transitive (OPT) verbs
e.g., “comer” (i.e. “to eat”)
Obligatorily Transitive (OT) verbs
e.g., “devorar” (i.e. “to devour”)
Existential IO ✓ x
Anaphoric IO ✓ ✓

The current work takes advantage of BP properties to investigate effects of context-
dependency on the comprehension of implicit objects. With obligatorily transitive verbs (OT; e.g.,
“devour”), only anaphoric IOs are possible. Therefore, when a referent is available in context, a

26
It is relevant to note that anaphoric implicit objects and overt pronouns in BP are not in free distribution, but rather
seem to display a certain division of labor: whereas overt pronouns like “ele” (he/him) and “ela” (she/her) are most
often used with animate antecedents (e.g., as patients/experiencers thematic roles), anaphoric IOs are generally used
with inanimate ones (e.g., as themes). There are also indications that other properties beyond animacy, such as
specificity, influence the choice between an overt pronoun and an anaphoric IO in BP (e.g., Cyrino 1996, Schwenter
& Silva 2002). In the experiment described in Section 4.3, antecedents were designed so that they were prototypical
anaphoric IO antecedents, which also match the properties of existential IOs (i.e., unspecific, inanimate, themes).

134
coreferential reading of the implicit object should occur (ex.4.10a). In contrast, when no referent
is available, there is no value from context to supply a meaningful interpretation of the anaphoric
IO, rendering it infelicitous (ex.4.10b).
On the other hand, with optionally transitive verbs (OPT; e.g., “eat”), both context-
independent existential IOs and context-dependent anaphoric IOs are possible. Thus, when a
referent is available, again a coreferential interpretation occurs (ex.4.11a). When no referent is
available, the implicit object can be interpreted existentially (independent of context, as ex.4.11b).
Because both types of implicit objects are identically invisible/unpronounced, it is expected
that the language parser must be able to use information about the verb (e.g., whether it is an OT
or OPT verb) to determine which kinds of IOs are allowed in a given context.
Crucially, based on previous findings related to the comprehension of overt pronouns in
the absence of antecedents (e.g., Van Berkum et al. 2004, Nieuwland and Van Berkum 2006,
Sanford et al. 2008, Filik et al. 2008), I hypothesize that processing sentences with OT verbs and
no available referent in context will lead to comprehension difficulties. That is because OT verbs
can only have anaphoric IOs, which are context-dependent. Thus, they may behave like the
context-dependent pronouns “she/he”, which have been found to cause processing difficulties
when there is no suitable referent in context. On the other hand, processing sentences with OPT
verbs and no available referents should not generate a processing burden. OPT verbs can have both
context-dependent and -independent IOs; thus, they may behave similarly to the pronoun “they”,
which does not generate comprehension difficulties in the absence of antecedents.
Whereas this work’s main research question pertains to the comprehension of IOs with no
antecedents, an exploratory question is also investigated regarding the comprehension of
anaphoric IOs with implicit antecedents. As discussed, anaphoric IOs with overt referent
135
antecedents are perfectly felicitous (ex.4.10a repeated below as ex.4.11a). In contrast, anaphoric
IOs with no antecedents are infelicitous (ex.4.10b repeated below as ex.4.11b). What happens,
though, when the only antecedent available is implicit? More specifically, in (ex.4.11c), the
anaphoric IO of the verb “devour” can only refer back to the existential IO of the verb “cook”: i.e.,
meaning Julia cooked something and quickly devoured the something she cooked.

(4.11) a. Julia recebeu um bolok e devorou ∅k rapidinho.
Julia received a cakek and devoured ∅k quickly.

b. #Julia devorou ∅ rapidinho.
Julia devoured ∅ quickly.

c. ??Julia cozinhou ∅k? e devorou ∅k? rapidinho.
Julia cooked ∅k? and devoured ∅k? quickly.

Does the comprehension of anaphoric IOs with existential implicit antecedents lead to
comprehension difficulties relative to the comprehension of anaphoric IOs with overt antecedents?
Chapters 1-2 indicate that existential IOs are generally less accessible than overt noun phrases to
subsequent referring expressions. This is compatible with prior work showing that other implicit
arguments, such as implicit agents, are also less accessible in discourse (e.g., Koenig & Mauner
1999). Thus, an existential IO such as the one for “cooked” in (ex.4.11c) should be less accessible
to the subsequent anaphoric IO relative to the overt antecedent “a cake” (ex.4.11a), which could
potentially lead to difficulties in retrieving the antecedent during comprehension.
136
However, prior experimental work suggests that the comprehension of reference to implicit
antecedents does not necessarily lead to difficulties during comprehension. In particular, Cornish
et al. (2005) found that comprehension difficulties related to reference to implicit antecedents is
modulated by how central the antecedent is in relation to the event described. In their definition,
a central or nuclear implicit referent is conceptually closely related to an event and fills an
argument slot: such is the case with existential implicit objects. Peripheral implicit referents, on
the other hand, are those that would be typically expressed by adverbials and are not as tightly
conceptually linked to the event: for example, implicit instruments (e.g., such as the instrument
related to an event of ‘cutting’). Cornish et al. (2005) conducted two sentence-by-sentence reading
experiments
27
in French and English and found that pronominal reference to implicit antecedents
only caused difficulties (as seen in slower reading times) when the antecedent was peripheral. In
contrast, pronominal reference to nuclear implicit antecedents were read in a similar pace to
pronominal reference to overt antecedents. That is, pronominal reference to nuclear implicit
antecedents – like existential implicit objects - did not cause comprehension difficulties relative to
pronominal reference to overt antecedents.
Cornish et. al’s (2005) results were replicated by Järvikivi et al. (2019) in a self-paced
listening experiment in German: they found a processing cost when the pronoun antecedent was
implicit and peripheral, but not when it was implicit and nuclear (relative to overt antecedents).
Whereas prior work focused on reference made with overt pronouns, the exploratory
inquiry presented in this chapter investigates reference made with an implicit referring expression,
the anaphoric implicit object. Thus, this line of inquiry explores whether the reference to existential

27
In which the time it takes the read a whole sentence is measured in milliseconds.
137
implicit objects made with anaphoric implicit objects follows the same patterns as those made
with overt pronouns (as seen in Cornish et al. 2005 and Järvikivi et al. 2019).
Thus, this work also tests whether processing ease of antecedent-demanding IOs depends
on whether the antecedent is overt or implicit. Although existential implicit objects are relatively
less accessible for subsequent reference especially with reduced expressions (as seen in Chapters
1-2), psycholinguistic work suggests that their relative lower degree of accessibility may not
necessarily contribute a burden to comprehension processes (e.g., Cornish et al. 2005, Järvikivi et
al. 2019). If anaphoric implicit objects behave similarly to other context-dependent expressions
like overt pronouns “she/he”, it is expected that processing of context-dependent anaphoric
implicit objects with implicit (less accessible) antecedents will not be more difficult than overt
(more accessible) ones.
In sum, the experiment presented in this chapter investigates two facets of anaphoric
implicit objects’ context-dependence during comprehension: effects caused by the complete
absence of an antecedent (main “no antecedent” inquiry), as well as effects caused by the relatively
lower accessibility of implicit antecedents (exploratory “existential implicit object antecedent”
inquiry).
The current experiment also taps into existential implicit objects’ context-independence in
predicting no difficulties in processing contexts without antecedents (main inquiry), and in
examining the effects of existential implicit objects’ lower accessibility to reduced referring
expressions during comprehension (exploratory inquiry).

138
4.3. Investigating effects of context-dependency during the comprehension of implicit objects
in a self-paced reading task

A word-by-word self-paced reading task in Brazilian Portuguese was designed to
investigate the effects of context-dependency on the comprehension of implicit objects. In this
experiment, verb type is manipulated: verbs were either obligatorily transitive (OT) or optionally
transitive (OPT) verbs. OT verbs without overt objects can only have anaphoric implicit objects,
whereas OPT verbs without overt objects can have either anaphoric or existential implicit objects
(see Table 1). The availability of an antecedent was also manipulated (either available or
unavailable).
It is expected that the comprehension of anaphoric IOs without available antecedents will
yield slower reading times following the verb, relative to when an antecedent is available. That is
because the interpretation of a context-dependent implicit object without a referent to provide its
meaning will lead to comprehension difficulties (as seen in slower reading times), similarly to
what has been found regarding the comprehension of “she/’he” pronouns without antecedents (e.g.,
Van Berkum et al. 2004, Nieuwland and Van Berkum 2006, Sanford et al. 2008, Filik et al. 2008).
In contrast, when both existential and anaphoric IOs are possible but there is no available
antecedent, it is expected that people will have no comprehension difficulties because an
existential IO interpretation does not require a referent to provide its meaning. This expectation is
parallel to what has been found regarding the comprehension of the pronoun “they”, which can
occur in contexts without specific antecedents (e.g., “institutional they”) without causing
processing difficulties (e.g., Sanford et al. 2008, Filik et al. 2008).
In addition, this experiment includes an exploratory investigation on the comprehension of
anaphoric implicit objects with existential implicit object antecedents. As prior work has found
139
that pronominal reference to nuclear implicit antecedents does not cause an extra processing
burden, it is expected that reference to existential implicit objects made with anaphoric implicit
objects as the referring expression will also not generate comprehension difficulties – despite their
relative lower accessibility in discourse.

Participants

47 native speakers of Brazilian Portuguese participated in the experiment in Rio de Janeiro,
Brazil. 2 participants were excluded from analysis due to low performance on filler comprehension
questions (less than 80% accuracy), leaving a total of 45 participants.
4.3.1. Methods

Materials and Design

The experiment contained 20 targets and 32 filler items. Target sentences are provided in
Appendix 5. They were manipulated so that they either had an obligatorily transitive (OT) or an
optionally transitive (OPT) verb. OT verbs (e.g., folhear, meaning “to flip through”) allow only
anaphoric implicit objects, whereas OPT verbs (e.g., ler, “to read”) allow both anaphoric and
existential implicit objects. In addition, the availability of an antecedent in context was also
manipulated: sentences either had an antecedent available or they did not.
These manipulations allow for the investigation of our “no antecedent” inquiry: does the
comprehension of context-dependent anaphoric implicit objects with no supporting context (i.e.,
no available antecedent) leads to comprehension difficulties, relative to when an antecedent is
140
available?
Sentences with OT verbs and no available antecedents are expected to generate slower
reading times immediately following the verb relative to sentences with available antecedents,
because OT verbs can only yield context-dependent implicit objects. This expectation is guided by
findings showing that processing overt context-dependent referring expressions like “she/he” with
no available antecedents leads to difficulties in comprehension (e.g., Van Berkum et al. 2004,
Nieuwland and Van Berkum 2006).
In contrast, sentences with OPT verbs and no available antecedents are not expected to
generate difficulties relative to those with available antecedents. That is because OPT verbs can
have both context-dependent and -independent implicit objects. Thus, the comprehension of OPT
verbs with implicit objects is expected to mirror the processing of overt expressions like the
pronoun “they”, which does not lead to difficulties when it occurs in contexts without antecedents.
To address this question, reading times at the critical region (starting at the OPT/OT verb,
e.g. ‘read’ or ‘flip-through’, plus three following words as the spill-over region; underlined in
Table 4.2) were analyzed.
In addition, target sentences had another verb which was always OT (e.g., guardou, “put
away”
28
). This verb allows for our exploratory inquiry on “Existential implicit object antecedents”:
does the comprehension of context-dependent (anaphoric) implicit objects with existential IO
antecedents (Table 4.2, ex.4.12) leads to comprehension difficulties, relative to when they refer
back to overt antecedents (Table 4.2, ex.4.13)?
Crucially, in regard to this exploratory inquiry, only conditions in which the initial verb is
optionally transitive (OPT) are relevant. Only OPT verbs yield existential implicit objects, which

28
In the example here, guardou means ‘put away’; in other contexts it can also mean ‘saved’ or ‘kept’.
141
are the antecedent of interest in this exploratory inquiry. Thus, the contrast of interest here is
between OPT conditions without antecedents (ex.4.12), which yield an existential implicit object
as the antecedent for the subsequent final verb “guardou”, and OPT conditions with an antecedent
(ex.4.13), which offer an overt referent as the antecedent for “guardou”. Consequently, conditions
with initial OT verbs are not informative for this line of inquiry
29
.
Moreover, because the OPT condition with an overt antecedent (ex.4.13) contains both (i)
an overt referent antecedent and (ii) an anaphoric IO (the implicit object of the OPT verb which
refers back to the overt antecedent), an additional control condition with only an overt antecedent
and no IO antecedent was included (ex.4.16). To address this exploratory question, reading times
at the final OT verb and the three following words were analyzed.
In sum, target sentences had the following manipulations: (i) verb type (either optionally
transitive, i.e., OPT or obligatorily transitive, i.e., OT) and (ii) antecedent (either available or
unavailable). An additional ‘control’ condition with an intransitive verb (e.g., ‘sit’) instead of an
OPT or OT verb was included to address the exploratory question.
Table 4.2 displays an example target item in all conditions: the critical region for the “no
antecedent” inquiry is underlined (it begins at the verb and includes the three following words as
the spill over region to detect effects that may ‘spill over’ beyond the verb). The critical region for
the exploratory question on “existential implicit object antecedents” begins at the final OT verb
(e.g., guardou, “put away”) and includes the three following words as spill over.

29
This is fortunate because it is precisely the OT conditions that are expected to yield effects that could potentially
‘carry over’ from one critical region to another.
142
Table 4.2 Example target items (BP on top, English translation below; ∅ indicates an IO. The critical region
for the main question is underlined. The critical region for the exploratory question begins at “guardou” and
includes the threw following words)
4.12. OPT verb, no antecedent

A Clara entrou, leu no sofá da sala, depois guardou na estante e saiu.
Clara entered, read ∅ on the couch of living-room, then put-away ∅ on shelf and left.
4.13. OPT verb, with antecedent

A Clara escolheu um livro, leu no sofá da sala, depois guardou na estante e saiu.
Clara picked a book, read ∅ on the couch of living-room, then put-away ∅ on shelf and left.
4.14. OT verb, no antecedent

A Clara entrou, folheou no sofá da sala, depois guardou na estante e saiu.
Clara entered, flipped-through ∅ on the couch of living-room, then put-away ∅ on shelf and
left.
4.15. OT verb, with antecedent

A Clara escolheu um livro, folheou no sofá da sala, depois guardou na estante e saiu.
Clara picked a book, flipped-through ∅ on the couch of living-room, then put-away ∅ on shelf
and left.
4.16. Control

A Clara escolheu um livro, sentou no sofá da sala, depois guardou na estante e saiu.
Clara picked a book, sat on the couch of living-room, then put-away ∅ on shelf and left.

Each sentence was followed by a yes/no comprehension question. Comprehension
questions on target trials were intended to assess participants’ interpretation of implicit objects.
For example, after reading a sentence such as “Clara grabbed a book, read on the couch and put
away.”, participants were asked “Did Clara read a book?” (following prior work, e.g., Christianson
et al. 2001). This question taps into whether participants interpreted the implicit object of ‘read’
anaphorically or not (‘yes’ and ‘no’ answers, respectively). This assessment is particularly relevant
in the case of sentences with OPT verbs, which can have both anaphoric and implicit objects: “yes”
answers indicate that although both possibilities are available, the implicit object was interpreted
143
anaphorically. In the additional control condition, the question assessed the interpretation of the
second verb’s IO (“Did Clara put away a book?”).

Procedure

In a self-paced reading task, participants read sentences word-by-word on a computer
screen at their own pace. Words are hidden by dashes until participants press the spacebar,
revealing one word with each pressing of the spacebar (previous words return to being covered by
dashes; only one word is visible at a given time). This method allows for the measurement of
reaction times in milliseconds, i.e. how long it takes for the participant to read each word (e.g.,
Aaronson & Scarborough 1976; Mitchell & Green 1978). The experiment was programmed on
Linger
30
entirely in Brazilian Portuguese.
4.3.2. Results and Discussions

Reading times and answers to the comprehension questions were analyzed.
First, this section reports on the analysis on reading times regarding the main question:
does the comprehension of context-dependent anaphoric implicit objects with no
antecedent/referent available in context leads to comprehension difficulties?
Secondly, I turn to reading times related to the exploratory question: does the
comprehension of anaphoric implicit objects that refer back to existential implicit objects leads to
comprehension difficulties, relative to when they refer back to overt objects?

30
Developed by TedLab and Doug Rohde (http://tedlab.mit.edu/~dr/Linger/), now supported by Lauren Ackerman
(as of 2019; https://github.com/VerbingNouns/linger3).
144
Finally, I report participants’ answers to the comprehension questions assessing
interpretation of the implicit objects.
Section 4.4. presents a general discussion.

Results: Reading Times for “No Antecedent” inquiry

Extreme reading times (RTs faster than 100ms or slower than 4000ms) were excluded from
analysis (.01% of total data). RTs +/- 2.5 standard deviations from the mean for any given word
were transformed to 2.5 standard deviations from the mean (2.33% of total data). RTs were
analyzed with mixed-effects regression using the lme4 package (Bates, Maechler, Bolker &
Walker 2019) in R (R Core Team 2019), with random slopes and intercepts for subjects and items
when supported (Barr 2013).
The critical region for testing effects of the absence of antecedents begins at the OPT/OT
verb and includes the three following words as the spillover region (e.g., “ler/folhear no sofá da”;
see Table 4.2, underlined).
RTs for the three spill-over words of the critical region (e.g., “no sofá da”) appear in the
predicted direction: in sentences with OT verbs – which can only yield anaphoric implicit objects
–, there is a slowdown when no antecedent is available relative to when an antecedent is available
(Table 4.3, Figure 4.1). These findings confirm the prediction that comprehension of context-
dependent IOs without the supporting context for interpretation leads to comprehension
difficulties. On the other hand, in sentences with OPT verbs - which allow both anaphoric or
existential implicit objects – we see no effect of the presence/absence of antecedents. The lack of
any slowdowns in the no-antecedent condition indicates that with OPT verbs, people can interpret
145
the implicit object as a context-independent existential IO.

Figure 4.1 Word-by-word reading times +-1 SE by verb and antecedent type (English translation: The Clara
[entered/selected a book], [read/flipped through/sat] on couch of living-room, then put-away on shelf and left)

Table 4.3 Mean RTs (in milliseconds) and SE for words in the critical region by verb and antecedent type
Word
OPT + no antec

OPT + with antec OT + no antec OT + with antec
Mean RT SE Mean RT SE Mean RT SE Mean RT SE
Verb
(leu/
folheou)
406.620 13.672 407.911 13.813 420.703 13.197 412.974 13.078
“no” 368.950 11.073 356.327 9.586 379.883 11.785 357.887 9.822
“sofá” 379.106 12.399 368.059 11.592 404.706 14.827 375.289 11.868
“da” 381.823 10.112 365.411 8.684 410.665 11.817 368.974 9.857

Table 4.4 Reading Time Results for words in the critical region, “No Antecedent” inquiry (“*”: p < 0.05)
Word
Main Effects Interaction
Verb Type Antecedent Verb Type*Antecedent
b SE t p b SE t p b SE t p
Verb
(leu/
folheou)
4.744 5.019 0.945 0.345 -1.715 5.025 -0.341 0.733 -2.379 5.022 -0.474 0.636
“no” 3.088 4.361 0.708 0.479 -8.591 4.364 -1.969 0.049* -2.333 4.363 -0.535 0.593
“sofá” 8.147 4.821 1.690 0.091 9.880 4.826 -2.047 0.041* -4.556 4.823 -0.945 0.345
“da” 7.977 3.928 -3.693 0.042* -14.531 3.935 -2.047 >0.001* -6.429 3.932 -1.635 0.102
146
Table 4.5 Planned Comparisons for words in the critical region, “No Antecedent” inquiry (“*”: p < 0.05)
Word Contrasts Results
b SE t p
“no” OPT+Antec vs. OPT-Antec -6.425 5.930 -1.083 0.279
OT+Antec vs. OT-Antec -10.998 6.426 -1.712 0.088
“sofá” OPT+Antec vs. OPT-Antec -5.587 6.314 -0.885 0.377
OT+Antec vs. OT-Antec -14.528 7.264 -2.000 0.046*
“da” OPT+Antec vs. OPT-Antec -8.485 5.070 -1.674 0.095
OT+Antec vs. OT-Antec -20.724 6.067 -3.416 >0.001*

At the critical verb, there were no RT differences between conditions with OT and OPT
verbs with or without an antecedent (Figure 4.1, Table 4.4). At this point, people might still be
expecting an overt object to follow
31
and therefore have not yet incurred an implicit object
interpretation.
At the first word of the spillover region (e.g., no, “on the”), RTs are slower in no-
antecedent conditions than in antecedent-containing conditions (main effect of antecedent, see
Tables 4.3, 4.4). We find no main effect of verb type and no interaction. However, planned
pairwise comparisons show that conditions with OT verbs with no antecedent were read marginally
slower than those with an antecedent, whereas conditions with OPT verbs showed no effect of
presence/absence of antecedent (see Tables 4.4, 4.5 for statistical details).
At the second word of the spillover region (e.g., sofá, “couch”), RTs were again longer
in no-antecedent conditions (main effect of antecedent). There is no significant effect of verb type
and no interaction. However, pairwise analyses show that absence of an antecedent lead to a
slowdown in OT conditions but not in OPT conditions (Tables 4.4, 4.5).
At the third and final word of the spillover region (e.g., da, “of the”), RTs were again
longer in no-antecedent conditions (main effect of antecedent) and for OT verbs (main effect of

31
Chapter 4 offers further insight into this expectation: even when an antecedent is available, if a noun phrase follows
the verb, the noun phrase in canonical overt object position is more likely to be interpreted as the object (relative to
the anaphoric implicit object interpretation with the antecedent). That is, an overt object interpretation seems to be
preferred over an anaphoric implicit object interpretation.
147
verb type). Despite no interaction, pairwise analyses show once again that the lack of an antecedent
cause slowdowns in OT conditions but not in OPT ones (Tables 4.4, 4.5)
All words outside the critical region were analyzed as well, for the sake of completeness.
These analyses revealed a spurious main effect of verb type at the first word and a main effect of
antecedent at the comma (sixth word). Details are given in Appendix 5. However, as these results
have no bearing on the hypotheses being tested in this chapter, we do not discuss them further.

Discussion (Reading Times for “No Antecedent” inquiry)

The reading time patterns show that in conditions with obligatorily transitive (OT) verbs
(which can only have anaphoric implicit objects), absence of an antecedent triggers RT
slowdowns. In contrast, optionally transitive (OPT) verbs (which can have both anaphoric and
existential implicit objects) show no such slowdowns. These patterns support the hypothesis that
the comprehension of context-dependent anaphoric implicit objects with no supporting context for
the anaphoric interpretation leads to comprehension difficulties. This parallels previous findings
regarding the comprehension of context-dependent overt pronouns “she/he”: the absence of a
referent leads to difficulties (e.g., Van Berkum et al. 2004, Nieuwland and Van Berkum 2006). In
contrast, when the implicit object can be interpreted existentially, no difficulties ensue – this
parallels the comprehension of pronouns like “they”, which do not always require an antecedent
(e.g., Sanford 2008, Filik et al. 2008).

148
Results: Reading Times for “Existential implicit object antecedent” inquiry

Reading times for the relevant conditions and critical region regarding the exploratory
question on the effects of having an Existential implicit object as the antecedent for an Anaphoric
implicit object were analyzed in a parallel way to those pertaining to the main research question.
The critical region for the exploratory research question begins at the final OT verb (e.g.,
guardou, “put away”) and includes the three following words as the spillover region (e.g., na
estante e, “on shelf and”; see Table 4.2). To address the exploratory question on whether the
comprehension of anaphoric implicit objects that refer back to existential implicit objects leads to
comprehension difficulties, reading times in this critical region for the relevant conditions (e.g.,
ex.4.12, ex.4.13, ex.4.16) were analyzed.
As can be seen in Figure 4.1 and Tables 4.6 and 4.7, RTs in the critical region reveal no
effect of antecedent on OPT verbs. Thus, there is no evidence supporting the idea that
comprehension of an anaphoric implicit object (e.g. the implicit object of ‘put away’) is harder
when the antecedent is an existential implicit object (ex.4.12; e.g. the existential IO of ‘read’, i.e.
the unmentioned something being read) relative to when the antecedent is an overt object (e.g., “a
book”), regardless of whether the overt object was only mentioned once (ex.4.16) or mentioned
once and picked up again by an anaphoric IO (ex.4.13). There are also no effects of antecedent
when comparing the no-antecedent OPT condition with the control condition (Tables 4.6, 4.7).

Table 4.6 Mean RTs in milliseconds and SE for words in the exploratory critical region (Existential implicit
object antecedent inquiry) by verb type and antecedent type
Word
OPT + no antec OPT + with antec Control

Mean RT SE Mean RT SE Mean RT SE
“guardou” 371.984 11.723 369.267 11.63 375.779 12.9
“na” 377.168 11.028 366.376 9.859 368.533 10.487
149
“estante” 376.934 10.154 376.487 10.382 380.103 10.172
“e” 387.98 9.277 377.137 8.773 387.248 9.021

Table 4.7 Reading Time Results for words in the exploratory critical region, “Existential implicit object
antecedent inquiry” inquiry (“*”: p < 0.05)
Word
Main effect of Antecedent
OPT + antec vs. OPT - antec OPT - antec vs. Control
b SE t p b SE t p
“guardou” -1.364 6.059 -0.225 0.822 1.776 6.418 0.277 0.782
“na” -5.421 5.876 -0.923 0.357 -4.541 6.077 -0.747 0.455
“estante” 0.072 5.439 0.013 0.989 1.576 5.278 0.299 0.765
“e” -5.59 4.699 -1.19 0.235 -0.481 4.649 -0.103 0.918

Discussion (Reading Times for “Existential implicit object antecedent” inquiry)

This line of inquiry investigated whether having an existential implicit object as the
antecedent of an anaphoric implicit object would cause difficulties during comprehension. Results
revealed no evidence supporting this idea, and parallel findings from prior work showing that
pronominal reference to nuclear implicit antecedents does not cause comprehension difficulties
(e.g., Cornish et al. 2005, Järvikivi et al. 2019): reference to existential implicit objects (which are
nuclear to the event) made with anaphoric implicit objects also does not seem to generate an
additional processing burden, despite existential implicit objects’ lower accessibility in discourse.

Results: Comprehension Questions

After each sentence, participants answered a comprehension question. These questions
were meant to assess people’s interpretations of the implicit object (following Christianson et al.
2001). The goal of this assessment was to confirm that conditions with available antecedents did
indeed lead to anaphoric interpretations of the implicit object, which was particularly relevant in
sentences with OPT verbs because they can potentially yield both anaphoric and existential
150
interpretations.
In conditions with OPT and OT verbs, the question inquired about the identity of the
implicit object of the OPT/OT verb: e.g., “Did Clara read/flip-through a book?”. Answers provide
evidence regarding whether people interpreted the implicit object anaphorically (‘yes’ response)
or not (‘no response’). In control conditions (which did not contain the initial OPT/OT verbs), the
question inquired about the identity of the final OT verb’s implicit object: e.g., “Did Clara put
away a book?”. Similarly, answers provide evidence regarding whether people interpreted this
final implicit object anaphorically or not.
As expected, when an antecedent (e.g., “a book”) was mentioned and people were asked if
the antecedent matched their interpretation of the implicit object’s identity (e.g., “Did Clara read
a book?”), most responses were ‘yes’ (see Table 4.8, Figure 4.2). In contrast, when there was no
antecedent present, people responded with ‘yes’ significantly less often (only about 24% of the
time). There were no significant differences between verb types or the control condition. These
patterns confirm that the availability of an antecedent did lead to anaphoric interpretations
of implicit objects. This anaphoric interpretation obtains even with OPT verbs, which allow both
anaphoric and existential implicit objects.
The statistical analysis with OPT and OT verbs shows no main effect of verb type, a
significant main effect of antecedent, and no interaction. A comparison between the control
condition and the OPT condition with an antecedent shows no effect of verb type, just like a
comparison between the control condition and the OT condition with an antecedent (see Table
4.9).

151
Table 4.8 Proportion of ‘yes’ responses +-1 SE by verb type and antecedent
OPT verb OT verb Control
Prop. SE Prop. SE Prop. Se
With antecedent 0.904 0.021 0.941 0.017 0.856 0.025
No antecedent 0.229 0.03 0.245 0.031 -

Figure 4.2 Proportion of ‘yes’ responses to comprehension question +-1 SE

Table 4.9 Comprehension Question Results (“*”: p < 0.05)
Contrast
Main Effects Interaction
Verb Type Antecedent Verb Type*Antecedent
b SE t p b SE t p b SE t p
OPT vs. OT

0.171 0.159 1.385 0.166 2.248 3.916 14.064
>
0.001*
0.116 0.123 0.937 0.348
Control vs.
OPT + Antec
-0.277 0.179 -1.55 0.121 -
Control vs.
OT + Antec
-0.139 0.188 -0.741 0.459 -

152
Discussion (Comprehension Questions)

Participants’ responses to target comprehension questions confirmed that in BP, whenever
a suitable antecedent is available, an anaphoric implicit object interpretation occurs: that is,
responses show that people interpret the implicit object as coreferring with the previously
introduced antecedent.
In particular, the anaphoric interpretation occurred with both obligatorily transitive and
optionally transitive verbs (including the final obligatorily transitive verbs in control conditions).
These patterns assuage concerns that implicit objects of optionally transitive verbs could have been
interpreted existentially despite the presence of the antecedent.
4.4. General Discussion

The current chapter expands on a body of work that has mainly focused on how the absence
of explicit antecedents affects the comprehension of overt context-dependent expressions such as
pronouns. Here, I take a closer look at a less understood phenomena, i.e. context-dependency
effects on the comprehension of implicit expressions.

Main question: Implicit objects with no antecedents

The main question of this work focused on the effects of context-dependency in the
comprehension of implicit objects in Brazilian Portuguese. Anaphoric implicit objects (IOs) are
context-dependent expressions, thus it was expected that sentences with anaphoric IOs and no
available antecedents would lead to comprehension difficulties. This expectation is based on
153
previous findings on the comprehension of overt context-dependent referring expressions, which
have shown that processing the pronouns “she/he” in contexts with no available antecedents
generates difficulties (e.g., Van Berkum et al. 2004, Nieuwland and Van Berkum 2006, Sanford et
al. 2008, Filik et al. 2008).
In contrast, for sentences that allowed an existential implicit object (IO) interpretation -
which does not require a referent/antecedent - it was expected that no comprehension difficulties
would ensue when no antecedent was available. This expectation is parallel to what happens with
the overt pronoun “they”, which can occur in contexts without specific antecedents without
generating a processing burden (e.g., Sanford et al. 2008, Filik et al. 2008).
As predicted, reading times revealed that sentences with context-dependent anaphoric IOs
and no available referent in context lead to comprehension difficulties (as seen in slower reading
times) relative to when a referent is available. These difficulties arise because there is no
antecedent to supply meaning for the interpretation of the context-dependent IO.
The slowdown in conditions with anaphoric IOs and no antecedent appears in the critical
region (following the verb) and is detectable even after the critical region, suggesting that
difficulties caused by the absence of a referent persist for a while. This persistence in slower
reading times may occur because people take a while to recover from this difficulty, or because
people do not immediately ‘give up’ on finding a referent for the anaphoric IO and keep waiting
for one to appear in discourse.
In contrast, sentences in which a context-independent interpretation of the implicit object
is possible did not lead to comprehension difficulties: even when no referent is available in context,
there was no slowdown in reading times, because the implicit object could be interpreted
existentially and independently of context.
154
The complete absence of a slowdown for sentences with OPT verbs – which can yield both
anaphoric and existential implicit objects – indicates that the existential interpretation is readily
available to people upon encountering the verb.
These findings suggest that the language processor is able to efficiently identify which
types of implicit objects different types of verbs (OPT and OT) allow.

Exploratory question: Anaphoric implicit objects with implicit antecedents

The exploratory question investigated whether the comprehension of anaphoric IOs with
existential, implicit antecedents would also yield comprehension difficulties relative to those with
overt antecedents. The results revealed that no differences in reading times occurred between
conditions with implicit (IO) and overt antecedents.
These findings are compatible with prior work by Cornish et al. (2005) and Järvikivi et al.
(2019), which found no difficulties associated with the comprehension of pronominal reference to
nuclear implicit antecedents. That is, despite existential implicit objects’ relatively lower
accessibility in discourse, referring back to them with a reduced expression such as an anaphoric
implicit object does not seem to generate an additional processing burden.
In addition, these findings highlight another way in which the comprehension of anaphoric
implicit objects parallels that of context-dependent overt pronouns like “she/he”: both types of
expressions are able to refer back to nuclear implicit antecedents without causing added difficulties
relative to when they refer back to overt antecedents.
Although these findings are in tune with prior research on reference to implicit antecedents,
it is important to acknowledge that the exploratory critical region occurred towards the end of
155
experimental target sentences. Thus, there is a possibility that effects may have been washed away
due to readers ‘rushing’ towards the end.

Comprehension Questions

After reading each sentence, people answered a comprehension question such as “Did
Clara read/flip-through/put away a book?”. The question was meant to assess how people
interpreted the IO and confirm that the availability of an antecedent would lead to anaphoric
interpretations. Thus, when an antecedent was available, it was expected that the majority answers
would be “yes”; if not, it was expected that most answers would be “no”. The answer patterns
occurred as expected, with majority of answers “yes” for sentences with a referent present and
“no” when there was no referent.
It is interesting to note that sentences with no referents available still yielded about 25%
“yes” responses. That is, after reading sentences such as “Clara entered, read on the couch...” - in
which no “book” was mentioned - people still answered the question “Did Clara read a book?”
with “yes” approximately 25% of the time. Upon closer inspection of responses by items and by
participants, some interesting patterns emerged. Firstly, three items (out of 20) displayed unusually
high proportion of “yes” answers for sentences with no referent (above 50% “yes” responses).
These were items in which the sentences provided enough cues that people could reasonably expect
the answer to be “yes” even though there was no explicit mention of the referent. For example,
one of these cases were sentences with the verbs “drive/wash” that ended with “…left in the
garage”. People were then asked “Did he drive/wash a car?”. In these cases, the identity of the
existential implicit object was not explicit but it was considerably narrowed down by the context
156
provided
32
, i.e., it seems reasonable that in these cases people inferred that what was being
driven/washed then put in a garage was indeed a car. In regards to the participants, there were at
least five (out of 45) who answered “yes” unusually often (above 50% of answers) across
conditions. I speculate that some people are more willing to be accommodating and accept the
information presented by the question itself. That is, for these “accommodators”, positive answers
are justified if the referent presented in the question can be reasonably assumed to match the
underspecified existential implicit object. The sentence might not have explicitly mentioned a
book, but because the question explicitly asks about a book, “accommodators” might be more
likely than most to accept/infer that what was read was in fact a book. Nonetheless, the majority
of people did not present this behavior.
Lastly, as expected, there were no differences in the proportion of yes/no answers based on
whether the verb was optionally or obligatorily transitive. This lack of differences between verb
types supports the idea that even with optionally transitive verbs, which do allow existential
implicit objects, an anaphoric interpretation obtains whenever there is a referent available in
context.

In sum, although existential and anaphoric implicit objects are both equally
invisible/unpronounced, the experiment reported in this chapter reveals that the language parser is
able to use information about the verb rapidly during processing (e.g., whether it is optionally or
obligatorily transitive) to assess which kinds of implicit objects – existential or anaphoric – are
allowed in a certain structure. The findings reported here contribute to a robust body of
psycholinguistic work looking at how verb information (e.g., its argument structure) is accessed

32
As we know from Chapter 2, context can influence the interpretation of existential IOs.
157
by the language processor during comprehension (e.g., Boland et al. 1990, Trueswell et al. 1993,
Boland 1993). More specifically, I show that the parser is able to quickly access verb information
regarding its implicit arguments and their context dependency properties. In doing so, the current
study illustrates a rather uncommon case in which it is the verb (and not a referring expression
such as a pronoun or a noun phrase) that guides the parser in the process of reference tracking, by
signaling whether or not its implicit object requires a referent for interpretation.

158
Chapter 5 : Implicit objects in comprehension: Garden-path sentences

5.1. Introduction

Chapter 4 explored effects of context-dependency on the comprehension of existential and
anaphoric implicit objects in Brazilian Portuguese. Chapter 5 continues to look into the
comprehension of implicit objects by investigating how the availability of existential and
anaphoric implicit object interpretations affects the comprehension of temporarily ambiguous
sentences (i.e., garden-paths).
In particular, the current work utilizes garden-path sentences in which a noun phrase (e.g.,
“a book”, in bold) is temporarily ambiguous between object and subject position (ex.5.1a):

(5.1) a. After I read a book fell on my foot.

b. After I read, a book fell on my foot.

Prior work shows that when reading sentences like (ex.5.1a) people tend to initially parse
the noun phrase “a book” as the object of “read” (e.g., Ferreira et al. 1991, Christianson et al. 2001,
Patson et al. 2009, Slattery et al. 2013). This tendency occurs despite the fact that the verb “read”
is optionally transitive: it can appear without an overt object, i.e., with an implicit object. This is
more clearly exemplified in (ex.5.1b). Crucially, prior work revealed that there is a bias in
comprehension towards (i) preferring an overt object when one is available over (ii) postulating
an implicit object for the verb “read” and assuming that the noun phrase “a book” is the subject of
159
a new clause (an instance of Late Closure bias; see Frazier 1978, Frazier & Rayner 1982).
The current work exploits properties of Brazilian Portuguese (BP) to investigate how the
interpretation of implicit objects (IOs) affects the comprehension of garden-path sentences. In BP,
there are two types of invisible/unpronounced direct objects: existential and anaphoric IOs.
Although they share an identical null form, their discourse functions are very different: anaphoric
IOs are context-dependent, i.e., they require a referent in discourse to supply their meaning for
successful interpretation. Existential IOs, on the other hand, are context-independent: they do not
require any context and are interpreted in a non-specific way (e.g., as ‘something’).
Taking a closer look at Brazilian Portuguese allows us to probe how the availability of the
anaphoric implicit object interpretation – which is not available in English – influences the
comprehension of garden-path sentences. The initial erroneous parsing of the ambiguous noun
phrase (in (ex.5.1a), “a book”) as an object and the subsequent need to re-analyze it as the subject
of the following verb generates comprehension difficulties, and the initial misparse interpretation
often lingers even after people finish reading the sentence (e.g., Ferreira et al. 1991, Christianson
et al. 2001, Patson et al. 2009, Slattery et al. 2013). However, prior work on garden-paths has
focused exclusively on garden-paths for which re-analysis involves existential implicit objects, not
anaphoric ones. The current work aims to fill this gap by probing into the comprehension of
Brazilian Portuguese.
The current chapter investigates how the interpretation of implicit objects affects the
comprehension of garden-path sentences. In particular, I focus on two main aims: (i) the first is to
investigate whether a mismatch between the object referent in the initial garden-path misparse
(e.g., a book in ex.5.1a) and the object referent in the successful interpretation (e.g., an implicit
‘something’ in ex.5.1a) affects the comprehension of these structures. (ii) The second is to
160
investigate whether a potential discourse link between the implicit object and a subsequent noun
phrase affects the comprehension of these structures.
More specifically, the first aim explores the idea that it might be easier for people to reject
the initial misparse during garden-path re-analysis if this initial misparse and the new parse are
incompatible with one another (i.e. they cannot both be true about the same event) relative to when
the initial misparse and the new parse are compatible and can logically coexist. In the current
study, this idea is tested by looking at garden-paths in which the object referent of the initial
misparse does not match the object referent of the new parse post re-analysis (e.g., in one case the
person is reading a book and in the other case the person is reading a letter). The idea that a
mismatch regarding object referents across parses may affect the comprehension of garden-paths
will henceforth be called the Alternative Referent Hypothesis:
This hypothesis is investigated here by comparing garden-paths where re-analysis results
in existential implicit objects vs. ones where re-analysis results in anaphoric implicit objects. With
an existential interpretation (ex.5.2a), the initial misparse has the object referent “a book”, and the
new successful parse has no specified referent (e.g., he read something): these are not incompatible,
as the “something” I read could be a book. In contrast, with an anaphoric interpretation (ex.5.2b),
the misparse has the object referent “a book”, but the successful parse has the object referent “a
letter”: these parses are incompatible because they cannot both be true about the same event, as “a
book” is most likely not the same object as “a letter”.

(5.2) a. Depois que eu li ∅ um livro caiu no meu pé. [existential IO]
After I read ∅ a book fell on my foot.
English Paraphrase: After I read something a book fell on my foot.
161

b. Depois que eu recebi uma cartak e li ∅k um livro caiu no meu pé. [anaphoric IO]
After I received a letterk and read ∅k a book fell on my foot.
English Paraphrase: After I received a letter and read it/the letter a book fell on
my foot.

The second aim of this chapter is to explore whether the availability of a discourse link
between the implicit object and a following expression can facilitate parsing. For instance, when
a noun phrase is definite, it indicates givenness and thus can be inferred to corefer with a referent
in prior discourse. Thus, in (ex5.3b), one may infer that “the book” that fell on my foot is the same
that I had read (this inference of coreference is what I call a discourse link for the purposes of this
Chapter). In contrast, an indefinite noun phrase (ex.5.3a) typically encodes discourse newness and
does not promote the same inference, i.e., the same discourse link (e.g., Heim 1982, Prince 1992).
In (ex.5.5a), it is much less likely that one could infer that the fallen book is the same as the one I
read:

(5.3) a. Depois que eu li ∅ um livro caiu no meu pé.
After I read ∅ a book fell on my foot.

b. Depois que eu li ∅ o livro caiu no meu pé.
After I read ∅ the book fell on my foot.

In garden-paths such as (ex.5.1a or ex.5.2b), there is no discourse link between the implicit
162
object (e.g., the IO of “read”) and any subsequent expression. However, if a discourse link were
to arise in the successful parse but not in the initial misparse, this discourse link could potentially
function as an additional cue pushing the language processor towards the successful parse –
because it’s the only parse in which the discourse link obtains. This idea will be henceforth referred
to as the Discourse Link Cue hypothesis:
In the current chapter, the Discourse Link Cue hypothesis is tested by manipulating the
definiteness of the temporarily ambiguous noun phrase. The definite noun phrase, due to its
association with discourse givenness, may trigger the search for a possible link in prior discourse.
In these garden-paths, the only prior referent available for this discourse link is the implicit object
that arises only in the ultimately successful parsing of the garden-path. Thus, definite noun phrases
may provide a cue towards this successful parse. In contrast, when the noun phrase is indefinite, it
is unlikely to be interpreted as given in discourse, or as correfering with a prior referent. Thus,
indefinites do not promote a search for a prior referent nor a discourse link with prior discourse,
and consequently provide no cue towards successful re-analysis.
In order to investigate aims 1 and 2, a self-paced reading task in Brazilian Portuguese was
designed to investigate the role implicit objects play in the comprehension of sentences with a
temporary object/subject ambiguity (i.e., garden-paths).
By taking a closer look at how the interpretation of implicit objects affects the processing
of garden-path sentences, the current work aims to shed light on how these invisible, unmentioned
arguments influence the mental operations involved in language comprehension.
This chapter is structured as follows: Section 5.2 introduces the background on the
comprehension of garden-path sentences with temporary object/subject ambiguities, and discusses
how implicit object interpretation may play a role in the comprehension of these constructions.
163
Section 5.3 describes a self-paced reading experiment designed to investigate how (i) an anaphoric
interpretation of the implicit object and (ii) a discourse link between the implicit object and a
subsequent expression affect the comprehension of these temporarily ambiguous constructions.
Section 5.4 concludes with a general discussion on the findings.
5.2. The comprehension of Garden-Path sentences with a temporary object/subject
ambiguity

Garden-path sentences are temporarily ambiguous structures. They are known as “garden-
paths” because during comprehension people are initially ‘led down the wrong path’: the language
parser initially selects what ultimately turns out to be an incorrect parse/interpretation of the
sentence. When enough information is encountered to determine that the initial parsing is incorrect,
this initial misparse can be revised and the correct parse selected.
There are various kinds of garden-path sentences (e.g., see Frazier 1978, Frazier & Rayner
1982, Ferreira et al. 1991, Britt et al. 1992, Christianson et al. 2001, Patson et al. 2009, Slattery et
al. 2013). The current work focuses on garden-paths with a temporary object/subject ambiguity
such as (ex.5.1a), repeated below as (ex.5.4a):

(5.4) a. After I read a book fell on my foot.

b. After I read, a book fell on my foot.

164
When sentences such as (ex.5.4a) are presented in writing and without a comma between
the first verb and the subsequent noun phrase, readers typically face comprehension difficulties
33

(e.g., Frazier and Rainer 1982, Ferreira et al. 1991, Christianson et al. 2001). These difficulties
occur because people tend to interpret the noun phrase “a book” as the direct object of the
preceding verb, “read”. It is only when people reach the following verb, “fell”, that it becomes
clear that this initial parse must be revisited: the noun phrase must be interpreted as the subject of
“fell”, otherwise the second clause would be left without a subject and consequently the structure
would be ungrammatical in English. As a result, “read”, which is an optionally transitive verb,
must be re-analyzed as not having an overt direct object, but rather an existential implicit object
(as in (ex.5.4b)).
The initial misparse with garden-path sentences occurs because – other things being equal
– the parser has a tendency to use available input to continue a clause that it is currently processing
rather than using that input to begin a new clause. This tendency is known as Late Closure (see
Frazier and Rayner 1982). Thus, when the parser encounters the noun phrase following a verb that
allows a direct object, late closure bias guides the parser towards interpreting the noun phrase as
the object rather than initiating a new sentence.
The need to reanalyze the initial misparse in garden-paths is known to cause difficulties
during comprehension starting at the point of the disambiguating verb (e.g., Ferreira et al. 1991,
Christianson et al. 2001). This disruption in comprehension also causes lingering
misinterpretations: even after the sentence has been disambiguated, people often respond to
questions about the sentence with answers that are compatible with the initial misparse. For

33
These difficulties (i.e., known as the garden-path effect) also occur in spoken language when prosody does not match
that of the correct parsing. When the appropriate prosodic cues are present, prosody can facilitate garden-path
processing (e.g., Nagel et al. 1996, Kjelgaard et al. 1999).
165
example, when people are asked a question such as “Did I read a book?” after reading a garden-
path such as (ex.5.4a), they often respond “yes” – despite the fact that after successful re-analysis,
“the book” is no longer the object of “read” (e.g., Christianson et al. 2001, Patson et al. 2009,
Slattery et al. 2013).
Why do people often exhibit lingering misinterpretations after reading garden-path
sentences? Slattery et al. (2013) looked into this matter and found that lingering misinterpretations
occur because the initial erroneous interpretation of garden-path sentences remains active in
memory. They conducted two experiments which revealed that although people do ultimately
achieve successful reanalysis of these structures, the initial misparse is not always fully erased
from memory – i.e., it lingers.
5.2.1. Garden-Path reanalysis: implicit object interpretation

The current work investigates whether the interpretation of the implicit object in the
ultimately correct parse of garden-path sentences (e.g., the implicit object of “read” in (ex.5.6a))
affects the comprehension of these sentences. More specifically, this work’s first aim is to
investigate how the possibility of an anaphoric interpretation of the implicit object as coreferential
with an antecedent affects the comprehension of garden-path sentences. The second aim is to
examine whether the possibility of a discourse link between the implicit object and a subsequent
definite noun phrase influences the comprehension of these structures.

166
Implicit object interpretation in Garden-Paths

Prior studies have examined the comprehension of garden-path sentences with temporary
object/subject ambiguities and found that they lead to processing difficulty and lingering
misinterpretations (e.g., Ferreira et al. 1991, Christianson et al. 2001, Patson et al. 2009, Slattery
et al. 2013). In this prior body of work, successful reanalysis of garden-paths leads to a parse in
which the initial clause has an existential implicit direct object (e.g., the IO of “read” in (ex.5.6a));
that is, an implicit object that entails the existence but does not specify the identity of a referent.
As such, existential IOs are better paraphrased with “something”, e.g., “After I read something a
book fell on my foot” (e.g., Fillmore 1986, Anderbois 2012, Williams 2015).
To the extent of my knowledge, Christianson et al. (2001) presents the work who has most
directly examined how the interpretation of the implicit object affects garden-path comprehension.
Christianson et al. (2001) looked at garden-paths with two different kinds of implicit objects: (i)
existential implicit objects, which arise with optionally transitive verbs like “read” (ex.5.4a) and
“hunt” (ex.5.5a) and (ii) reflexive implicit objects, which arise with reflexive absolute verbs like
“bathe” (ex.5.5b). The crucial difference between these garden-paths lies on how the initial verb’s
implicit object is interpreted after reanalysis: in (ex.5.5a), the implicit object of “hunt” is an
existential IO, and can be paraphrased with “something”. In (ex.5.5b), the implicit object of
“bathe” is reflexive (refers to subject), and could be paraphrased with an overt reflexive like
“herself”. In the first case the implicit object has no specified referent, whereas in the latter case
the implicit object has a specific referent (the subject of the clause, “the mother”).

(5.5) a. While the man hunted the deer ran into the woods.
167

b. While the mother bathed the child laughed.

Christianson et al. (2001) tested both optionally transitive and reflexive absolute verbs to
assess whether lingering misinterpretations were truly due to the initial misparse or whether they
could be attributed to general reasoning about the identity of the implicit object. With optionally
transitive verbs (e.g., “hunt”), the existential implicit object post-reanalysis is left unspecified
(ex.5.6a). When asked “Did the man hunt the deer?” participants may say with “yes” not because
they retained the original misparse, but because they use general reasoning to infer that the
‘something’ being hunted is, in fact, the deer that ran into the woods.
34
In contrast, garden-paths
with reflexive verbs do not allow for the same inference: after reanalysis, the implicit object of
“bathe” must be interpreted as “herself” (ex.5.6b). This final reflexive interpretation is
incompatible with inferences that what the mother bathed was the subsequently mentioned child.

(5.6) a. While the man hunted ∅ the deer ran into the woods. [∅ = “something”]

b. While the motherk bathed ∅k the child laughed. [∅ = “herself”]

Christianson et al. (2001) indeed found evidence that both garden-paths with optionally
transitive and reflexive absolute verbs (ex.5.6a-b) lead to lingering misinterpretations. Even after
reading garden-paths with reflexive verbs, people displayed interpretations compatible with the

34
Note that this inference is supported by the definiteness of the noun phrase “the deer”. Definiteness typically
correlates with givenness in discourse, and definite noun phrases can be inferred to link back to an entity previously
evoked in discourse. Further discussion on this topic can be found below on the subsection titled Second aim:
Investigating the Discourse Link Cue hypothesis.
168
initial misparse (e.g., the mother bathed the child). This suggests the ‘misinterpretations' are truly
remnants of the initial misparse and cannot simply be attributed to inferences about the identity of
the implicit object.
Interestingly, Christianson et al. (2001)’s study also found that garden-paths with reflexive
verbs lead to significantly less lingering misinterpretations than garden-paths with optionally
transitive verbs: 57.3% “yes” responses for reflexives and 71.9% “yes” responses for optionally
transitive cases (where “yes” responses indicates an interpretation compatible with the initial
misparse)
35
. It is unclear whether the discrepancy between verb types is due to (i) inference cases
leading to the additional “yes” responses when the verb is optionally transitive or (ii) reflexives
yielding overall fewer lingering misinterpretations.
I speculate that reflexives could potentially lead to less lingering misinterpretations because
reanalysis with a new referent (e.g., “herself” instead of “the baby”) may more efficiently help
erase the initial misparse, relative to a reanalysis in which no referent is specified (e.g.,
“something” instead of “the deer”). I return to this below in my discussion of the Alternative
Referent hypothesis.
In addition, Christianson et al. (2001) did not directly compare the online comprehension
of garden-paths with optionally transitive and reflexive verbs - so it is also not clear whether having
a referent for the implicit object, in the case of reflexives, affects the real time reanalysis of these
structures in any way.
Brazilian Portuguese (BP) provides a fruitful testing ground to build on Christianson et al.
(2001)’s findings and further investigate how the interpretation of the implicit object in garden-
paths affects comprehension. In English, optionally transitive verbs like “read” can only give rise

35
Christianson et al. (2001)’s experiment 3b.
169
to existential IOs. In BP, optionally transitive verbs such as “read” can give rise to existential and
anaphoric IOs. This provides an opportunity to investigate the interpretation of the implicit object
while controlling for verb type (i.e., there is no need to compare two different verb types to one
another, such as optionally transitive and reflexive verbs).
As we already saw in Chapter 4, anaphoric IOs in BP occur when a salient antecedent is
available: in (ex.5.7a), the implicit object of li (‘read’) is interpreted anaphorically as the
previously mentioned referent um livro (‘a book’). This sentence can be best paraphrased as “I
read it/the book I grabbed in the kitchen”. When no referent is available, an existential
interpretation of the implicit object takes hold: (ex.5.7b) is best paraphrased as “I read something
in the kitchen”.

(5.7) a. Eu peguei um livrok e li ∅k na cozinha.
I grabbed a bookk and read ∅k in the kitchen.

b. Eu li ∅ na cozinha.
I read ∅ in the kitchen.

The current work’s first aim is to examine how having another referent for the implicit
object interpretation – such as with reflexives (e.g., Christianson et al. 2001) – affects the
comprehension of these structures. To do so, I turn to anaphoric implicit objects in Brazilian
Portuguese. The second aim is to take a closer look at whether a discourse link associated with the
implicit object can function as a cue towards the correct parse. In the course of investigating these
aims, the current work also provides further confirmation that Late Closure bias (e.g., Frazier and
170
Rayner 1982) occurs even when the direct object can be interpreted as an anaphoric implicit object.
This confirmation of late closure bias, as well as each of the current work’s two main aims,
are further explored in more detail below.

Late Closure bias confirmation

In addition to investigating its aims 1 and 2, the current work also assesses whether the late
closure bias occurs even in contexts that allow for an anaphoric interpretation of the implicit object.
Doing so is valuable because it provides additional validation of general parsing principles (in this
case, Late Closure) in a previously untested construction – garden-paths with anaphoric IOs.
For instance, the sentence in (ex.5.2a, repeated below as ex.8a) offers two possibilities in
regards to how the direct object of the verb li (‘read’) is interpreted: (i) The noun phrase following
the verb (um livro, ‘a book’) can be interpreted as the object, leading to a parse that will need to
be revised at the disambiguating verb caiu (‘fell’). Alternatively, (ii) the verb can have an
existential implicit object that has no referent; this is the eventual parsing in English post-
reanalysis.
Prior garden-path studies have shown that, due to the language parser’s Late Closure bias,
option (i) is selected over option (ii): the parser prefers to use new input to continue processing the
current structure when that is possible, relative to ‘closing’ the current structure and starting a new
one (e.g., Frazier and Rayner 1982, Ferreira et al. 1991). Therefore, in (ex.5.8a) the noun phrase
“a book” tends to be interpreted as the object of the clause currently being parsed rather than as
the subject of a following clause. This tendency leads to the initial misparse and subsequent
revision that make up the classic garden-path effect.
171

(5.8) a. Depois que eu li ∅ um livro caiu no meu pé. [existential IO]
After I read ∅ a book fell on my foot.
English Paraphrase: After I read something a book fell on my foot.

b. Depois que eu recebi uma cartak e li ∅k um livro caiu no meu pé. [anaphoric IO]
After I received a letterk and read ∅k a book fell on my foot.
English Paraphrase: After I received a letter and read it/the letter a book fell on
my foot.

In contrast, the sentence in (ex.5.2b, repeated here as ex.5.8b) contains an available
antecedent for the implicit object (uma carta, ‘a letter’): thus, this construction offers two
possibilities to the parser: either (i), the same as for (ex.5.10a), or another option: (iii) The verb
can have an anaphoric implicit object that is corefential with the antecedent uma carta (‘a letter’).
I expect that, just as the parser prefers option (i) over (ii) due to late closure, this very same
principle will guide the parser to prefer option (i) over (iii) in sentences such as (ex.8b). This
expectation is supported by previous findings regarding garden-paths with reflexive verbs:
although the reflexive interpretation of the implicit object was available, people still displayed a
bias towards interpreting the subsequent noun phrase as an overt direct object, leading to lingering
misinterpretations (e.g., Christianson et al. 2001). Thus, it seems that as long as upcoming input
(e.g., “a book”) can be incorporated into the structure being currently parsed, late closure predicts
that this input will indeed be incorporated.

172
First aim: Investigating the Alternative Referent hypothesis

The first aim of this study is to investigate whether the possibility of interpreting the
implicit object anaphorically when an antecedent is available affects the comprehension of garden-
paths with object/subject ambiguities.
In garden-paths with object/subject ambiguities, the initial misparse has an overt direct
object. This initial misparse must be revised to a parse with an implicit object. In (ex.5.8a), this IO
is existential and has no specific referent. In contrast, when the IO is interpreted anaphorically
(ex.5.8b), the final parse contains an specific object referent that is different than the object referent
in the initial misparse. For instance, in (ex.5.8b) the object referent is a book in the misparse and
a letter in the final parse.
The mismatch between referents in the initial and final parses may have some influence on
the comprehension of these structures. More specifically, this mismatch could potentially make it
easier for the parser to reject the initial misparse during garden-path re-analysis. Evidence
supporting this idea comes from Christianson et al. (2001), who found that garden-paths with
reflexive verbs like “bathe” – which result in a mismatch between object referents in the initial
and final parses – led to fewer lingering misinterpretations.
The current work investigates whether garden-paths that provide a specific referent for the
implicit object interpretation, like reflexives, may facilitate garden-path re-analysis and lead to
fewer lingering misinterpretations. This may occur because with an alternative referent, the
ultimately successful interpretation is less compatible with the initial misparse: the two parses
cannot both be true of the same event. In contrast, the successful parse without a specific object
referent is relatively more compatible with the initial misparse, because both parses can accurately
173
describe the same event.
Garden-path re-analysis encompasses a process in which the initial misparse is revised in
favor of the successful parse. Prior work has shown that even when re-analysis is successful (i.e.,
when people reach the successful parse), the initial misparse still often lingers and remains active
in memory: these lingering misinterpretations are attributed to the parser nor fully erasing the
initial parse (e.g., Slattery et al. 2013). Thus, the current work examines whether the
incompatibility between parses (in this case, due to the object referent) may help the parser
reject/erase the initial misparse in favor of the correct one, as in this case, only one of the parses
can accurately describe the event. This idea is summarized as the Alternative Referent hypothesis:

(5.9) Alternative Referent hypothesis

A garden-path reanalysis that provides a specific object referent (e.g., “letter”) that
does not match the object referent in the initial misparse (e.g., “book”) may (i)
facilitate garden-path re-analysis and (ii) reduce lingering misinterpretations.

For instance, with reflexive verbs (ex.5.6b), the misparse leads to the interpretation that the
child was bathed by the mother, and the successful parse leads to the interpretation that the mother
bathed herself: the object referent is either the child or herself/the mother. These interpretations
are not compatible with one another, i.e., they cannot both accurately describe the same event of
bathing. In contrast, with optionally transitive verbs in English (ex.5.6a), the misparse leads to the
interpretation that the man hunted the deer, whereas in the successful parse the man hunted
something unspecified, for which there is no referent. In this latter case, the two parses are not as
174
incompatible: both “hunting the deer” and “hunting [something]” can accurately describe the same
hunting event.
In the current work, the Alternative Referent hypothesis is tested by assessing the
comprehension of garden-paths in Brazilian Portuguese. BP allows for an investigation of the
Alternative Referent hypothesis while keeping the verb itself constant. In BP, when no antecedent
referent is available, the implicit object has an existential interpretation (ex.5.8a). When a referent
is available, the implicit object has an anaphoric interpretation (ex.5.8b). The latter case leads to
parses that have incompatible object referents: thus, the hypothesis predicts that this mismatch
may facilitate re-analysis, i.e., facilitate the rejection of the initial misparse in favor of the
successful parse. In addition, the object referent mismatch across parses should also lead to fewer
cases of lingering misinterpretations: it may be easier to let go of an initial misparse that is not
compatible with the final parse than to let go of an initial misparse that is relatively more
compatible with the final parse.
Testing the Alternative Referent hypothesis sheds light on the different ways the
interpretation of implicit objects may play a significant role in comprehension. Furthermore, it can
also provide a clearer view of which factors (e.g., referent mismatch) affect the language parser’s
ability to reject/erase misparses in the re-analysis of temporarily ambiguous structures.

Second aim: Investigating the Discourse Link Cue hypothesis

The second aim of this work is to investigate if a discourse link between the implicit object
and a subsequent noun phrase functions as a cue regarding the ultimately correct parse, and
whether this cue can facilitate garden-path reanalysis.
175
In garden-path sentences with object/subject ambiguities, the final successful parse
contains an implicit object followed by a noun phrase (the temporarily ambiguous noun phrase).
Typically, garden-path studies have used constructions in which this noun phrase is definite (e.g.,
Ferreira et al. 1991, Christianson et al. 2001). Definite noun phrases often signal givenness in the
discourse, that is, they indicate that a referent has been previously evoked in discourse (e.g., Heim
1982, Prince 1992).
36
Due to this association between definiteness and givenness in the discourse,
people are able to infer the existence of a discourse link between a definite noun phrase and a
previously-evoked referent in discourse: for example, in (ex.5.10), one may infer that the definite
noun phrase “the book” corefers with the previously introduced implicit object of the verb
“reading” (i.e., people may infer that the book titled “Implicit Arguments in Syntax” is indeed the
same book Ana was reading in the park. For more on inferencing, see e.g., Trabasso & Van Den
Broek 1985, McKoon & Ratcliff 1992, Birner 1997).

(5.10) Ana was reading ∅ in the park. The book was titled “Implicit Arguments in Syntax”.

In garden-paths such as (ex.5.11a), the definite noun phrase “the deer” cannot be linked to
any previous referents in the initial misparse, because there aren’t any available. However, the
successful parse allows the definite noun phrase to be linked to the previous implicit object of
“read”. Due of the association between definiteness and givenness in discourse, people can
potentially interpret this sentence as conveying that what the man hunted (i.e., the ‘something’

36
The relationship between definiteness and newness in discourse (e.g., indefinite-new and definite-old) is not a
categorical one, but rather a robust tendency in discourse (e.g., Heim 1982, Prince 1992).
There are other factors that can influence whether a referent is introduced with an indefinite or a definite noun phrase
in discourse. For example, a referent such as “the moon” can be introduced as a definite without having been evoked
or mentioned in previous discourse, presumably because it’s uniquely identifiable (e.g., Ward & Birner 1995).
176
conveyed by the existential implicit object) was the deer that ran into the woods.

(5.11) a. After the man hunted the deer ran into the woods.

b. After the man hunted a deer ran into the woods.

Compare (ex.5.11a) to a garden-path in which the temporarily ambiguous noun phrase is
indefinite (ex.5.11b): the construction with an indefinite noun phrase does not as strongly suggest
that what the man was hunting was the deer that ran into the woods. In (ex.5.13b), “a deer” is more
easily interpreted as a new referent in discourse (e.g., Heim 1982, Prince 1992), non-coreferential
with the previously introduced implicit object. Consequently, it is much less likely that there would
be a discourse link between this indefinite noun phrase and the IP, in both the initial misparse and
the final parse.
The current work explores how the possibility of a discourse link that arises in the final
parse may affect the comprehension of these garden-path sentences. More specifically, this
discourse link may work as a cue signaling the language parser towards the ultimately correct
parse. That is, the definite noun phrase may prompt the parser to search for a previously introduced
referent that is associated with the noun phrase. In the initial misparse this search fails. However,
in the final parse the search can succeed, confirming the givenness presupposition that is often
associated with definiteness. Thus, the parser may see the success of this search as a cue that this
final parse is indeed the correct one, and that the initial misparse should be rejected. This idea is
introduced here as the Discourse Link Cue Hypothesis:

177
(5.12) Discourse Link Cue hypothesis

A garden-path reanalysis that promotes a discourse link in the ultimately successful
parse but not in the initial misparse provides a cue that may facilitate garden-path
reanalysis (relative to a garden-path reanalysis that promotes no discourse link at all
in the successful parse).

In the current work, the Discourse Link Cue hypothesis is tested by manipulating the
definiteness of the temporarily ambiguous noun phrase (as in (ex.5.11a-b)). Definites are often
associated with givenness, and thus may be interpreted as a link to a previously introduced referent.
This discourse link is only successful in the final parse, and as such may function as a cue towards
this final (and successful parse). Indefinites, on the other hand, are unlikely to be interpreted as
given in the discourse and promote no discourse link with prior referents (in both the initial and
final parses). Consequently, indefinites would not provide a cue toward the successful parse.
If the discourse link does indeed function as a cue towards the final parse, it is possible that
it would not only facilitate garden-path re-analysis (as stated in (ex.5.12)) but it could also affect
the proportion of lingering misinterpretations. This cue towards the correct parse could potentially
help the parser let go of the initial misparse. Unfortunately, this possibility cannot be investigated
in the current study. Here, what is being manipulated is the definiteness of the ambiguous noun
phrase itself (e.g., “a/the deer” in (ex.5.11a-b)). Therefore, a final interpretation in which the
ambiguous noun phrase is the direct object (e.g., that the man hunted the deer) could be due to
either (i) a lingering misinterpretation from the initial misparse or (ii) an interpretation in which a
discourse link between the implicit object and the noun phrase has been inferred. In the current
178
work, it is impossible to disentangle between these two possibilities, therefore we cannot assess
whether the Discourse Link Cue hypothesis can also help explain lingering misinterpretations.
In sum, the current work’s second aim is to investigate the Discourse Link Cue hypothesis
by assessing whether definite ambiguous noun phrases facilitate garden-path reanalysis relative to
indefinite ones. Testing this hypothesis allows us once again to shed light on how the interpretation
of the implicit object can affect the comprehension of sentences with temporary object/subject
ambiguities; this time, by focusing on the discourse link between the implicit object and a
subsequent noun phrase (made possible by the association between definiteness and givenness in
the discourse). In addition, this line of inquiry also contributes to our understanding of which
factors (e.g., a discourse cue) affect the language parser’s ability to revise an initial misparse in
temporarily ambiguous structures.
5.3. Investigating how implicit object interpretation affects the comprehension of garden-
path sentences in a self-paced reading task

A word-by-word self-paced reading task in Brazilian Portuguese was designed to
investigate how the interpretation of implicit objects affects the comprehension of garden-path
sentences with object/subject ambiguities. More specifically, this task was designed to explore two
main aims.
The first aim is to investigate the Alternative Referent hypothesis, by assessing whether
garden-path sentences that give rise to an anaphoric implicit object interpretation can facilitate
reanalysis and reduce lingering misinterpretations.
When an antecedent is available, garden path re-analysis leads to a final parse in which the
implicit object refers back to the antecedent (e.g., “After I received a letter and read [it/the letter
I received], a book fell”). In this case, there’s a mismatch between the object referent in the initial
179
parse (e.g., a book) and the final parse (e.g., a letter). The Alternative Referent hypothesis predicts
that the incompatibility between these two parses will make it easier to reject the initial misparse,
facilitating garden-path re-analysis and reducing lingering misinterpretations.
When no antecedent is available, garden-path re-analysis leads to a parsing in which the
initial verb has an existential, unspecified implicit object (e.g., “After I read [something], a book
fell”). In this case, the initial parse has a specific object referent (e.g., the book) and the final parse
does not (e.g., ‘something’). These two parses are not as incompatible with one another: the
something being read could be a book. Thus, the Alternative Referent hypothesis predicts no
facilitation effect in the comprehension of garden-paths with no available antecedent.
The second aim of the current work is to investigate the Discourse Link Cue hypothesis,
by examining whether garden-paths that give rise to a discourse link involving the implicit object
may facilitate the comprehension of these structures. This hypothesis proposes that when a
discourse link (such as the link between a definite noun phrase and a previously-evoked referent)
succeeds in the final - but not the initial – parse, this discourse link may function as a cue
encouraging the selection of this final parse.
This second aim is investigated by manipulating the definiteness of the temporarily
ambiguous noun phrase. Definite noun phrases are often associated with givenness in the discourse
and can link back to a previously evoked referent. Thus, when the ambiguous noun phrase is
definite, it can promote a discourse link between itself and a previous implicit object. The
Discourse Link Cue hypothesis predicts that the parser may use this link as a cue signaling the
correct parse, facilitating garden-path re-analysis. On the other hand, when the noun phrase is
indefinite, no discourse link is promoted. The hypothesis then predicts that no facilitation in
comprehension will occur.
180
Participants

48 native speakers of Brazilian Portuguese participated in the experiment in Rio de Janeiro,
Brazil. One participant was excluded from analysis due to low performance on filler
comprehension questions (less than 80% accuracy), leaving a total of 47 participants (who
performed with 94% accuracy on average).
5.3.1. Methods

Materials and Design

The experiment contained 32 targets and 32 filler items. Targets were constructed from a
set of 16 optionally transitive (OPT) verbs. In BP, OPT verbs can occur with overt objects, with
existential implicit objects (when there’s no antecedent/referent) or with anaphoric implicit objects
(when there’s an antecedent/referent for the anaphoric interpretation). Each of the sixteen verbs
was used twice, producing a total of 32 target items.
The experiment featured a 2x2x2 design, in which three dependent variables were
manipulated along two levels: (i) ambiguity (comma or no comma), (ii) presence of an antecedent
(present or absent) and (iii) definiteness of the ambiguous noun phrase (definite or indefinite). In
regard to the (i) ambiguity manipulation, target sentences either had a comma making the sentence
unambiguous (i.e., not a garden-path) or they lacked a comma, making the following noun phrase
temporarily ambiguous between object or subject position (i.e., the classic garden-path). Targets
also varied in terms of (ii) antecedent availability, so that half of the sentences introduced a noun
phrase before the verb that could function as a referent for an anaphoric implicit object
181
interpretation, and the other half did not introduce such a noun phrase. Lastly, the (iii) definiteness
of the ambiguous noun phrase was also manipulated so that it was either a definite or an indefinite
noun phrase.
Table 5.1 displays an example target item in all conditions. The critical region for the main
research questions (aims 1 and 2) is underlined. The current work aims to probe the reanalysis
process of these temporarily ambiguous sentences: therefore, the critical region starts at the
disambiguating verb (e.g., ‘caiu’, fell), which is the point when people have access to enough
information to realize the need to revise their initial misparse
37
. The critical region also includes
the following three words (e.g., ‘no meu pé’, on my foot), because effects may not all appear
immediately at the verb, but also in subsequent words.
Each item in the experiment was followed by a yes/no comprehension question. Filler
comprehension questions were used to ensure participants were paying attention throughout the
experiment: participants with lower than 80% accuracy on filler comprehension questions were
excluded from data analysis. Target comprehension questions inquired about the misparse
interpretation, i.e. the interpretation that takes the temporarily ambiguous noun phrase to be the
direct object of its preceding verb: e.g., “Did I read a book?” (Table 5.1, (ex.5.23)).
Sentences were all entirely in Brazilian Portuguese. Prior work has shown that in Brazilian
Portuguese, just like in English, people display a garden-path effect when processing these
temporarily ambiguous sentences, and they also yield lingering misinterpretations (e.g., Ribeiro
2004, Ferrari 2017).

37
The disambiguating verb in target sentences has third person agreement to make clear that the sentence contains an
implicit argument which is an object of the verb “read”, and not an implicit subject coreferential with the first person
“I” (which is a possibility in Brazilian Portuguese; see also Ribeiro 2004, Ferrari 2017 for garden-path studies in BP
that used parallel manipulations with number and gender agreement respectively).
182
Table 5.1 Example target items (BP on top, English translation below; ∅ indicates an implicit object)
(5.15) Unambiguous, no antecedent, indefinite

Depois que eu li, um livro caiu no meu pé esquerdo.
After I read ∅, a book fell on my foot left.
38

(5.16) Unambiguous, no antecedent, definite

Depois que eu li, o livro caiu no meu pé esquerdo.
After I read ∅, the book fell on my foot left.
(5.17) Unambiguous, with antecedent, indefinite

Depois que eu recebi uma carta e li, um livro caiu no meu pé esquerdo.
After I received a letter and read ∅, a book fell on my foot left.
(5.18) Unambiguous, with antecedent, definite

Depois que eu recebi uma carta e li, o livro caiu no meu pé esquerdo.
After I received a letter and read ∅, the book fell on my foot left.
(5.19) Ambiguous, no antecedent, indefinite

Depois que eu li um livro caiu no meu pé esquerdo.
After I read ∅ a book fell on my foot left.
(5.20) Ambiguous, no antecedent, definite

Depois que eu li o livro caiu no meu pé esquerdo.
After I read ∅ the book fell on my foot left.
(5.21) Ambiguous, with antecedent, indefinite

Depois que eu recebi uma carta e li um livro caiu no meu pé esquerdo.
After I received a letter and read ∅ a book fell on my foot left.
(5.22) Ambiguous, with antecedent, definite

Depois que eu recebi uma carta e li o livro caiu no meu pé esquerdo.
After I received a letter and read ∅ the book fell on my foot left.
(5.23) Target comprehension question
Eu li um livro?
Did I read a book?

38
Adjectives are mostly post-nominal in BP, hence the word order for “left foot” is actually “foot left”.
183
Procedure

Participants read sentences word-by-word on a computer screen, at their own pace. Words
are hidden by dashes until participants press the spacebar, revealing one word with each pressing
(previous words return to being hidden behind dashes; only one word is visible at any given time).
This method allows for the measurement of reaction times in milliseconds, i.e., how long it takes
for each word to be read (e.g., Aaronson & Scarborough 1976; Mitchell & Green 1978). The
experiment was programmed on Linger
39
entirely in Brazilian Portuguese.

Predictions

The first aim of this work is to investigate the Alternative Referent hypothesis, which states
that garden-path reanalysis may be facilitated when the initial and final parses have conflicting
referents (i.e., when the object referent in the initial misparse and the object referent in the
successful parse are different specific referents, like “a letter” vs. “a book”). This hypothesis
predicts that when an anaphoric IO interpretation occurs in the final parse it will lead to (i) a
facilitation in reanalysis, seen as faster reading times, and (ii) a reduction of lingering
misinterpretations, seen as a lower proportion of “yes” answers to a question about the initial
misparse. The antecedent effect is predicted to appear only in ambiguous conditions, which are the
ones that yield a garden-path effect. Thus, this hypothesis predicts a main effect of antecedent and
an interaction between antecedent and ambiguity.

39
Developed by TedLab and Doug Rohde (http://tedlab.mit.edu/~dr/Linger/), now supported by Lauren Ackerman
(as of 2019; https://github.com/VerbingNouns/linger3).
184
The second aim of this work is to investigate the Discourse Link Cue hypothesis. This
hypothesis proposes that when garden-path reanalysis brings about a discourse link in the final
parse, this link may function as a cue towards the selection of this parse as the ultimately correct
one and facilitate re-analysis. This facilitation is expected to be seen in faster reading times, and
is also predicted to appear only in ambiguous conditions because they are the ones that yield
reanalysis processes. Therefore, this hypothesis predicts a main effect of definiteness and an
interaction between definiteness and ambiguity.
Furthermore, by testing the comprehension of garden-paths with anaphoric implicit
objects, the current work may provide evidence towards confirming that late closure bias occurs
in a previously untested environment. Late Closure (e.g., Frazier 1978, Frazier & Rayner 1982,
Ferreira et al. 2001) states that the parser has a bias towards incorporating upcoming input as part
of the clause it is currently processing whenever this is possible. In the current study, late closure
predicts that the temporarily ambiguous noun phrase that follows the noun will initially be parsed
as the direct object of the preceding verb – even in instances that allow for an anaphoric
interpretation of the implicit object. Thus, late closure predicts a main effect of ambiguity (the
garden-path effect).
5.3.2. Results

Reading times and answers to the comprehension questions were analyzed.
First, the analysis on reading times for target sentences is reported in regard to Late Closure
bias confirmation and aims 1 and 2 of this work, followed by a brief discussion on the findings.
Second, the analysis on answers to the comprehension questions is reported, followed by a
brief discussion.
185
Section 5.4. presents a general discussion on the findings.

Results: Reading Times

Extreme reading times (faster than 100ms or slower than 4000ms) were excluded from
analysis (0.02% of total data). Reading times more than 2.5 standard deviations from the mean for
any given word in target sentences were transformed to 2.5 standard deviations from the mean
(2.76% of total data). Reading times were analyzed with mixed-effects regression using the lme4
package (Bates, Maechler, Bolker & Walker 2019) in R (R Core Team 2019), with random slopes
and intercepts for subjects and items when supported (e.g., Barr 2013).
The critical region begins at the disambiguating verb (e.g., caiu, “fell”) and includes the
three following words as the spillover region (e.g., no meu pé, “on my foot”; see Table 1,
underlined). Reading times analyses for (i) ambiguity, (ii) antecedent availability and (iii)
definiteness of the ambiguous noun phrase are described in the subsections below.
A main effect of ambiguity is predicted by late closure bias: ambiguous conditions are
expected to lead to slower reading times in the critical region, relative to unambiguous (comma-
less) conditions.
Both a main effect of antecedent and an interaction between ambiguity and antecedent are
predicted by the Alternative Referent hypothesis, the focus of the current work’s first aim.
Conditions with antecedents are expected to be read faster than those without, but only in
ambiguous conditions (which require re-analysis).
Finally, both a main effect of definiteness and an interaction between definiteness and
ambiguity are predicted by the Discourse Link Cue hypothesis, the focus of the second aim of this
186
work. Conditions with definite noun phrases are expected to be read faster than those with
indefinites, but again only in ambiguous conditions.

Late Closure Bias Confirmation

In addition to investigating its two main aims, the current work also verifies that late
closure bias occurs in the comprehension of a previously untested garden-path construction: BP
sentences with a temporary object/subject ambiguity that yield anaphoric implicit objects in the
final parse. Late closure bias predicts that in the comprehension of ambiguous conditions – but not
unambiguous ones - people will parse the temporarily ambiguous noun phrase as the overt direct
object of its preceding verb, leading to the classic garden-path effect (seen in a slowdown in RTs
for ambiguous conditions relative to unambiguous ones).
RTs for all four words in the critical region (i.e., the disambiguating verb and three words
in the spillover region: caiu no meu pé, “fell on my foot”) appear in the predicted direction
regarding ambiguity effects: words in ambiguous sentences (without a comma) were read slower
than those in unambiguous sentences (main effect of ambiguity; see Tables 5.2, 5.3, 5.4 and Figure
5.1).

First aim: Investigating the Alternative Referent Hypothesis

The Alternative Referent hypothesis suggests that the reanalysis of garden-path sentences
may be facilitated when the initial misparse and the final parse have conflicting referents (in this
case, the direct object). This hypothesis predicts that ambiguous conditions with antecedents
187
should lead to a facilitation during comprehension (seen as faster reading times) relative to
conditions without antecedents.
Reading times in the critical region do not support the Alternative Referent hypothesis
predictions regarding a facilitation during online comprehension. Sentences with antecedents did
not lead to faster reading times relative to those without antecedents. In fact, RTs at the
disambiguating verb appear in the opposite direction: sentences with antecedents were read
slightly slower at the disambiguating verb relative to sentences without antecedent (Table 5.3;
main effect of antecedent and ambiguity, see Tables 5.4 and 5.5; pairwise comparisons show main
effect of antecedent in ambiguous conditions, but not unambiguous ones, see Table 5.7).
On the three words following the disambiguating verb, there were no significant
differences in the reading times of sentences with and without antecedents.

Second aim: Investigating the Discourse Link Cue hypothesis

The Discourse Link Cue hypothesis proposes that when garden-path reanalysis yields a
final parse with a discourse link involving the implicit object, this link may function as a cue
towards the selection of this parse, facilitating reanalysis. This hypothesis predicts that ambiguous
conditions with definite noun phrases (that promote the discourse link) will lead to faster reading
times, relative to conditions with indefinite noun phrases (which do not promote a discourse link).
RTs at the critical region reveal that sentences with definite noun phrases were indeed read
faster than those with indefinites; however, that was true for both ambiguous and unambiguous
sentences (Tables 5.2 and 5.3). That is, definite noun phrases were not only faster to process
when reanalysis was needed, but rather in all cases. Analyses reveal a main effect of
188
definiteness at the verb and the following two words (Table 5.4). At the verb, a three-way
interaction between ambiguity, definiteness and antecedent also emerges (Table 5.6): planned
pairwise comparisons further reveal a significant main effect of definiteness for ambiguous
conditions, but in unambiguous conditions only conditions without antecedents show a significant
effect of definiteness at the verb (unambiguous conditions with antecedents also revealed a
significant effect of definiteness, which appeared before the critical region). At the third word in
the critical region, in addition to the main effect of definiteness, there was also a significant
interaction between definiteness and antecedent: planned pairwise comparisons show that at this
point, the definiteness effect occurs only in conditions with antecedents.

Table 5.2 Mean RTs and SE for words in the critical region by antecedent and definiteness for
UNAMBIGUOUS (with comma) conditions
Word
No antecedent With antecedent
indefinite definite indefinite definite
Mean RT SE Mean RT SE Mean RT SE Mean RT SE
“caiu” 519.443 21.875 456.937 15.996 481.705 15.611 488.075 17.629
“no” 476.730 13.862 455.938 13.143 490.312 14.918 458.684 13.360
“meu” 460.149 12.497 449.916 13.293 481.532 18.132 455.094 13.004
“pé” 454.255 12.461 433.305 11.582 455.623 13.908 450.307 11.826

Table 5.3 Table 5.3. Mean RTs and SE for words in the critical region by antecedent and definiteness for
AMBIGUOUS (no comma) conditions
Word
No antecedent With antecedent
indefinite definite indefinite definite
Mean RT SE Mean RT SE Mean RT SE Mean RT SE
“caiu” 708.770 36.011 668.183 32.413 854.544 52.207 731.614 41.765
“no” 578.432 20.454 549.535 20.140 577.813 22.811 561.345 20.332
“meu” 503.475 15.897 498.705 14.368 551.798 21.950 488.142 14.489
“pé” 478.074 14.859 499.513 15.469 485.816 16.391 465.345 13.934

189
Figure 5.1 Word-by-word reading times +-1 SE by ambiguity, antecedent and definiteness (English translation:
After that I [received a letter and] read [,] [a/the] book fell on my foot left).

Table 5.4 Reading Time Results: Main Effects for words in the critical region (“*”: p < 0.05)
Word
Main Effects
Ambiguity Antecedent Definiteness
b SE t p b SE t p b SE t p
“caiu” 127.956 20.157 6.348 0.001* 25.592 9.146 2.798 0.005* -28.438 9.146 3.109 0.001*
“no” 48.177 7.018 6.864 0.001* 3.354 5.474 0.613 0.540 -12.517 5.474 -2.286 0.022*
“meu” 24.344 4.742 5.134 0.001* 8.127 4.741 1.714 0.086 -13.293 4.742 -2.803 0.005*
“pé” 16.952 4.107 4.127 0.001* -1.195 4.107 -.291 0.771 -3.152 4.107 -0.767 0.442

Table 5.5 Reading Time Results: 2-way interactions for words in the critical region (“*”: p < 0.05)
Word
Interactions
Ambiguity*Antecedent Ambiguity*Definiteness Antecedent*Definiteness
b SE t p b SE t p b SE t p
“caiu” 27.335 9.146 2.989 0.002* -15.362 9.147 -1.679 0.093 -1.998 9.146 -0.218 0.827
“no” -0.756 5.474 -0.138 0.890 .785 5.474 0.143 0.885 0.370 5.474 0.068 0.946
“meu” 1.782 4.742 0.376 0.707 -3.792 4.742 -0.800 0.424 -9.410 4.741 -1.985 0.047*
“pé” -5.703 4.107 -1.388 0.165 3.440 4.107 0.837 0.402 -3.165 4.107 -.771 0.441

Table 5.6 Reading Time Results: 3-way interaction for words in the critical region (“*”: p < 0.05)
Word
Interaction
Ambiguity*Antecedent*Definiteness
b SE t p
“caiu” -19.155 9.147 -2.094 0.036*
“no” 3.229 5.474 0.590 0.555
“meu” -5.535 4.742 -1.167 0.243
“pé” -7.270 4.107 -1.770 0.076

190
Table 5.7 Planned Comparisons for Verb position (“*”: p < 0.05)
Word
Con-
trasts

Antecedent Definiteness Antecedent*Definiteness
b SE t p b SE t p b SE t p
Verb
“ca-
iu”
Ambig.
only

52.85 19.05 2.775
0.008
*
-44.16 16.49 -2.678
0.007
*
-21.90 16.49 -1.328 0.184
Unamb.
only
-1.947 7.243 -0.269 0.788 -13.415 8.75 -1.533 0.132 17.188 7.243 2.373
0.017
*
Unamb.
-Antec
only
- -30.50 12.89 -2.365
0.020
*
-
Unamb.
+Antec
only
- 3.288 9.711 0.339 0.735 -

Table 5.8 Planned Comparisons for Spill 2 position (“*”: p < 0.05)
Word Contrasts

Ambiguity Definiteness Ambiguity*Definiteness
b SE t p b SE t p b SE t p
“so-
fá”
No Antec.
only
22.974 6.100 3.766
0.001
*
-3.957 5.797 -0.683 0.495 1.730 5.797 0.298 0.765
W/ Antec.
only
26.089 7.480 3.488
0.001
*
-22.593 7.479 -3.021
0.002
*
-9.259 7.480 -1.238 0.216

All words outside the critical region were analyzed as well, for the sake of completeness.
These analyses revealed three noteworthy findings: first, an effect of antecedent occurs at the 8
th

and 9
th
words (e.g., li plus the following comma; “read,”), possibly due to the antecedent
manipulation that takes place immediately before these words. Secondly, a main effect of
definiteness appears in the 10
th
and 11
th
words (e.g., um/o livro, “a/the book”), but only in
unambiguous conditions: this finding is further discussed in Section 3 of this chapter, as it parallels
a definiteness effect that appears in the critical region for ambiguous conditions. Lastly, a main
effect of ambiguity occurs at the 11
th
word, immediately preceding the disambiguating verb. This
effect is likely due to differences in processing noun phrases in different grammatical positions
(e.g., object vs. subject). Details are given in Appendix 6. As these results occur prior to the critical
region and have no bearing on the hypotheses being tested in this chapter, they are not discussed
in depth here.
191
Discussion of Reading Times

RT data show that Late Closure bias occurs in a previously untested environment: garden-
paths that yield anaphoric implicit objects. That is, the possibility of resolving the direct object as
an IO referring back to an antecedent does not prevent the parser from incorporating the noun
phrase that follows the verb as a direct object.
Regarding the current work’s first aim, findings from the self-paced reading experiment
do not support the Alternative Referent hypothesis’ prediction regarding a facilitation during
comprehension. The incompatibility between the initial and the final parses did not facilitate online
re-analysis of the garden-path.
The lack of a facilitation as described in the Alternative Referent hypothesis may have
occurred because during comprehension, once the parser reaches the disambiguating verb and
decides reanalysis is needed, other factors – like the mismatch between referents across parses –
do not play a critical role in the process.
Alternatively, there are a few characteristics of garden-paths with anaphoric implicit
objects that could have potentially made them more burdensome to process relative to those with
existential implicit objects. Firstly, whereas garden-paths with existential IOs simply require
garden-path re-analysis, garden-paths with anaphoric IOs require re-analysis and they require
realizing that the implicit object is anaphoric to an antecedent rather than existential; this additional
step could have an influence on the speed with each these structures are parsed. Secondly, garden-
paths with anaphoric IOs could incur an additional processing cost of building the silent syntax
associated with these structures. These possibilities are further addressed in Section 5.4: General
Discussion.
192
Regarding the current work’s second aim, findings from the self-paced reading time did
not support the Discourse Link Cue hypothesis’ predictions about a facilitation effect in cases
where reanalysis promotes a discourse link, i.e., when the ambiguous noun phrase was definite.
This hypothesis only predicted a definiteness facilitation during reanalysis, i.e., in ambiguous
conditions; however, facilitation occurred also in unambiguous conditions.
There is no evidence that the discourse link promoted by definite noun phrases does
function as a cue towards the correct parse. It is possible that a general facilitation occurred with
definites because it might be generally easier to process because they did not introduce a new
referent to the discourse (they just referred back to an already introduced referent). Indefinite noun
phrases, on the other hand, did introduce a new referent to the discourse, which could be a more
burdensome computation. Reasons for this general ‘definiteness advantage’ and for the differences
in timing regarding where the definiteness effect appears across conditions are further explored in
section 5.4: General Discussion.

Results: Comprehension Questions

After each sentence, participants answered a yes/no comprehension question. The question
was intended to probe sentence interpretations as compatible or incompatible with the original
misparse, as done in earlier work (e.g., Christianson et al. 2001, Slattery et al. 2013). For example,
after reading a target sentence such as the ones in Table 5.1, participants were asked: “Did I read
a book?”. “Yes” responses match the initial misparse, in which “a book” has been interpreted as
the object of “read”. Thus, “yes” responses are typically indicative of lingering misinterpretations,
and “no” responses are considered the accurate response most compatible with successful
193
reanalysis.
Crucially, in sentences with definite noun phrases, “yes” responses are ambiguous: they
could be due to either (i) lingering misinterpretations or (ii) people inferring a discourse link of
correference between the definite noun phrase and the implicit object (e.g., inferring that the book
that fell must be the same something being read). On the other hand, “yes” responses after
sentences with indefinites are unambiguous: indefinites do not promote a discourse link and thus
eliminate option (ii). Therefore, only conditions with indefinites were analyzed in regards to
lingering misinterpretations. Answers were analyzed with generalized linear mixed-effects
regression using the lme4 package (Bates, Maechler, Bolker & Walker 2019) in R (R Core Team
2019), with random slopes and intercepts for subjects and items when supported (e.g., see Barr
2013).

First aim: investigating the Alternative Referent Hypothesis

To address the current work’s first aim – to investigate the Alternative Referent Hypothesis
– responses to comprehension questions were analyzed
40
. The Alternative Referent hypothesis
suggests that garden-paths in which the initial and final parses have conflicting (object) referents
may make it easier for the parser to reject/erase the initial misparse. Consequently, this hypothesis
predicts that conditions with antecedents will lead to a reduction in lingering misinterpretations,
seen as a lower proportion of “yes” responses relative to conditions without antecedents.

40
The second aim of this work, investigating the Discourse Link Cue hypothesis, did not include predictions about
comprehension questions. However, section 3 of this chapter includes a discussion on how the answers to
comprehension questions can still be informative regarding a discourse link involving the implicit object when the
noun phrase is definite.
194
Answers to target comprehension questions do support the Alternative Referent hypothesis
predictions regarding a reduction in lingering misinterpretations. As predicted, sentences with
antecedents led to a reduction in the proportion of “yes” responses, which indicate lingering
misinterpretations (main effects of ambiguity, antecedent and an interaction; see Table 5.9, 5.10
and Figure 5.2. Planned pairwise comparisons reveal a main effect of antecedent in ambiguous
conditions, see Table 5.11. Unambiguous conditions also show an effect of antecedent, see Table
5.12; the difference between conditions here is much smaller, and this finding is further discussed
in Section 5.4). Ambiguous conditions without antecedents yielded 51.6% “yes” responses to
questions such as “Did I read a book?”. In contrast, ambiguous conditions with antecedents yielded
approximately 22.9% of “yes” responses. Lingering misinterpretations were so greatly reduced in
conditions with antecedents, that in fact they were virtually eliminated: unambiguous conditions
with antecedents, which do not lead to misparses, display almost the same proportion of “yes”
responses relative to their ambiguous counterparts (18.6% and 22.9% respectively; an additional
analysis reveals that these conditions do not significantly differ from one another).

Table 5.9 Proportion of ‘yes’ responses +-1 SE by ambiguity, antecedent availability and definiteness of NP
Ambiguous Unambiguous
Indefinite Definite Indefinite Definite
Prop. SE Prop. SE Prop. SE Prop. SE
W/ antecedent
0.229 0.030 0.229 0.030 0.186 0.028 0.181 0.028
No antecedent
0.516 0.036 0.585 0.036 0.277 0.032 0.511 0.036

195
Figure 5.2 Proportion of ‘yes’ responses to comprehension question +-1 SE

Table 5.10 Comprehension Question Results: Main effects and interactions (Indefinites only; “*”: p < 0.05)

Ambiguity Antecedent Ambiguity*Antecedent
b SE z p b SE z p b SE z p
0.459 0.102 4.486 0.001* 0.653 0.105 6.166 0.001* -0.294 0.101 -2.894 0.003*

Table 5.11 Comprehension Question Results: Planned Pairwise comparisons (Indefinites only; “*”: p < 0.05)

Contrasts

Antecedent
b SE z p
Ambiguous
Only
0.921 0.154 5.964 0.001*
Unambiguous
Only
0.391 0.159 2.451 0.014*

Discussion (Comprehension Questions)

Answers to comprehension questions support the Alternative Referent hypothesis’
predictions about a reduction in lingering misinterpretations when there is a mismatch between the
object referent in the initial and the final parses.
196
Although the referent mismatch did not facilitate re-analysis online, it did practically
eliminate the occurrence of lingering misinterpretations. These findings support the idea that the
alternative referent can help the parser erase the lingering initial misparse from active memory
after re-analysis. This finding as well as additional observations that arise from the answers to
comprehension questions are further discussed below in Section 5.4: General Discussion.
5.4. General Discussion

Things that are unsaid can still affect language comprehension in unexpected ways,
however not much is known about how various kinds of implicit arguments - such as existential
and anaphoric implicit objects - participate in comprehension processes. The current chapter aims
to shed light on this topic by examining implicit objects’ role in a particular construction: garden-
path sentences with object/subject ambiguities. More specifically, this chapter takes a closer look
at how the interpretation of an implicit object - either as an existential, as anaphoric to an
antecedent or inferred as coreferential with a subsequent noun phrase - affects garden-path
processing and reanalysis, as well as the likelihood that the initial misparse will linger.
More specifically, I focused on two main aims related to the comprehension of implicit
objects in garden-path sentences. The first aim was to investigate the Alternative Referent
hypothesis, which predicts that a garden-path reanalysis that yields an anaphoric IO with a referent
that differs from the initial misparse referent may (i) facilitate reanalysis and (ii) reduce lingering
misinterpretations. The second aim was to explore the Discourse Link Cue hypothesis, which
proposes that a garden-path reanalysis that yields a discourse link involving the implicit object
may facilitate reanalysis. In the course of investigating these two aims, the current work also
assessed late closure bias in a previously untested structure.
197
Late Closure Bias confirmation

Firstly, the experiment confirmed that late closure bias occurs in a previously untested
environment, i.e., BP garden-paths that allow an anaphoric implicit object interpretation post
reanalysis. Results showed that a noun phrase following the verb in canonical direct object position
is indeed processed as the direct object in the initial parse, even when there was an antecedent
available for the anaphoric implicit object interpretation.

First aim: Investigating the Alternative Referent hypothesis

When an anaphoric implicit object interpretation is available, the initial misparse (in which
the ambiguous noun phrase is the object) and the ultimately successful parse (in which the
antecedent is the object) are incompatible. In this case, there is a mismatch between object
referents: e.g., in the initial parse “a book” is read, whereas in the final parse it is “a letter” that is
read. Due to this incompatibility – the two parses cannot both be true of the same event – , the
Alternative Referent hypothesis predicts that an anaphoric implicit object interpretation could
potentially make it easier to reject/erase the initial misparse, (i) facilitating garden-path reanalysis
and (ii) reducing lingering misinterpretations
The Alternative Referent hypothesis’s prediction regarding (i) a facilitation during online
comprehension was not born out: the anaphoric implicit object interpretation did not facilitate
garden-path processing. In fact, contrary to expectations, garden-paths with antecedents were
read slower than those without.

198
Why didn’t the parse with an alternative referent (the anaphoric implicit object) facilitate
garden-path reanalysis? It may be that once the parser has enough information to realize reanalysis
is necessary (i.e., once the disambiguating verb is reached), other factors play little to no role in
this process during comprehension.
There are other reasons that could have potentially contributed to conditions with
antecedents leading to slower reading times. For instance, sentences with antecedents offer one
additional route for re-analysis: in these cases, once people reach the disambiguating verb and
realize the noun phrase cannot be the overt object of the previous verb, they now have two options
to select from: the preceding verb either has an existential IO or an anaphoric IO. In contrast,
sentences without antecedents offer only one option for re-analysis: an existential IO. Perhaps the
extra burden of having to select between two options instead of just one is responsible for the
relatively slower reading times for sentences with antecedents.
Another possible explanation as to why garden-paths with antecedents were read slower
than those without antecedents may have something to do with their syntactic structure: garden-
paths with antecedents may incur an additional processing cost due to the extra burden of
integrating the anaphoric implicit object in the syntax (as null anaphora or as ellipsis; see Kato
1993, Cyrino 1996, 2016). This extra burden does not occur with existential implicit objects
because they are not represented in the syntax (e.g., Williams 2015).
41

When it comes to the analysis of lingering misinterpretations, findings confirm the
Alternative Referent hypothesis’ prediction: garden-paths that yield anaphoric implicit objects

41
This “syntactic burden” account does not entirely match findings from Chapter 4, which had revealed no differences
in readings times for verbs like “read” with and without antecedents (i.e., with anaphoric and existential IOs).
However, if there is indeed a “syntactic burden” related to the syntactic integration of an anaphoric implicit object, it
could have gone by unnoticed in the prior experiment. It is possible that in the current experiment, difficulties related
to the garden-path effect and difficulties related to this syntactic burden occurring simultaneously added up and
amplified the syntactic burden to the point of statistical significance.
199
drastically reduced the occurrence of lingering misinterpretations. That is, despite the absence
of an antecedent facilitation during online comprehension, antecedent availability did play a
significant role in preventing lingering misinterpretations.
Slattery et al. (2013) presents evidence indicating that lingering misinterpretations occur
because the initial misparse interpretation is still active in memory, i.e. it has not been fully erased.
Findings from the current work suggest that when the initial and final parses have conflicting
referents, the initial misparse is more easily erased from memory – in accordance to the Alternative
Referent hypothesis. This hypothesis also explains why Christianson et al. (2001) found less
lingering misinterpretations with reflexives relative to optionally transitive verbs: reflexives
provide an alternative referent.
Interestingly, the anaphoric implicit object interpretation in the current study seemed to
have reduced lingering misinterpretations at a higher rate than reflexives in Christianson et. al
(2001)’s study. They found that ambiguous conditions with reflexives yielded 57.3% “yes”
responses, whereas optionally transitive verbs (without antecedents) led to 71.9% “yes” responses.
However, note that Christianson et al. found an overall higher rate of “yes” responses in ambiguous
conditions: whereas their garden-paths with optionally transitive verbs like “read” had 71.9%
“yes” responses, the ones in the current study had 58.5% at the most. Differences between the two
studies may help explain this discrepancy: Christianson et al. (2001) used long ambiguous noun
phrases in their experiment (e.g., “the letter that was long and eloquent”), which they have shown
increases the overall proportion of lingering misinterpretations relative to short phrases like the
ones used in the current study (e.g., “the letter”).
What would the Alternative Referent hypothesis predict for cases in which an anaphoric
coreferential relationship between the implicit object and an antecedent is not available (e.g., with
200
optionally transitive verbs in English, like “read”)? In such cases there is still the possibility that a
final parse could have an alternative referent via inferencing (see Trabasso & Van Den Broek 1985,
McKoon & Ratcliff 1992, Birner 1997). In “After I received a letter and read a book fell on my
left foot.”, although the anaphoric implicit object interpretation is not allowed in English, one can
still infer that what I read was the letter I received. The matter of how often inferences like this
occur and whether they can prevent lingering misinterpretations at a similar rate as anaphoric IOs
remains an empirical question.

Second aim: investigating the Discourse Link Cue hypothesis

The second aim of this work was to investigate the Discourse Link Cue hypothesis. This
hypothesis suggests that when the final parse promotes a discourse link involving the implicit
object, it could function as a cue towards the ultimately correct parse, facilitating reanalysis. Thus,
garden-paths in which the temporarily ambiguous noun phrase is definite – and as such can
promote this discourse link – may lead to a facilitation in reanalysis.
Surprisingly, definites facilitated the comprehension of both ambiguous and
unambiguous sentences relative to indefinites. These findings do not support the Discourse Link
Cue hypothesis, because definites did not just facilitate reanalysis – they also led to faster reading
times when no reanalysis was needed.
One possible explanation for this general facilitation effect is that definite noun phrases
were easier to process because they indicate givenness and promoted a discourse link with a
previously evoked referent. In contrast, indefinite noun phrases were interpreted as introducing a
new referent in discourse. That is, it might be less burdensome for the parser to try to incorporate
201
a referent that is related to previous discourse than to add a new, unrelated referent to the discourse
model. Answers to comprehension questions support the assumption that definite noun phrases,
but not indefinite ones, were interpreted as referring to the same entity as the implicit object.
Moreover, the timing of the facilitation generated by definite noun phrases also supports
this idea: in unambiguous sentences, differences in reading times for definites and indefinites
begins even earlier than the critical region, at the ambiguous noun phrase itself. This early effect
possibly occurs because the comma makes it clear from earlier on that the preceding verb has an
implicit object. At the point people encounter the noun phrase, they already know there was a
previously evoked implicit object that the definite noun phrases can corefer with.
In ambiguous sentences, people parse the noun phrase as an overt direct object until they
reach the disambiguating verb: only then, at the point of re-analysis, do people find out that the
previous verb had an implicit object. That is why it is only at the disambiguating verb that we see
a difference in reading times between definite and indefinite noun phrases in ambiguous
conditions.
Note also that the definiteness advantage appears both in conditions with and without
antecedents: this suggests that even when there already was a viable referent to corefer with the
implicit object (the antecedent), subsequent definite noun phrases were still read faster and
potentially evaluated as linked to the implicit object. That is, the anaphoric IO interpretation was
not set in stone by the time the noun phrase was parsed.
The Discourse Link Cue hypothesis could not be assessed in regard to lingering
misinterpretations: when the noun phrase is definite, “yes” responses cannot be straightforwardly
attributed to lingering misinterpretations. In the current experiment, it is the temporarily
ambiguous noun phrase itself that has been manipulated in terms of definiteness (e.g., “a book”
202
vs. “the book”). As previously discussed, when the noun phrases is definite, people can infer a
discourse link between the noun phrase and the previous implicit object. Consequently, “yes”
responses following conditions with definite noun phrases are ambiguous: they could be due to
either (i) a lingering misinterpretation from the initial misparse or (ii) an interpretation in which a
discourse link between the implicit object and the noun phrase has been inferred. In the current
work, these two possibilities cannot be disentangled.
However, comparing the answers following conditions with indefinites and conditions with
definites is still informative, insofar as it sheds light on whether or not people do infer a discourse
link between the noun phrase and the previous implicit object. To understand how, we must first
turn to unambiguous conditions without antecedents:
Unambiguous conditions contain a comma separating the verb from the following noun
phrase, and thus do not tend to generate misparses nor lingering misinterpretations. Moreover, in
conditions without antecedents, the only remaining referent available to be interpreted as the direct
object is the temporarily ambiguous noun phrase itself. Thus, in unambiguous conditions without
antecedents, “yes” responses to questions such as “Did I read a book?” suggest that people
interpreted the implicit object as having the same referent as the subsequent noun phrase (i.e., via
a discourse link).
Results indicate that in unambiguous conditions without antecedents, people answered
“yes” much more often when the noun phrase was definite than when it was indefinite (51% and
27% “yes” responses respectively). These patterns confirm that people are indeed much more
likely to infer a discourse link between the noun phrase and the implicit object when the noun
phrase is definite, indicating discourse givenness.
42

42
Why do indefinites, which do not indicate discourse givenness, yielded 27% “yes” responses in unambiguous
conditions without antecedents? These responses are likely to be due to errors/lack of attention and/or not noticing the
203
Responses to comprehension questions raise an interesting additional point about how the
implicit object is interpreted, specifically regarding how often the implicit object was interpreted
as linked to the antecedent (aim 1) versus how often the implicit object was interpreted as linked
to the subsequent definite noun phrase (aim 2). This point can be observed by taking a closer look
at unambiguous conditions: responses seem to indicate that people adopted anaphoric
interpretations more often than they inferred a link between the implicit object and a subsequent
definite noun phrase. When an antecedent was present, people answered “no” about 82% of the
time - which suggests they adopted the antecedent as the object referent at this rate. This pattern
is the same for conditions with definites and indefinites, indicating that the antecedent is
interpreted as the object referent even when there is the possibility of a discourse link between the
implicit object and the subsequent noun phrase (i.e., when the noun phrase is definite). In contrast,
when no antecedent was present and the noun phrase was definite, people answered “yes” about
51% of the time, which can be taken as a measure of how often people inferred a discourse link
between the noun phrase and the implicit object (or, at least, a ceiling of how often).
These patterns suggest that the parser prefers to connect the implicit object with a referent
that came before (82%) relative to one that comes after it (51%). This preference could be due to
two reasons: (i) the implicit object may be interpreted as anaphoric when preceded by an
antecedent and as existential when there is no antecedent, in which case these patterns suggest that
strict corefence via an anaphoric relation (either as pro or ellipsis) takes precedence over an
inference of a discourse link, and that strict coreference also seems to occur more easily and
frequently than this inference. Alternatively, (ii) the implicit object may be interpreted as anaphoric
in both cases; this would suggest that anaphora is preferred over cataphora for anaphoric implicit

comma while reading target sentences. Prior work has found similar rates of “yes” responses for unambiguous garden-
path conditions with commas (e.g., see Christianson et al. 2001).
204
objects in BP.

In sum, although they share the same invisible/unpronounced form, anaphoric and
existential implicit objects affect comprehension in varied ways. The current experiment shows
that in garden-path sentences with temporary object/subject ambiguities, final parses with
anaphoric implicit objects lead to a drastic reduction in the occurrence of lingering
misinterpretations (relative to final parses with existential implicit objects). The online reanalysis
of garden-paths with anaphoric and existential implicit objects also seems to operate differently –
as seen in the slower reading times for garden-paths with anaphoric implicit objects than those
with existential ones – however, the reasons behind those differences are not entirely clear.
Moreover, the current experiment also shed more light on the parser biases’ regarding the
interpretation of implicit objects in Brazilian Portuguese: although implicit objects can be inferred
to participate in a discourse link with a subsequent definite noun phrase, this interpretation occurs
less frequently than one in which the implicit object is coreferential with a preceding antecedent.
And, when both options are available, the latter seems to be chosen.

205
Chapter 6 : General Discussion and Conclusions

This dissertation set out to deepen our understanding of how arguments that have not been overtly
expressed participate in people’s mental models of the discourse and comprehension processes. A
series of experiments were conducted to investigate existential implicit objects’ behavior in
discourse, i.e., their accessibility and persistence. Moreover, experiments assessed how existential
and anaphoric implicit objects affect language comprehension.
Existential implicit arguments, such as the direct object in “Ana has been writing.”, have
generally been regarded as discourse “inert” because they do not seem to interact much with the
wider discourse. They don’t refer to specific referents (they are left unspecified), and it is hard to
refer to them with reduced expressions like pronouns. Thus, existential implicit objects seem like
they are barely even represented in people’s mental models of the discourse: as if the direct object,
when left unsaid, becomes entirely inconsequential to the unfolding of the discourse. However,
this is not really the case: these unmentioned implicit objects are not as inaccessible to reduced
referring expressions as initially thought, and they are sometimes referred to with pronouns.
Moreover, despite being left ‘unsaid’, existential implicit objects can still “persist” and often come
up again in subsequent discourse - under some conditions, they appear in subsequent discourse as
often as overt objects (Chapters 2 and 3).
Whether or not existential implicit objects persist in discourse seems to be modulated by
prior context: when the context preceding an implicit object helps ‘narrow down’ its identity, the
implicit object is brought up again in subsequent discourse more often. Not only are these implicit
objects not simply ‘insignificant’ to people’s mental discourse models, their perceived relevance
to the structuring of discourse interacts with other factors, such as prior context (Chapter 3).
206
This dissertation also investigated how the language processor deals with these ‘unsaid’
arguments during comprehension. For example, how can the parser differentiate between
existential and anaphoric implicit objects, given that they share the same unspoken/invisible form?
It turns out that the language processor is quite efficient in identifying which types of implicit
objects different verbs allow. Verbs that allow only anaphoric implicit objects but occur in
sentences with no available referent for the anaphoric interpretation lead to comprehension
difficulties, confirming that these types of implicit objects are truly context-dependent. Just like
with the personal pronouns “she” or “he”, the parser faces difficulties when it cannot identify a
value from context to supply the meaning for the referential expression. In contrast, verbs that
allow both existential and anaphoric implicit objects can occur in contexts without referents
available and do not generate comprehension difficulties, because the parser identifies that the verb
can have a context-independent implicit object interpretation. The latter parallels what occurs with
the pronoun “they”, which can be used both to refer to a specific referent or in a more generic
sense, like “institutional they”. When used in a generic sense, “they” is context-independent.
Consequently, when the pronoun “they” occurs without an available referent in context, no
comprehension difficulties occur (Chapter 4).
Whether an implicit object is interpreted as existential or anaphoric can also have
ramifications regarding other aspects of language comprehension. In particular, how an implicit
object is interpreted affects the comprehension of garden-path sentences with temporary
object/subject ambiguities. Because of their temporary ambiguity, these types of sentences are
generally erroneously parsed at first; then, this initial misparse is re-analyzed in favor of the correct
parsing. However, even after re-analysis the initial misparse often ‘lingers’, presumably because
it stays active enough in memory and is not fully replaced by the ultimate parsing. An experiment
207
in this dissertation shows that this is only the case when the implicit object in the ultimate parsing
is an existential implicit object; if this implicit object can be interpreted anaphorically, lingering
misinterpretations completely go away. These findings suggest that having conflicting specific
referents in the initial misparse and the ultimate parsing facilitates erasing the initial misparse from
memory (Chapter 5).
Overall, this dissertation sheds light on how the human language processor is able to deal
with these ‘unsaid’ objects by recruiting various kinds of resources during comprehension. For
instance, the processor utilizes general syntactic parsing biases such as Late Closure to first
establish whether the direct object is overt or not. The parser is also shown to quickly and
effortlessly be able to access information about the verb and its argument structure, including its
implicit arguments and their specific referring functions. Moreover, the parser is able to pinpoint
and recruit information from the context in order to (i) identify suitable referents that can supply
meaning to context-dependent implicit objects and (ii) assess the domain of interpretation of
context-independent implicit objects. Finally, the parser also utilizes knowledge about discourse
structure to select appropriate referring expressions to refer back to implicit objects, and to
determine the likelihood that implicit objects may appear in subsequent discourse. These examples
show that the comprehension of these ‘unsaid’ elements requires the language processor to employ
a wide range of resources.

More general insights
The research presented in this dissertation delved into a relatively underexplored domain,
that of implicit objects in discourse and comprehension. Taking a closer look at implicit objects
208
has also shed light on - and raised questions about - more general phenomena beyond the particular
realm of implicit objects.
For example, the fact that prior context affects existential implicit objects’ persistence in
discourse, but not their accessibility (Chapter 3) highlights a dissociation between these two
discourse measures that remains poorly understood. Although persistence has been proposed to be
a deterministic factor in people’s assessments of referents’ accessibility, the findings reported in
this dissertation pile onto a series of prior studies which have shown that these two measures do
not always walk hand in hand.
Moreover, while prior work has found instances of the dissociation between accessibility
and persistence in regard to other types of referential expressions, the current work reveals that
this dissociation expands to the domain of implicit expressions; and in addition, the current work
reveals a novel finding: that a factor other than the form or grammatical position of the referent
itself – i.e., prior context – is able to affect referential persistence but not accessibility (Chapter 3).
This suggests that when people are building a mental model of the discourse, they might make use
of different sources of information when they assess whether or not a referent will appear again in
subsequent discourse (persistence) relative to when they assess how activated a referent is and
which expressions to use when referring to it (accessibility).
Notably, prior context affected the persistence of implicit objects, but not the persistence
of overt objects. Again, this suggests that some aspects about a referent - for example, its
form/implicitness - influences which other sources of information are considered in the assessment
of referents’ persistence in discourse.
Investigations about implicit objects in comprehension have also provided more general
insights about comprehension processes. For instance, it had been shown previously that the
209
language processor is quite efficient at recognizing verbs’ argument structures during
comprehension. This turns out to be the case also in regard to the kinds of implicit arguments that
are part of a verb’s argument structure: when there is no overt object, the language processor is
able to quickly assess whether a verb had an existential or anaphoric implicit object (Chapter 4).
Investigations on the interpretation of implicit objects have also provided insights
regarding the phenomena of ‘lingering misinterpretations’. These typically occur following the
comprehension of temporarily ambiguous sentences, also known as ‘garden-path’ sentences.
‘Lingering misinterpretations’ are instances in which an erroneous initial misparse remains active
in memory even after the misparse is re-analyzed and a different parsing is achieved. However, it
seems that whether or not the initial misparse remains active depends on features of the final
parsing: if the final parsing has an existential implicit object, lingering misinterpretations occur. If
the final parsing has an anaphoric implicit object, lingering misinterpretations are greatly reduced,
practically eliminated (Chapter 5). These findings suggest that whether or not an initial parsing
remains active in memory post-reanalysis might be related to the nature of the final parsing: when
the two parses are incompatible with one another (e.g., each contains a different referent
interpreted as the direct object), the initial misparse is more easily put aside in favor of the new
parsing.
Looking at implicit objects in the context of garden-path sentences has also expanded our
knowledge about the language processor’s biases during comprehension. One such bias, ‘late
closure’, refers to the idea that the processor has a tendency to use new input to continue parsing
a structure whenever possible, rather than starting a new structure. This occurs in garden-path
sentences with a temporary object/subject ambiguity: when a noun phrase follows an optionally
transitive verb, this noun phrase tends to be interpreted as the verb’s direct object rather than the
210
subject of a new clause. The current work shows that this tendency occurs even when there is an
antecedent allowing an anaphoric implicit object interpretation for the optionally transitive verb.
That is, even when the processor could potentially ‘close’ the initial clause by processing the direct
object of a transitive verb as an anaphoric implicit one, the processor remains biased towards
incorporating new input - the overt noun phrase - as part of the clause it is currently processing
(Chapter 5).

A note on syntax
As discussed in Chapter 1, the syntactic status of implicit objects is not entirely clear.
Anaphoric IOs are thought to be represented in the syntax (as either null anaphora or DP ellipsis;
e.g., Farrell 1990, Kato 1993, Cyrino 1996, Cyrino & Lopes 2016). The status of existential IOs is
murkier: due to their inability to exert syntactic control and be felicitous antecedents for pronouns,
they have been suggested to not have a presence in the syntax at all (e.g., Martí 2011, Williams
2015). An alternative view proposes that existential IOs might be what Landau (2010) describes
as “weak implicit arguments” (WIAs), in contrast with “strong implicit arguments” (SIAs) like
anaphoric IOs. According to this view, SIAs have a full syntactic representation, consisting of a
determiner phrase and bundle of semantic features. This determiner allows SIAs to refer back to
previous referents and exert syntactic control. Crucially, WIAs lack the determiner – and that
would explain why they exhibit a different, ‘weaker’ syntactic behavior relative to SIAs (such as
the inability to control or be antecedents of pronouns).
Although assessing implicit objects’ syntactic status was not a goal of this dissertation, its
findings help shed light on this matter. In particular, there are two main findings that speak to
existential IO’s presence in the syntax:
211
Firstly, the finding that pronouns are used to refer back to existential IOs (as seen in
Chapter 3). Existential IOs may differ from other existential implicit arguments in this regard –
hence why empirical research with various kinds of implicit arguments is still needed and will help
clarify this issue. Nonetheless, the claim that existential IOs’ inability to be pronoun antecedents
indicates they are missing from the syntax loses its strength once we see that they can, in fact, be
pronoun antecedents.
Secondly: the responses in the sluicing completion task show that the locality preference
in the resolution of wh-remnants occurs even when the most ‘local’ antecedent is an existential IO
(as seen in Chapter 2 Experiment 3). This locality bias hinges on antecedents’ proximity to the
wh-remnant in the syntactic structure (see e.g. Frazier and Clifton 1998, Carlson et al. 2009, Harris
et al. 2015) – thus, it would be surprising that existential IOs were chosen as antecedents more
often than overt subjects if they have no presence in the syntax at all.
Together, these findings suggest that perhaps existential IOs might not be completely
negligible in the syntax: they might just have a ‘weaker’ representation, as proposed by Landau
(2010).

Future research

This dissertation has opened up many paths for future research that can help further our
understanding of how the human mind builds and manages representations of the discourse that
they’re engaged with, in particular when it comes to referents that were implicitly introduced into
the discourse.
212
This dissertation has shown that implicit objects can (albeit somewhat rarely) be referred
to with pronouns, but we do not know which factors, if any, contribute to whether or not reference
with pronouns is felicitous. Is implicit objects’ accessibility in discourse fully determined by their
implicit nature? And if so, are all implicit arguments equal when it comes to their participation in
the discourse model?
While this dissertation focused on the domain of implicit objects, it is still unclear if all
phenomena that fall under the ‘implicit argument’ umbrella exhibit the same discourse behavior.
Crucially, implicit arguments vary in how much information about them can be retrieved from
what has been said/written overtly: for instance, implicit objects are always accompanied by their
verbs, which provide some information about the nature of the argument (e.g., the implicit object
of “bake” is a typically baked good, the implicit object of “read” is something people typically
read like a book or magazine, etc.). In contrast, implicit agents of short passives have very little
information about them provided, other than that they are agents (e.g., in “The ship was sunk.”,
we infer that someone or something has done the sinking). Implicit instruments, on the other hand,
seem more parallel to implicit objects: a verb such as “cut” (e.g., in “She cut the thing”) implies
that something typically used in cutting events has been used. However, in this case, it seems like
the object of the verb also contributes relevant information: the object used in an event such as
“She cut the cheese” is likely very different from an object in an event like “She cut the paper” or
“She cut the wood”. Does the amount of information that people can infer about an implicit
argument affect how they perceive their status in discourse? Chapter 3 shows that a prior context
that helps ‘narrow down’ the identity of the implicit object affects its persistence in discourse.
Thus, is it possible that other sources of information about the implicit argument may influence
how likely they are to participate in subsequent discourse.
213
Similarly, even within implicit objects there are differences that could potentially influence
their discourse behavior. Whereas the inquiries in this dissertation focused on the class as a whole,
it may be worthwhile to also dig deeper into their differences. Some verbs yield more narrow
inferences regarding the identity of the implicit object than others: the verb “bake” implies the
baking of any typical baked good, whereas the verb “drive” suggests that the subject drove a car,
and more specifically, their own car/the car they typically drive. Relatedly, the implicit object of
“bake” may be singular or plural (e.g., a cake or a batch of cupcakes) but the implicit object of
“drive” is most likely singular. Finally, verbs that allow existential implicit objects also vary in
their relationship to the object itself: some, like “bake”, are creation verbs (the object of baking
comes into existence via the event of baking it) and others are not (e.g., “drive”, “read”). These
differences may affect how people incorporate these various implicit objects into their mental
models of the discourse, and how people assess their accessibility and persistence.
Additionally, while this dissertation presented initial steps in revealing more about the
comprehension of existential and anaphoric implicit objects, there is much more to be uncovered
in this domain.
Chapter 4 presented an inquiry on whether an anaphoric implicit object referencing back
to an existential implicit object (e.g., in Brazilian Portuguese, something like “He cooked ∅k then
put ∅k in the fridge”) would cause any slowdowns relative to when an anaphoric implicit object
refers back to an overt referent. No effects were found, indicating either (i) that anaphoric implicit
objects can refer as easily to overt referents as to implicit ones or (ii) that the task was not sensitive
enough to detect any differences between these two scenarios (which is not unlikely given that the
effect when there was no antecedent at all was already quite small). Additional inquiries are needed
to clarify this matter, and also to understand whether reference with other expressions such as
214
pronouns would provide similar results during online comprehension (e.g., “He cooked ∅k then
put itk in the fridge”).
Another intriguing matter that has come up in the investigations from this dissertation
relates to how the language processor finds a suitable referent when interpreting an anaphoric
implicit object. Findings from Chapter 5 reveal that people were very likely to interpret the implicit
object anaphorically when the antecedent preceded the implicit object, as expected. However, non-
garden-path sentences in which a suitable definite referent appeared after the implicit object led to
fewer interpretations in which the object correfered with the referent. Is this pattern simply an
indication that anaphoric implicit objects in Brazilian Portuguese disprefer cataphora? Further
experimental work on this domain will help clarify how the language processor operates when
interpreting an anaphoric implicit object, and could also shed light on its syntactic nature (i.e.,
whether they are null pro or cases of ellipsis).

215
References

Aaronson, D., & Scarborough, H. S. (1976). Performance theories for sentence coding: Some
quantitative evidence. Journal of Experimental Psychology: Human Perception and
Performance, 2(1), 56.
Almor, A. (1999). Noun-phrase anaphora and focus: The informational load
hypothesis. Psychological review, 106(4), 748.
AnderBois, S. (2012). Indefiniteness and the typology of implicit arguments. In Proceedings of
the West Coast Conference on Formal Linguistics (WCCFL 30) (pp. 232-240).
Ariel, M. (1990). Accessing Noun Phrase antecedents. London: Routledge.
Ariel, M. (2001). Accessibility theory: An overview. Text representation: Linguistic and
psycholinguistic aspects, 8, 29-87.
Arnold, J. E. (2001). The effect of thematic roles on pronoun use and frequency of reference
continuation. Discourse processes, 31(2), 137-162.
Asher, N., & Lascarides, A. (2003). Logics of Conversation. Cambridge: Cambridge University
Press.
Bach, E. (1979). Control in Montague grammar. Linguistic inquiry, 10(4), 515-531.
Barr, D. J., Levy, R., Scheepers, C., & Tily, H. J. (2013). Random effects structure for confirmatory
hypothesis testing: Keep it maximal. Journal of memory and language, 68(3), 255-278.
Besserman, A., Kaiser, E. (2017) Reading or Reading a Book? Comprehenders’ expectations about
verb transitivity. Talk at the 27th Annual Meeting of the Society for Text & Discourse
(ST&D2017) in Philadelphia PA.
216
Besserman, A., Kaiser, E. (2018). The discourse status of existential implicit arguments:
Investigating referent accessibility through sluicing. In Proceedings of the 54
th
Annual Meeting
of the Chicago Linguistic Society (CLS) at University of Chicago, IL.
Bhatt, R., & Pancheva, R. (2006). Implicit arguments. The Blackwell companion to syntax, 2, 554-
584.
Birner, B. J. (1997). The Linguistic Realization of Inferrable Information. Language &
Communication, 17(2), 133-47.
Bock, J. K., & Warren, R. K. (1985). Conceptual accessibility and syntactic structure in sentence
formulation. Cognition, 21(1), 47-67.
Bock, K. (1987). Exploring levels of processing in sentence production. In Natural language
generation (pp. 351-363). Springer, Dordrecht.
Boland, J. E., Tanenhaus, M. K., & Garnsey, S. M. (1990). Evidence for the immediate use of verb
control information in sentence processing. Journal of Memory and Language, 29(4), 413-432.
Boland, J. E. (1993). The role of verb argument structure in sentence processing: Distinguishing
between syntactic
and semantic effects. Journal of Psycholinguistic Research, 22(2), 133-152.
Brennan, S. E., Friedman, M. W., & Pollard, C. J. (1987). A centering approach to pronouns.
In Proceedings of the 25th annual meeting on Association for Computational Linguistics (pp.
155-162). Association for Computational Linguistics.
Britt, M. A., Perfetti, C. A., Garrod, S., & Rayner, K. (1992). Parsing in discourse: Context effects
and their limits. Journal of memory and language, 31(3), 293-314.
217
Brocher, A., Chiriacescu, S. I., & von Heusinger, K. (2018). Effects of information status and
uniqueness status on referent management in discourse comprehension and
planning. Discourse Processes, 55(4), 346-370.
Carlson, K., Dickey, M. W., Frazier, L., and Clifton, C. Jr. (2009). Information structure
expectations in sentence comprehension. Q. J. Exp. Psychol. 62, 114–139.
Carlson, L. (1983). Dialogue Games: An Approach to Discourse Analysis. Reidel, Dordrecht.
Chafe, W. L. (1976). Givenness, contrastiveness, definiteness, subjects, and topics and point of
view. In C. Li (Ed.), Subject and topic (pp. 27-55). New York: Academic Press.
Chambers, C. G., & Smyth, R. (1998). Structural parallelism and discourse coherence: A test of
centering theory. Journal of Memory and Language, 39(4), 593-608.
Chiriacescu, S., & von Heusinger, K. (2010). Discourse prominence and pe-marking in
Romanian. International Review of Pragmatics, 2(2), 298-332.
Chiriacescu, S. I. (2014). The discourse structuring potential of indefinite noun phrases: special
markers in English, German and Romanian (doctoral dissertation). University of Stuttgart,
Germany.
Christianson, K., Hollingworth, A., Halliwell, J. F., & Ferreira, F. (2001). Thematic roles assigned
along the garden path linger. Cognitive psychology, 42(4), 368-407.
Chung, S., Ladusaw, W. A., and McCloskey, J. (1995). Sluicing and logical form. Nat. Lang.
Semantics 3, 239–282. doi: 10.1007/BF01248819
Cyrino, S. M. L. (1996). O objeto nulo do português brasileiro. DELTA: Documentação e Estudos
em Linguística Teórica e Aplicada, 12(2).
Cyrino, S., & Lopes, R. (2016). Null objects are ellipsis in Brazilian Portuguese. The Linguistic
Review, 33(4), 483-502.
218
Cole, P. (1987). Null objects in universal grammar. Linguistic inquiry, 18(4), 597-612.
Condoravdi, C., and J. M. Gawron (1996) “The Context-Dependency of Implicit Arguments,” in
M. Kanazawa, C. J. Pin ̃on, and H. de Swart, eds., Quantifiers, Deduction, and Context, CSLI,
Stanford, CA, 1–32.
Comrie, B. (1976). Aspect: An introduction to the study of verbal aspect and related
problems (Vol. 2). Cambridge university press.
Cornish, F., Garnham, A., Cowles, H. W., Fossard, M., & André, V. (2005). Indirect anaphora in
English and French: A cross-linguistic study of pronoun resolution. Journal of Memory and
Language, 52(3), 363-376.
Crawley, R. A., & Stevenson, R. J. (1990). Reference in single sentences and in texts. Journal of
Psycholinguistic Research, 19(3), 191-210.
Creswell, M. J. (1992). Logics and Languages. London: Methuen.
Davies, Mark. (2008-) The Corpus of Contemporary American English (COCA). Available online
at https://www.english-corpora.org/coca/.
Dickey, M. W., & Bunger, A. C. (2011). Comprehension of elided structure: Evidence from
sluicing. Language and Cognitive Processes, 26(1), 63-78.
Farkas, D. (1997). Towards a semantic typology of noun phrases. In Colloque de syntaxe et
sémantique de Paris, Université Paris (Vol. 7).
Ferrari, L. R. (2017). O processamento sintático em idosos: compreensão de sentenças ambíguas
e correlação com funções executivas. Dissertação (Mestrado), Pontifícia Universidade
Católica do Rio de Janeiro, Brasil.
Farrell, P. (1990). Null objects in Brazilian Portuguese. Natural language & linguistic
theory, 8(3), 325-346.
219
Ferreira, F., & Henderson, J. M. (1990). Use of verb information in syntactic parsing: Evidence
from eye movements and word-by-word self-paced reading. Journal of Experimental
Psychology: Learning, Memory, and Cognition, 16(4), 555.
Ferreira, F., & Henderson, J. M. (1991). Recovery from misanalyses of garden-path
sentences. Journal of Memory and Language, 30(6), 725-745.
Ferreira, M. B. (2000). Argumentos nulos em português brasileiro. Dissertação de doutorado,
Unicamp SP, Brasil.
Ferreira, M. (2009). Null subjects and finite control in Brazilian Portuguese. Minimalist essays on
Brazilian Portuguese syntax, 17-49.
Filik, R., Sanford, A. J., & Leuthold, H. (2008). Processing pronouns without antecedents:
Evidence from event-related brain potentials. Journal of Cognitive Neuroscience, 20(7), 1315-
1326.
Fillmore, C. J. (1986). Pragmatically controlled zero anaphora. In Proceedings of Berkeley
Linguistics Society 12, 95–107. Berkeley, Calif.: Berkeley Linguistics Society.
Frazier, L. (1978). On comprehending sentences: Syntactic parsing strategies. Doctoral
dissertation, University of Connecticut.
Frazier, L., & Rayner, K. (1982). Making and correcting errors during sentence comprehension:
Eye movements in the analysis of structurally ambiguous sentences. Cognitive
psychology, 14(2), 178-210.
Frazier, L., and Clifton, C. Jr. (1998). Comprehension of sluiced sentences. Lang. Cogn.
Process. 13, 499–520.
220
Frazier, L., and Clifton, C. Jr. (2011). D-linking and memory retrieval: the annoying case of
sluicing. University of Massachusetts Occasional Papers in Linguistics: Processing Linguistic
Structure, 38, 37-52.
Fukumura, K., & Van Gompel, R. P. (2010). Choosing anaphoric expressions: Do people take into
account likelihood of reference?. Journal of Memory and Language, 62(1), 52-66.
Gagnon, M. R. (2013). Anaphors and the missing link. Doctoral dissertation, University of
Maryland, MA.
Garrod, S., Freudenthal, D., & Boyle, E. (1994). The role of different types of anaphor in the on-
line resolution of sentences in a discourse. Journal of memory and language, 33(1), 39-68.
Gernsbacher, M. A., & Shroyer, S. (1989). The cataphoric use of the indefinite this in spoken
narratives. Memory & cognition, 17(5), 536-540.
Gerrig, R. J. (1986). Process models and pragmatics. Advances in cognitive science, 1, 23-42.
Gerrig, R. J., Horton, W. S., & Stent, A. (2011). Production and comprehension of unheralded
pronouns: A corpus analysis. Discourse Processes, 48(3), 161-182.
Gillon, B. (2006). English relational words, context sensitivity and implicit arguments. Ms.,
McGill University.
Ginzburg, Jonathan. 1996. Dynamics and the semantics of dialog. In J. Seligman, editor,
Language, Logic and Computation, Volume 1. CSLI, Stanford.
Givón, T. (1983). Topic continuity in discourse: A quantitative cross-language study (Vol. 3).
John Benjamins Publishing.
Glass, L. (2020). Verbs describing routines facilitate object omission in English. Proceedings of
the Linguistic Society of America, 5(1), 44-58.
221
Goldberg, A. E. (2001). Patient arguments of causative verbs can be omitted: The role of
information structure in argument distribution. Language sciences, 23(4-5), 503-524.
Gordon, P. C., Grosz, B. J., & Gilliom, L. A. (1993). Pronouns, names, and the centering of
attention in discourse. Cognitive science, 17(3), 311-347.
Grober, E. H., Beardsley, W., & Caramazza, A. (1978). Parallel function strategy in pronoun
assignment. Cognition, 6, 117–133.
Gundel, J. K., Hedberg, N., & Zacharski, R. (1993). Cognitive status and the form of referring
expressions in discourse. Language, 274-307.
Hagoort, P., Brown, C.M., Osterhout, L., 1999. The neurocognition of syntactic processing. In:
Brown, C.M., Hagoort, P. (Eds.), The Neurocognition of Language. Oxford Univ. Press,
Oxford, pp. 273–316.
Hankamer, J., & Sag, I. (1976). Deep and surface anaphora. Linguistic inquiry, 7(3), 391-428.
Harris, J. A. (2015). Structure modulates similarity-based interference in sluicing: An eye tracking
study. Frontiers in psychology, 6, 1839.
Harris, J. A. (2019). Alternatives on demand and locality: Resolving discourse-linked wh-phrases
in sluiced structures. In: Clifton, C., Fodor, J.D., & Carlson, K. Grammatical Approaches to
Language Processing - Essays in Honor of Lyn Frazier. Springer Studies in Theoretical
Linguistics.
Heim, I. (1982). The Semantics of Definite and Indefinite Noun Phrases. Doctoral dissertation,
University of Massachusetts, Amherst.
Hofmeister, P. (2011). Representational complexity and memory retrieval in language
comprehension. Language and cognitive processes, 26(3), 376-405.
222
Hopper, P. J., & Thompson, S. A. (1980). Transitivity in grammar and discourse. language, 251-
299.
Huang, C.-T. J. (1982) Logical Relations in Chinese and the Theory of Grammar, Doctoral
dissertation, MIT, Cambridge, Massachusetts
Ionin, T. (2006). This is definitely specific: specificity and definiteness in article systems. Natural
language semantics, 14(2), 175.
Kaan, E. (2001). Effects of NP type on the resolution of word-order ambiguities. Journal of
Psycholinguistic Research, 30(5), 529-547.
Kaiser, E. (2010). Investigating the consequences of focus on the production and comprehension
of referring expressions. International Review of Pragmatics, 2(2), 266-297.
Kaiser, E. (2011). Focusing on pronouns: Consequences of subjecthood, pronominalisation, and
contrastive focus. Language and Cognitive Processes, 26(10), 1625-1666.
Kaplan, D. (1989). Demonstratives: An Essay on the Semantics, Logic, Metaphysics, and
Epistemology of Demonstratives and Other Indexicals. In Joseph Almog, John Perry and
Howard K. Wettstein, eds., Themes from Kaplan, Oxford University Press, 481-563.
Kato, M. A. (1993). The distribution of pronouns and null elements in object position in Brazilian
Portuguese. In Linguistic Perspectives on Romance Languages (p. 225). John Benjamins.
Kehler, A., Kertz, L., Rohde, H., & Elman, J. L. (2008). Coherence and coreference
revisited. Journal of semantics, 25(1), 1-44.
Kjelgaard, M. M., Speer, S. R. (1999). Prosodic Facilitation and Interference in the Resolution of
Temporary Syntactic Closure Ambiguity. Journal of Memory and Language, 40(2), 153–194.
Koenig, J. P. (1999). ‘On a tué le président!’ Ultra-indefinites and the nature of passives. Cognition
and function in language, 235-251.
223
Koenig, J. P., & Mauner, G. (1999). A-definites and the discourse status of implicit
arguments. Journal of Semantics, 16(3), 207-236.
Koenig, J. P., Mauner, G., & Bienvenue, B. (2003). Arguments for adjuncts. Cognition, 89(2), 67-
103.
Järvikivi, J., Schimke, S., & Pyykkönen-Klauck, P. (2019). Understanding indirect reference in a
visual context. Discourse Processes, 56(2), 117-135.
Lewis, D. (1980). Index, context, and content. In Philosophy and grammar (pp. 79-100).
Springer, Dordrecht.
McKoon, G., & Ratcliff, R. (1992). Inference during reading. Psychological review, 99(3), 440.
Magliano, J. P., & Schleich, M. C. (2000). Verb aspect and situation models. Discourse
processes, 29(2), 83-112.
Martí, L. (2011). Implicit indefinite objects: grammar, not pragmatics. Unpublished ms., Queen
Mary University of London, UK.
Merchant, J. (2001). The Syntax of Silence: Sluicing, Islands, and the Theory of Ellipsis. Oxford:
Oxford University Press.
Mitchell, D. C., & Green, D. W. (1978). The effects of context and content on immediate
processing in reading. Quarterly Journal of Experimental Psychology, 30(4), 609-636.
Mittwoch, A. (1982). On the difference between eating and eating something: Activities versus
accomplishments. Linguistic Inquiry, 113-122.
Nagel, H. N., Shapiro, L. P., Tuller, B., & Nawy, R. (1996). Prosodic influences on the resolution
of temporary ambiguity during on-line sentence processing. Journal of Psycholinguistic
Research, 25(2), 319-344.
224
Nieuwland, M. S., & Van Berkum, J. J. (2006). Individual differences and contextual bias in
pronoun resolution: Evidence from ERPs. Brain Research, 1118(1), 155-167.
Omaki, A., & Schulz, B. (2011). Filler-gap dependencies and island constraints in second-
language sentence processing. Studies in Second Language Acquisition, 33(4), 563-588.
Patson, N. D., Darowski, E. S., Moon, N., & Ferreira, F. (2009). Lingering misinterpretations in
garden-path sentences: Evidence from a paraphrasing task. Journal of Experimental
Psychology: Learning, Memory, and Cognition, 35(1), 280.
Pesetsky, D. (1987). Wh-in-situ: Movement and unselective binding. In: E. Remand and G.B. ter
Meulen (eds.). The Representation of (in)definiteness: 98-129. Cambridge. MA: MIT Press.
Pesetsky, D. (2000). Phrasal movement and its kin. Cambridge, Mass.: MIT Press.
Prince, E. F. (1981). Topicalization, focus-movement, and Yiddish-movement: A pragmatic
differentiation. In Annual meeting of the berkeley linguistics society (Vol. 7, pp. 249-264).
Prince, E. F. (1992). The ZPG letter: Subjects, definiteness, and information-status. Discourse
description: diverse analyses of a fund raising text, 295-325.
R Core Team (2017) R: A Language and Environment for Statistical Computing. R Foundation
for Statistical Computing. Vienna, Austria.
Raposo, E. (1986). On the null object in European Portuguese. In Osvaldo Jaeggli and Carmen
Silva-Corvalan, eds., Studies in Romance Linguistics, 373-390. Dordrecht: Fori.
Ribeiro, A. J. C. (2004). Late Closure em parsing no portuguê s do Brasil. Tese de Doutorado,
Universidade Federal do Rio de Janeiro, Brasil.
Rizzi, L. (1986). Null objects in Italian and the theory of pro. Linguistic inquiry, 17(3), 501-557.
R Core Team (2019) R: A Language and Environment for Statistical Computing. R Foundation
for Statistical Computing. Vienna, Austria.
225
Roberts, C. (1998). Information structure in discourse: Towards and integrated formal theory of
pragmatics. In J. H. Yoon and Andreas Kathol, editors, OSU Working Papers in Linguistics
49: Papers in Semantics (1996). Ohio State University, pages 91–136. Revised version as of
Sept. 1998.
Rodrigues, C. (2002). “Morphology and null subjects in Brazilian Portuguese”. In D. Lightfoot
(ed.), Syntactic effects of morphological change, 160–178. Oxford & New York: Oxford
University Press.
Rodrigues, C. (2004). Impoverished Morphology and A-movement out of Case Domains. Ph.D.
thesis, University of Maryland.
Rohde, H., Kehler, A., & Elman, J. (2006). Event structure and discourse coherence biases in
pronoun interpretation. Presented at the Proceedings of the 28th Annual Conference of the
Cognitive Science Society, Vancouver, July 26–29, 2006.
Rohde, H., Kehler, A., & Elman, J. (2007). Pronoun interpretation as a side effect of discourse
coherence. Presented at the Proceedings of the 29th Annual Conference of the Cognitive
Science Society, Nashville, TN, August 1–4, 2007.
Ross, J. R. (1967). Constraints on Variables in Syntax. Ph.D. thesis, MIT, Cambridge, MA.
Ross, J. R. (1969). “Guess who,” in Proceedings of the 5th Annual Chicago Linguistic Society, eds
R. I. Binnick, A. Davison, G. M. Green, and J. L. Morgan (Chicago, IL), 252–286.
Rudanko, J. (2015). On a Class of Exceptions to Bach’s Generalization. In Linking Form and
Meaning: Studies on Selected Control Patterns in Recent English (pp. 89-97). Palgrave Pivot,
London
226
Sanford, A. J., Filik, R., Emmott, C., & Morrow, L. (2008). Short Article: They're Digging up the
Road Again: The Processing Cost of Institutional They. Quarterly Journal of Experimental
Psychology, 61(3), 372-380.
Schütze, C. T., & Sprouse, J. (2014). Judgment data. Research methods in linguistics, 27.
Slattery, T. J., Sturt, P., Christianson, K., Yoshida, M., & Ferreira, F. (2013). Lingering
misinterpretations of garden path sentences arise from competing syntactic
representations. Journal of Memory and Language, 69(2), 104-120.
Snyder, E. (2013). Binding, genericity, and predicates of personal taste. Inquiry, 56(2-3), 278-306.
Stalnaker, R. C. (1972). Pragmatics. In Donald Davidson and Gilbert Harman, eds., Semantics of
Natural Language, 380-397. Dordrecht: Reidel.
Tomlin, R. S. (1985). Foreground-background information and the syntax of subordination. Text-
Interdisciplinary Journal for the Study of Discourse, 5(1-2), 85-122.
Trabasso, T., & Van Den Broek, P. (1985). Causal thinking and the representation of narrative
events. Journal of memory and language, 24(5), 612-630.
Trueswell, J. C., Tanenhaus, M. K., & Kello, C. (1993). Verb-specific constraints in sentence
processing: Separating effects of lexical preference from garden-paths. Journal of
Experimental Psychology: Learning, Memory, and Cognition, 19(3), 528–553.
Van Berkum, J. J. (2004). Sentence comprehension in a wider discourse: Can we use ERPs to keep
track of things. The on-line study of sentence comprehension: Eyetracking, ERPs and beyond,
229-270.
Ward, G., & Birner, B. (1995). Definiteness and the English existential. Language, 722-742.
Williams, A. (2015). Arguments in syntax and semantics. Cambridge University Press.

227
Appendix 1

Chapter 2, Experiment 1: 20 target items (20 Optionally Transitive verbs)

Bob handed Victoria several books to read.
Later, he saw her reading (a book), but he doesn't know (which book/which one).
Andrew handed Lucy several plates to clean.
Later, he saw her cleaning (a plate), but he's not sure (which plate/which one).
Mary sent William several melodies to play.
That night, she saw him playing (a melody), but she's not certain (which melody/which one).
Ellen gave Justin a few manuals to study.
That evening, she saw him studying (a manual), but she doesn't remember (which
manual/which one).
Brandon handed Catherine some fruits to eat.
Later, he saw her eating (a fruit), but he's not certain (which fruit/which one).
Peter gave Brenda some drink options to order.
Later, he saw her ordering (a drink), but he's not certain (which drink/which one).
Lisa gave Gabriel a few choreographies to perform.
That night, she saw him performing (a choreography), but she doesn't know (which
choreography/which one.)
228
Ann handed Bertie a few magazines to color.
That evening, she saw him coloring (a magazine), but she's not certain (which magazine/which
one).
John gave Marge a few cars to drive.
Later, he saw her driving (a car), but he doesn't remember (which car/which one).
Steven gave Phoebe several jingles to sing.
The next day, he saw her singing (a jingle), but he doesn't remember (which jingle/which one).
Rachel gave Albert a few essays to write.
That night, she saw him writing (an essay), but she doesn't know (which essay/which one).
Louis gave Vanessa some emails to type.
That evening, he saw her typing (an email), but he doesn't remember (which email/which one).
Maggie gave Jasper several landscapes to paint.
The next day, she saw him painting (a landscape), but she doesn't remember (which landscape/
which one).
Charlie sent Caitlin a few designs to sketch.
The next day, he saw her sketching (a design), but he's not sure (which design/which one).
Britney handed Mikey some toys to draw.
That night, she saw him drawing (a toy), but she's not sure (which toy/which one).
Karen gave Frank several figures to sculpt.
That evening, she saw him sculpting (a figure), but she doesn't know (which figure/which one).
229
Lucas handed Betty some glove patterns to knit.
The next day, he saw her knitting (a glove), but he's not sure (which glove/which one).
Carla sent Christopher some scarf patterns to crochet.
The next day, she saw him crocheting (a scarf), but she's not sure (which scarf/which one).
Bill gave Lauren several steak recipes to cook.
That night, he saw her cooking (a steak), but he doesn't know (which steak/which one).
Kate sent Kevin a few pie recipes to bake.
The next day, she saw him baking (a pie), but she's not certain (which pie/which one).

230
Appendix 2

Chapter 2, Experiments 1 and 2: 20 intransitive verbs used in ungrammatical control conditions

Sleeping
Sitting
Sneezing
Standing
Smiling
Crying
Laughing
Falling
Jogging
Swimming
Vomiting
Running
Resting
Walking
Giggling
Coughing
Sobbing
Screaming
Waiting
Relaxing
231
Appendix 3

Chapter 2, Experiment 2: Set Complexity Manipulation for 20 target items

Simple set - Complex set

Several books - Several rare old books
Several plates - Several dirty porcelain plates
Several melodies - Several original folk melodies
A few manuals – A few recent technical manuals
Some fruits – Some fresh tropical fruits
Some drink options – Some sweet alcoholic drink options
A few choreographies – A few new postmodern choreographies
A few magazines – A few worn out popular magazines
A few cars – A few Japanese electric cars
Several songs – Several British pop songs
A few essays – A few analytical academic essays
Some emails – Some private internal emails
Several landscapes – Several American urban landscapes
A few designs – A few black geometric designs
Some toys – Some striped plastic toys
Several figures – Several artistic mythological figures
Some glove patterns – Some colorful fingerless glove patterns
Some scarf patterns – Some plaid woolen scarf patterns
232
Several steak recipes – Several traditional French steak recipes
A few pie recipes – A few savory Southern pie recipes

233
Appendix 4

Chapter 2, Experiment 2: Verb + Noun phrases used in secondary task control items

Reading a newspaper
Cleaning a fork
Playing an anthem
Studying a textbook
Eating a cookie
Ordering a dessert
Performing a monologue
Coloring a napkin
Driving a truck
Singing a jingle
Writing a letter
Typing a memo
Painting a horse
Sketching a portrait
Drawing a lake
Sculpting an ashtray
Knitting a hat
Crocheting a sweater
Cooking a turkey
Baking a cake
234
Appendix 5

Chapter 4 – Target items

Note: first parentheses contain antecedent manipulation (without/with antecedent); second
parentheses contain verb type manipulation (OPT verb/OT verb/Intransitive control verb)

1.
A Ana (acordou/pegou um bife), (cozinhou/fritou/refletiu) durante uns dez minutos, depois
devorou rapidinho e foi trabalhar.

Translation: Ana (woke up/grabbed a steak), (cooked/fried/pondered) for about ten minutes, then
devoured quickly and went to work.

2.
O Joao (acordou/foi até um portão), (pulou/fechou/saiu) em segundos com facilidade, depois
abriu para os outros entrarem.

Joao (woke up/went up to a gate), (jumped/closed/left) in seconds with ease, then opened so
others could come in.

3.
A Clara (entrou/escolheu um livro), (leu/folheou/deitou) no sofá da sala, depois guardou na
estante e saiu.
235

Clara (entered/selected a book), (read/flipped-through/laid down) on the living room couch, then
put on the shelf and left.

4.
O Pedro (sentou/pensou num poema), (escreveu/digitou/levantou) com calma e concentração,
depois enviou para a sua namorada.

Pedro (sat down/thought of a poem), (wrote/typed/got up) calmly and with focus, then sent to his
girlfriend.

5.
A Luisa (chegou/preparou uma tela), (pintou/emoldurou/entrou) no estúdio de arte, depois
embrulhou de presente para uma amiga.

Luisa (arrived/prepared a canvas), (painted/framed/went in) at the art studio, then wrapped as a
gift for a friend.

6.
O Bruno (saiu/preparou um carro), (dirigiu/lavou/passeou) por umas duas horas, depois deixou
na garagem do prédio.

236
Bruno (left/prepared a car), (drove/washed/strolled) for about two hours, then left in the
building’s garage.

7.
A Maria (levantou/recebeu uma conta), (pagou/parcelou/chorou) no caixa da loja, depois jogou
no lixo da saida.

Maria (got up/received a bill), (paid/split-into-parcels/cried) at the store’s cashier, then threw in
the garbage by the exit.

8.
O Carlos (chegou/pegou um violao), (tocou/afinou/acenou) para o seu professor, depois botou no
armário da escola.

Carlos (arrived/picked up a guitar), (played/tuned/waved) to the teacher, then put in school’s
closet.

9.
A Nina (saiu/pediu um suco), (bebeu/provou/sorriu) bem devagar com gosto, depois derramou
no chão do restaurante.

Nina (left/ordered a juice), (drank/tasted/smiled) slowly and with gusto, then spilled on the
restaurant floor.
237

10.
O Thiago (chegou/fez um gol), (comemorou/repetiu/correu) rápido logo em seguida, depois
relembrou com os amigos do bar.

Thiago (arrived/scored a goal), (celebrated/repeated/ran) fast right after, then thought about with
the friends at the bar.

11.
A Julia (chegou/buscou uma camisa), (costurou/remendou/dormiu) no quarto de noite, depois
entregou para o seu marido.

Julia (arrived/picked up a shirt), (sewed/patched/slept) in the bedroom at night, then delivered to
her husband.

12.
O Luis (levantou/avistou um monumento), (fotografou/admirou/descansou) da/na varanda do
hotel, depois visitou cedo no dia seguinte.

Luis (got up/saw a monument), (photographed/admired/rested) from/at the hotel balcony, then
visited early the following day.

13.
238
A Tais (saiu/comprou um girassol), (plantou/regou/acordou) de manhã bem cedo, depois colocou
na janela da sala.

Tais (left/bought a sunflower), (planted/watered/woke up) very early in the morning, then put in
living room window.

14.
O Marcio (entrou/foi em um escritorio), (varreu/decorou/brincou) por um bom tempo, depois
arrumou com cuidado e saiu.

Marcio (entered/went in an office), (swept/decorated/played) for a good while, then put together
with care and left.

15.
A Dani (saiu/ganhou uma moto), (estacionou/lavou/escorregou) na garagem do prédio, depois
vendeu para o primo de um amigo.

Dani (left/got a motorcycle), (parked/washed/slipped) in the building garage, then sold to a
cousin of a friend.

16.
O Jose (levantou/mordeu um brownie), (comeu/mastigou/sentou) na mesa de jantar, depois
cuspiu na pia da cozinha.
239

Jose (got up/bit a brownie), (ate/chewed/sat) in the dinner table, then spit on kitchen sink.

17.
A Marta (entrou/achou um relatorio), (estudou/analisou/meditou) num canto da biblioteca,
depois revelou para os outros jornalistas.

Marta (entered/found a report), (studied/analyzed/meditated) in a corner at the library, then
revealed to the other journalists.

18.
O Andre (sentou/desdobrou uma toalha), (bordou/examinou/relaxou) por uns cinco minutos,
depois pendurou no gancho do banheiro.

Andre (sat/unfolded a towel), (embroidered/examined/relaxed) for about five minutes, then hung
in the bathroom hook.

19.
A Bia (levantou/recebeu uma coreografia), (ensaiou/memorizou/trabalhou) na escola de dança,
depois apresentou no festival de artes e dança.

Bia (got up/received a choreography), (rehearsed/memorized/worked) at the dance school, then
presented at the art and dance festival.
240

20.
O Lucas (acordou/imaginou um vestido), (desenhou/esbocou/cochilou) no caminho pra
faculdade, depois descreveu para todos os professores.

Lucas (woke up/imagine a dress), (drew/sketched/napped) on the way to college, then described
to all the professors.

241
Appendix 6

Chapter 4 - Words outside the critical region

For the self-paced reading analysis presented in Chapter 4, words not in the critical region
were also analyzed in a parallel way to words in the critical region.
At words before the critical region, there are two noteworthy findings:
Firstly, at the very first word, the word was read faster in conditions with OPT verbs
relative to those with OT verbs (significant main effect of verb type; see Tables 1, 2). This seems
to be a spurious effect, because at the first word there are no differences between conditions (which
have been latin squared across lists). Spurious effects occur sometimes in self-paced reading tasks
(e.g., see Ferreira & Henderson 1990, Kaan 2001, Omaki 2011). Crucially, this effect appears
before the manipulations and before the critical region, and is not related to the research question
and predictions for this experiment.
The second finding occurs at the first comma (sixth word): the comma was read slower in
conditions that had an antecedent relative to those that did not (significant main effect of
antecedent; see Tables 1, 2). That is, the comma was read slower when it appeared after “Clara
picked a book” relative to when it appeared after “Clara entered”. This effect is not completely
surprising, since it appears immediately after a region in which the conditions are considerably
different. For reasons that are not entirely clear, it may be that a comma is more expected after an
intransitive than after a transitive predicate; alternatively, this could be another spurious effect.
Crucially, this effect at the comma precedes the critical region and is not directly related to our
predictions. It is also worth noticing that this effect goes in the opposite direction than the effect
that was predicted and occurs in the critical region: at this comma position, conditions with an
242
antecedent are read slower, whereas two words later, in the critical region, it is the OT condition
without an antecedent that is read slower (as predicted). That is, the effect seen in the critical region
should not be considered an ‘extension’ of what is happening at the comma.
At all other words occurring before the critical region, there are no significant differences
in reading times between conditions.
At words that appear after the critical region, there are various instances indicating
that difficulties caused by comprehension of anaphoric implicit objects (of OT verbs) without
an antecedent available persist beyond the critical region.
In particular, at the third word following the critical region, reading times appear as
predicted: conditions with OT verbs without antecedents are read significantly slower than those
with antecedents. Moreover, at the second and fifth words following the critical regions, there are
marginal effects in the expected direction: conditions with OT verbs without antecedents are again
read slower than those with antecedents. And at the first and fourth word following the critical
region, no significant effects appear but numerically OT conditions without antecedents are still
read slower (see Tables 1, 2 and 3 of this Appendix).

Table 1. Mean RTs in milliseconds and SE for non-critical region words by verb type and antecedent type
Word OPT + no antec

OPT + with antec OT + no antec OT + with antec
Mean RT SE Mean RT SE Mean RT SE Mean RT SE
1 381.475 10.43 367.794 9.719 353.283 8.408 362.247 9.454
2 369.64 10.276 367.213 10.159 360.625 9.895 360.491 10.087
243
3 396.701 13.276 392.589 13.387 399.085 13.152 386.483 13.586
4 - - 406.156 12.757 - - 382.195 11.708
5 - - 464.948 18.515 - - 448.498 18.814
6 419.699 13.371 437.94 13.815 399.808 11.798 427.518 12.983
Critical Region (see Table 2)
11 428.025 16.396 417.672 14.974 452.767 17.715 420.391 14.802
12 427.98 14.242 411.226 11.492 459.983 15.251 433.067 14.226
13 400.693 11.368 381.914 10.582 413.751 11.657 379.975 10.24
14 371.984 11.723 369.267 11.63 400.964 14.632 377.817 12.534
15 377.168 11.028 366.376 9.859 376.859 10.239 358.937 10.223
16 376.934 10.154 376.487 10.382 367.761 8.311 369.531 9.975
17 387.98 9.277 377.137 8.773 382.715 8.522 369.137 8.656

Table 2. Reading Time Results for non-critical region words, “No Antecedent” inquiry (“*”: p < 0.05)
Word Main Effects Interaction
Verb Type Antecedent Verb Type*Antecedent
b SE t p b SE t p b SE t p
1 -8.450 4.078 -2.072 0.044* -1.207 3.648 -0.331 0.740 5.722 3.646 1.569 0.117
2
-3.926 3.535 -1.110 0.267 -0.666 3.537 -.189 0.851 0.562 3.536 0.159 0.874
244
3 -0.917 3.535 -0.195 0.845 -4.209 4.704 -0.895 0.371 -2.113 4.703 -0.449 0.653
4 -12.322 6.535 1.886 0.06 - - - - - - - -
5 -7.683 11.181 0.687 0.496 - - - - - - - -
6 -7.599 6.335 1.199 0.236 11.477 5.305 2.163 0.03* 2.290 5.303 .432 .666
Critical Region (see Table 3)
11 6.863 6.545 1.049 0.294 -10.969 6.555 -1.673 0.094 -5.994 6.550 -0.915 0.36
12 13.408 5.407 2.480 0.013* -10.804 5.416 -1.995 0.046* -2.888 5.411 -0.534 0.593
13 2.768 3.933 -3.32 0.481 -10.969 3.939 -1.673 0.001* -3.643 3.936 -0.926 0.354
14
9.398 4.819 1.950 0.051 6.501 4.821 1.349 0.178 -5.119 4.82 1.062 0.288
15 -1.822 3.909 -0.466 0.641 -7.247 3.916 -1.85 0.064 -1.852 3.913 -0.473 0.636
16 -4.207 3.602 -1.168 0.243 0.191 3.606 0.053 0.958 0.355 3.604 0.099 0.921
17 -3.289 3.362 -0.978 0.328 -6.483 3.367 -1.925 0.054 -0.863 3.365 -0.257 0.797

Table 3. Planned Comparisons for non-critical region words, “No Antecedent” inquiry (“*”: p < 0.05)
Word Contrasts Results
b SE t p
12 OPT+Antec vs. OPT-Antec -8.274 7.336 -1.128 0.26
OT+Antec vs. OT-Antec -13.23 7.97 -1.66 0.098
13 OPT+Antec vs. OPT-Antec -9.454 5.835 -1.62 0.106
245
OT+Antec vs. OT-Antec -16.806 5.442 -3.088 0.002*

246
Appendix 7

Chapter 5: Target items

Note: First parentheses contain antecedent manipulation (content between parenthesis either
present or absent); second parentheses contain comma manipulation (either with comma or
without); third parentheses contain definiteness manipulation (indefinite or definite)

1.
Depois que eu (andei até um muro e) pulei(,) (uma/a) cerca desabou em cima das flores e
arbustos.

Translation: After I (walked to a wall and) jumped(,) (a/the) fence fell on top of the flowers and
bushes.

2.
Depois que eu (recebi uma carta e) li(,) (um/o) livro caiu nos dedos do meu pé esquerdo.

After I (received a letter and) read(,) (a/the) book fell on the toes of my left foot.

3.
Depois que eu (planejei uma redação e) escrevi(,) (um/o) bilhete voou da minha mesa ate a
janela.

247
After I (planned an essay and) wrote(,) (a/the) note flew from my table to the window.

4.
Depois que eu (preparei uma tela e) pintei(,) (um/o) móvel ficou manchado de tinta verde e azul.

After I (prepared a canvas and) painted(,) (a/the) piece of furniture was stained with blue and
green paint.

5.
Depois que eu (entrei em um carro e) dirigi(,) (uma/a) van deslizou em uma vala na estrada.

After I (got in a car and) drove(,) (a/the) van slid in a ditch on the road.

6.
Depois que eu (achei um boleto e) paguei(,) (uma/a) conta apareceu na mesa do escritório.

After I (found a bill and) paid(,) (a/the) bill appeared on the office desk.

7.
Depois que eu (selecionei uma flauta e) toquei(,) (um/o) violão caiu da estante da escola de
música.

After I (selected a flute and) played(,) (a/the) guitar fell from the self of the music school.
248

8.
Depois que eu (comprei uma água e) bebi(,) (um/o) suco vazou dentro da minha bolsa de couro.

After I (bought a water and) drank(,) (a/the) juice leaked inside my leather purse.

9.
Depois que eu (ganhei uma coxinha e) comi(,) (um/o) pastel ficou na geladeira mais de duas
semanas.

After I (was gifted a pastry and) ate(,) (a/the) other-pastry remained in the fridge for over two
weeks.

10.
Depois que eu (assisti um gol e) comemorei(,) (uma/a) falta aconteceu perto do meio de campo.

After I (watched a goal and) celebrated(,) (a/the) penalty happened near the middle of the field.

11.
Depois que eu (achei um lençol e) costurei(,) (uma/a) camisa rasgou dentro da maquina de lavar.

After I (found some sheets and) sew(,) (a/the) shirt ripped inside the washing machine.

249
12.
Depois que eu (vi um foguete e) fotografei(,) (uma/a) estátua desabou na entrada do museu de
ciência.

After I (saw a rocket and) photographed(,) (a/the) statue fell at the science museum's entrance.

13.
Depois que eu (entrei em uma combi e) estacionei(,) (um/o) trator desceu ladeira abaixo e bateu
num poste.

After I (got into a van and) parked(,) (a/the) tractor went-down the hill and hit a lampost.

14.
Depois que eu (fui em um banheiro e) varri(,) (uma/a) sala ficou cheia de poeira no chão.

After I (went to a bathroom and) swept(,) (a/the) living room became full of dust on the floor.

15.
Depois que eu (lavei uma batata e) cozinhei(,) (um/o) salmão esquentou no microondas da
cozinha.

After I (washed a potato and) cooked(,) (a/the) salmon heated up in the kitchen microwave.
16.
250
Depois que eu (comprei um girassol e) plantei(,) (uma/a) rosa nasceu no quintal dos fundos.

After I (bought a sunflower and) planted(,) (a/the) rose sprouted in the backyard.

17.
Depois que eu (fui até uma grade e) pulei(,) (um/o) portão abriu e ficou aberto por horas.

After I (went to a fence and) jumped(,) (a/the) gate opened and remained opened for hours.

18.
Depois que eu (ganhei um convite e) li(,) (uma/a) revista escorregou do sofá pro chão do
consultório.

After I (was given an invite and) read(,) (a/the) magazine slipped from the couch to the floor of
the doctor's office.

19.
Depois que eu (pensei em um poema e) escrevi(,) (uma/a) crônica foi publicada com o mesmo
título.

After I (thought about a poem and) wrote(,) (a/the) essay-type was published with the same title.

20.
251
Depois que eu (descolori um cabelo e) pintei(,) (uma/a) unha acabou borrada com tinta vermelha.

After I (removed-color of a hair and) painted(,) (a/the) nail ended up stained with red paint.

21.
Depois que eu (entrei em uma ambulância e) dirigi(,) (um/o) caminhão bateu em vários carros na
rua.

After I (got into an ambulance and) drove(,) (a/the) truck hit on several cars on the street.

22.
Depois que eu (recebi uma multa e) paguei(,) (um/o) empréstimo foi oferecido pelo meu banco.

After I (received a fine and) paid(,) (a/the) loan was offered by my bank.

23.
Depois que eu (escolhi um funk e) toquei(,) (uma/a) sinfonia começou no rádio do carro.

After I (chose a funk song and) played(,) (a/the) symphony started on the car radio.
24.
Depois que eu (pedi um refrigerante e) bebi(,) (uma/a) cerveja apareceu bem na minha frente.

After I (ordered a soda and) drank(,) (a/the) beer appeared right in front of me.
252

25.
Depois que eu (fiz um sanduíche e) comi(,) (uma/a) pizza pareceu uma boa idéia também.

After I (made a sandwich and) ate(,) (a/the) pizza seemed like a good idea as well.

26.
Depois que eu (descobri uma conquista e) comemorei(,) (um/o) campeonato começou em uma
cidade vizinha.

After I (found out about an accomplishment and) celebrated(,) (a/the) championship started in a
neighboring town.

27.
Depois que eu (peguei uma meia e) costurei(,) (um/o) casaco apareceu na pilha de roupas velhas.

After I (grabbed a sock and) sew(,) (a/the) sweater appeared on the old clothes pile.

28.
Depois que eu (avistei uma águia e) fotografei(,) (um/o) pombo pousou na frente da câmera.

After I (saw an eagle and) photographed(,) (a/the) pigeon landed in front of the camera.

253
29.
Depois que eu (manobrei um ônibus e) estacionei(,) (uma/a) limosine parou na vaga ao lado da
minha.

After I (maneuvered a bus and) parked(,) (a/the) limo stopped on the spot next to mine.

30.
Depois que eu (entrei em uma varanda e) varri(,) (um/o) quarto ficou mais sujo do que antes.

After I (entered a balcony and) swept(,) (a/the) bedroom became dirtier than before.

31.
Depois que eu (preparei um perú e) cozinhei(,) (uma/a) abóbora queimou no forno do meu
vizinho.

After I (prepared a turkey and) cooked(,) (a/the) pumpkin burned in my neighbor's stove.

32.
Depois que eu (ganhei uma tulipa e) plantei(,) (um/o) cacto morreu no jardim da casa.

After I (was given a tulip and) planted(,) (a/the) cactus died in the garden of the house.

254
Appendix 8

Chapter 5: Words outside the critical region

For the self-paced reading analysis presented in Chapter 5, words not in the critical region
were also analyzed in a parallel way to words in the critical region.
In these positions, there are three noteworthy findings:
At the eighth and ninth words (e.g., the verb “li” i.e. “read” and the following comma in
unambiguous conditions), sentences with antecedents were read slower than those without
antecedents. These words in the sentence immediately follow the region that introduces the
antecedent. For example, the words following “After I picked a book and…” were read slower than
those following “After I…”. There are a few possible explanations for this difference in reading
times: The first is that sentences with antecedents are read slower at these positions because the
antecedent is being processed as an anaphoric implicit object, which could be costlier/more
burdensome than processing an existential implicit object in the antecedentless sentences.
Although attractive, this explanation does not match findings from the experiment in Chapter 3,
which showed no differences in reading times between optionally transitive verbs that had
antecedents and those that didn’t (i.e., those with anaphoric implicit objects and those with
existential implicit objects). However, the sentences used in the experiments in Chapter 2 and 3
differ slightly, so it is not completely unreasonable that this effect would appear here but not in the
previous study.
Alternatively, this difference between antecedent and antecedentless conditions at the eight
and ninth words could be a spurious effect. In any case, this effect precedes the critical region and
does not directly speak to the main purposes of this experiment.
255
At the tenth and eleventh words (e.g., the temporarily ambiguous noun phrase “um/o livro”
i.e. “a/the book”), sentences with indefinites (e.g. “a book”) were read slower than those with
definites (e.g. “the book”) - but only in unambiguous sentences, those that had a comma between
the preceding verb and this noun phrase. At the tenth word, this difference appears only in
unambiguous sentences with antecedents. At the eleventh word, this difference appears in both
unambiguous conditions (with and without antecedents). These patterns suggest that in
unambiguous conditions, in which the noun phrase is being treated as a subject, people are already
evaluating whether the noun phrase is coreferential with the previous implicit object: definites,
which can be coreferential, are read faster. Indefinites, which cannot corefer with the previous
implicit object and therefore are introducing a new referent, are read relatively slower.
At the eleventh word (e.g., “livro” i.e. “book”), ambiguous sentences are read slower than
unambiguous ones. This occurs before the disambiguating verb appears. This difference in reading
times may be due to differences in processing distinct grammatical positions: in unambiguous
conditions, the noun phrase is treated as the subject of a new sentence (i.e., the beginning of a new
structure). In ambiguous conditions, the noun phrase is - at that moment - treated as the object of
the current sentence (i.e., it is being incorporated into the structure currently being built).
At all other words not in the critical region, there are no significant differences in reading
times between conditions.

256
Table 1. Mean RTs and SE for non-critical region words by antecedent and definiteness of noun phrase for
UNAMBIGUOUS (with comma) conditions
Word No antecedent With antecedent
indefinite definite indefinite definite
Mean RT SE Mean RT SE Mean RT SE Mean RT SE
1 474.129 13.700 476.535 13.464 467.507 12.473 479.317 14.940
2 419.773 10.416 400.880 9.440 402.214 9.547 406.093 10.422
3 398.901 10.186 401.518 10.586 410.385 11.539 400.171 11.143
4 - - - - 472.114 15.572 441.324 14.340
5 - - - - 460.329 14.232 443.085 12.706
6 - - - - 538.294 20.187 521.099 20.030
7 - - - - 506.712 14.711 493.194 13.940
8 496.368 17.702 488.211 15.823 522.776 18.672 521.307 18.763
9 493.801 13.400 483.283 13.250 510.961 17.696 509.533 16.836
10 443.600 12.423 420.794 9.993 490.543 15.113 442.740 10.937
11 465.257 16.912 431.229 13.847 479.411 18.636 437.792 12.110

257
Table 2. Mean RTs and SE for non-critical region words by antecedent and definiteness of noun phrase for
AMBIGUOUS (no comma) conditions
Word No antecedent With antecedent
indefinite definite indefinite definite
Mean RT SE Mean RT SE Mean RT SE Mean RT SE
1 473.256 13.759 459.414 12.359 484.147 13.529 462.397 12.872
2 415.447 10.688 404.521 9.442 407.843 9.827 406.938 11.559
3 400.654 10.119 397.012 9.720 396.683 10.207 391.942 10.144
4 - - - - 445.709 12.800 449.284 13.351
5 - - - - 441.815 13.470 444.535 12.992
6 - - - - 542.382 22.676 534.911 21.094
7 - - - - 476.355 11.711 501.728 15.445
8 459.413 15.359 477.520 14.727 478.969 13.727 488.283 17.118
9 - - - - - -
- -
10 460.669 12.834 439.589 10.836 449.740 12.298 441.281 11.636
11 518.320 19.930 494.975 18.218 536.010 22.523 515.184 21.605

258
Table 3. Reading Time Results: Main Effects for non-critical region words (“*”: p < 0.05)
Word Main Effects
Ambiguity Antecedent Definiteness
b SE t p b SE t p b SE t p
1
-2.299 3.823 -0.602 0.548 1.150 3.823 0.301 0.764 -2.591 3.823 -0.678 0.498
2
-1.460 5.546 -0.263 0.792 -4.676 5.681 -0.823 0.411 -2.008 5.474 -0.367 0.714
3
-6.676 5.462 -1.222 0.222 -0.891 5.601 -0.159 0.874 -2.279 5.386 -0.423 0.672
4
11.889 9.937 1.196 0.234 - - - - -2.171 14.701 -0.148 0.884
5
6.502 13.012 0.500 0.620 - - - - 6.456 12.691 0.509 0.614
6
13.71 14.36 0.955 0.342 - - - - 14.40 14.15 1.018 0.311
7
25.375 12.906 1.966 0.051 - - - - 5.774 12.820 0.450 0.653
8
-10.398 7.204 -1.443 0.149 23.988 10.815 2.389 0.036* 6.220 7.109 0.875 0.382
9
- - - - 20.186 7.498 2.692 0.007* -3.999 7.498 -0.533 0.593
10
-0.807 3.465 -0.233 0.815 7.437 3.465 2.146 0.032* -12.592 3.465 -3.634 0.001*
11
31.648 7.925 3.993 0.001* 6.931 4.811 1.441 0.149 -14.727 4.811 -3.061 0.002*

Table 4. Reading Time Results: 2 way interactions for non-critical region words (“*”: p < 0.05)
Word Interactions
Ambiguity*Antecedent Ambiguity*Definiteness Antecedent*Definiteness
b SE t p b SE t p b SE t p
259
1 2.290 3.823 0.599 0.549 -6.211 3.823 -1.625 0.104 0.093 3.823 0.024 0.980
2 -4.965 5.464 -0.909 0.364 -4.122 5.496 -0.750 0.454 6.919 5.490 1.260 0.208
3 4.327 5.378 0.805 0.421 -2.992 5.411 -0.553 0.580 3.350 5.406 -0.620 0.536
4 - - - - 9.321 10.026 0.930 0.355 - - - -
5 - - - - 22.592 11.820 1.911 0.058 - - - -
6 - - - - 10.82 14.51 0.746 0.457 - - - -
7 - - - - 11.522 12.928 0.891 0.374 - - - -
8 3.439 7.135 0.482 0.630 -2.646 7.163 -0.399 0.690 0.950 7.134 0.133 0.894
9 - - - - - - - - -1.947 7.498 -0.260 0.795
10 -9.743 3.465 -2.812 0.004* 5.170 3.465 1.492 0.135 -1.488 3.465 -0.430 0.667
11 1.893 4.811 0.394 0.693 4.306 4.811 -0.168 0.370 -0.810 4.811 -0.430 0.866

Table 6. Reading Time Results: 3 way interaction for non-critical region words (“*”: p < 0.05)
Word Interaction
Ambiguity*Antecedent*Definiteness
b SE t p
1
-2.243 3.823 -0.587 0.557
2
-0.641 5.539 -0.116 0.908
3
-2.023 5.460 -0.371 0.711
4
- - - -
260
5
- - - -
6
- - - -
7
- - - -
8
-3.428 7.209 -0.476 0.634
9
- - - -
10
4.705 3.465 1.358 0.174
11
0.983 4.811 0.204 0.838

Table 7. Planned Comparisons for for non-critical region words (“*”: p < 0.05)
Word Con-
trasts

Antecedent Definiteness Antecedent*Definiteness
b SE t p b SE t p b SE t p
10 Ambig.
only

2.399 8.942 0.268 0.789 -2.430 8.586 -0.283 0.777 6.465 8.543 0.757 0.45
0
Unam.
only
17.201 4.773 3.604 0.001* -17.701 4.773 -3.709 0.001
*
-6.271 4.773 -1.314 0.18
9
Unam.
-Antec
only
- -11.505 6.088 -1.890 0.059 -
Unam.
+Antec
only
- -23.902 7.479 -3.196 0.001
*
-
261
11 Ambig.
only

17.213 12.807 1.344 0.180 -5.545 12.236 -0.453 0.650 21.86
5
12.161 1.798 0.07
3*
Unam.
only
-0.404 12.376 -0.033 0.973 -29.630 12.203 -2.428 0.016
*
-5.977 12.443 -0.480 0.63
1

Abstract (if available)

Abstract This dissertation aims to expand our knowledge on implicit linguistic phenomena, their discourse properties and how they participate in comprehension processes. In particular, this dissertation looks at the underexplored domain of direct objects that are introduced into the discourse implicitly, such as what Ana was writing in the sentence “Ana was writing.” Chapters 2 and 3 focus on the discourse status of existential implicit objects (such as the implicit object of ‘writing’ introduced above), which have been claimed to be “discourse inert”. Chapter 2 presents two acceptability judgement tasks and one forced-choice sentence completion task using sluicing constructions. Findings revealed that existential implicit objects are less accessible to reduced expressions, but they can be as discourse persistent as overt objects. Chapter 3 expands on this line of inquiry with a story continuation task, which reveals that although relatively less accessible and persistent, existential implicit objects do participate in the structuring of subsequent discourse and they are sometimes pronominalized (i.e., they are not completely discourse inert). Crucially, Chapter 3 also reveals that a prior context can affect implicit objects’ discourse persistence, but not their accessibility. Chapter 4 investigated how the context-dependency properties of implicit objects affects their comprehension. A self-paced reading task in Brazilian Portuguese (a language with both context-dependent and -independent implicit objects, i.e., anaphoric/existential) shows that the language processor is able to quickly identify which types of implicit objects are allowed by each verb. Moreover, this task shows that the comprehension of anaphoric implicit objects with no available antecedents in context leads to processing difficulties, however no difficulties occur when there is at least an implicit antecedent available. In contrast, the comprehension of existential implicit objects does not suffer in the absence of antecedents. Chapter 5 investigated how the interpretation of implicit objects in sentences with temporary subject/object ambiguities (garden-paths) affects sentence re-analysis and the occurrence of lingering misinterpretations. A self-paced reading task in Brazilian Portuguese shows that an anaphoric interpretation of the implicit object in garden path reanalysis leads to a large reduction in the proportion of lingering misinterpretations. In sum, this dissertation shows that existential implicit objects are not entirely “discourse inert” and do participate in people’s mental models of the discourse. Moreover, investigations on the comprehension of context-dependent and -independent implicit objects show that the language processor is able to quickly recruit information from various domains (e.g., verb argument structure, prior context, general syntactic parsing biases) when dealing with these implicit elements.

Linked assets

University of Southern California Dissertations and Theses

Conceptually similar

PDF

Discourse level processing and pronoun interpretation

PDF

Processing the dynamicity of events in language

PDF

Discourse-level processing of nominal possessive constructions

PDF

Vietnamese pronouns in discourse

PDF

Exploring the effects of Korean subject marking and action verbs’ repetition frequency: how they influence the discourse and the memory representations of entities and events

Asset Metadata

Creator Piani Besserman Vianna, Ana Elisa (author)

Core Title When things are left unsaid: existential and anaphoric implicit objects in discourse

School College of Letters, Arts and Sciences

Degree Doctor of Philosophy

Degree Program Linguistics

Publication Date 04/01/2021

Defense Date 02/25/2021

Publisher University of Southern California (original), University of Southern California. Libraries (digital)

Tag acceptability judgements,accessibility,antecedent,Brazilian Portuguese,coreference,direct objects,discourse,English,garden-path,implicit objects,language comprehension,null objects,OAI-PMH Harvest,persistence,psycholinguistics,reference resolution,self-paced reading,sentence continuation,sentence processing,sprouting

Language English

Contributor Electronically uploaded by the author (provenance)

Advisor Kaiser, Elsi (committee chair), Mintz, Toben H. (committee member), Pancheva, Roumyana (committee member), Zevin, Jason (committee member)

Creator Email aninhavianna@gmail.com,pianibes@usc.edu

Permanent Link (DOI) https://doi.org/10.25549/usctheses-c89-435476

Unique identifier UC11667434

Identifier etd-PianiBesse-9378.pdf (filename),usctheses-c89-435476 (legacy record id)

Legacy Identifier etd-PianiBesse-9378.pdf

Dmrecord 435476

Document Type Dissertation

Rights Piani Besserman Vianna, Ana Elisa

Type texts

Source University of Southern California (contributing entity), University of Southern California Dissertations and Theses (collection)

Access Conditions The author retains rights to his/her dissertation, thesis or other graduate work according to U.S. copyright law. Electronic access is being provided by the USC Libraries in agreement with the a...

Repository Name University of Southern California Digital Library

Repository Location USC Digital Library, University of Southern California, University Park Campus MC 2810, 3434 South Grand Avenue, 2nd Floor, Los Angeles, California 90089-2810, USA