Close
About
FAQ
Home
Collections
Login
USC Login
Register
0
Selected
Invert selection
Deselect all
Deselect all
Click here to refresh results
Click here to refresh results
USC
/
Digital Library
/
University of Southern California Dissertations and Theses
/
Evaluating heaviness: Relative weight in the spoken production of heavy-NP shift
(USC Thesis Other)
Evaluating heaviness: Relative weight in the spoken production of heavy-NP shift
PDF
Download
Share
Open document
Flip pages
Contact Us
Contact Us
Copy asset link
Request this asset
Transcript (if available)
Content
INFORMATION TO USERS
This manuscript has been reproduced from the microfilm master. UMI
films the text directly from the original or copy submitted. Thus, some
thesis and dissertation copies are in typewriter free, while others may be
from any type of computer printer.
The quality of this reproduction is dependent upon the quality of the
copy submitted. Broken or indistinct print, colored or poor quality
illustrations and photographs, print bleedthrough, substandard margins,
and improper alignment can adversely affect reproduction.
In the unlikely event that the author did not send UMI a complete
manuscript and there are missing pages, these will be noted. Also, if
unauthorized copyright material had to be removed, a note will indicate
the deletion.
Oversize materials (e.g., maps, drawings, charts) are reproduced by
sectioning the original, beginning at the upper left-hand comer and
continuing from left to right in equal sections with snail overlaps. Each
original is also photographed in one exposure and is included in reduced
form at the back o f the book.
Photographs included in the original manuscript have been reproduced
xerographically in this copy. Higher quality 6” x 9” black and white
photographic prints are available for any photographs or illustrations
appearing in this copy for an additional charge. Contact UMI directly to
order.
UMI
A Bell & Howell Information Company
300 North Zed) Road, Ann Aibor MI 48106-1346 USA
313/761-4700 800/521-0600
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
EVALUATING HEAVINESS: RELATIVE WEIGHT IN THE
SPOKEN PRODUCTION OF HEAVY-NP SHIFT
by
Lynne Marie Stallings
A Dissertation Presented to the
FACULTY OF THE GRADUATE SCHOOL
UNIVERSITY OF SOUTHERN CALIFORNIA
In Partial Fulfillment of the
Requirements for the Degree
DOCTOR OF PHILOSOPHY
(Linguistics)
August 1998
Copyright 1998 Lynne Marie Stallings
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
UMI Number: 9919108
UMI Microform 9919108
Copyright 1999, by UMI Company. All rights reserved.
This microform edition is protected against unauthorized
copying under Title 17, United States Code.
UMI
300 North Zeeb Road
Ann Arbor, MI 48103
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
UNIVERSITY OF SOUTHERN CALIFORNIA
THE GRADUATE SCHOOL
UNIVERSITY PARK
LOS ANGELES. CALIFORNIA 90007
This dissertatiorir written by
...
under the direction of K .& .C ... Dissertation
Committee, and approved by all its members;
has b e e n presented to and accepted by The
Graduate School in partial fulfillment of re
quirements for the degree of
DOCTOR OF PHILOSOPHY
Date
\^ O e a n of Graduate^tudies
August 18, 1998
DISSERTATION COMMITTEE
Chairperson
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Dedication
To Daniel, your support and never-ending encouragement are
invaluable and greatly appreciated. To Cinnamon & Daiquiri,
thanks for making typing much less of a chore. Finally, to
friends and family, thanks for your encouragement along the
way.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Acknowledgements
This research was supported by NSF Grants DBS-9120415 and
SBR95-11270 to Dr. Maryellen C. MacDonald. Special thanks to
Danielle Cianfrocca, Jerry Cortrite, Jared Layport, Miranda
Lim, Summer Montague, Mariana Sapera, Daniel Stallings, and
Jessica Orena for assistance in data collection and transcript
coding. I am grateful to members of my qualifying committee,
Elaine Andersen, Ed Finegan, and Roger Woodard and especially,
John Hawkins and Maryellen MacDonald for their guidance and
numerous discussions about my work. Thanks also to colleagues
in Coglab, Joe Allen, Laura Gonnerman, Mike Harm, Todd Haskell
Marc Joanisse, Frank Manis, Mark Seidenberg, Sarah Schuster,
Karen Stevens, and Robert Thornton as well as others too
numerous to mention who listened to talks, read several
versions of my work, offered helpful suggestions, and
encouraged me as my work progressed.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
T a b l e o f C o n t e n t s
Dedication...................................................... n
ACKNOWLEDGEMENTS......................................................ZZZ
Lzst ar Tables...................................................vz
List or Figures................................................. vzz
Abstract...................................................... vziz
Chapter 1: Heavy-NP Shift: Am Overview.............................1
Heavy-NP Shift in Production................................................................................6
Implications for weight effects.........................................................................8
Summary..................................................................................................................................10
Chapter 2: A Linguistic Review or Heavt-NP Shift....................12
I. Heavy-NP shift literature review..........................................................13
Heaviness Properties ..................................... 13
Grammatical Explanations ............................... 15
Syntactic Factors...................................... 15
Syntactic-Prosodic Processes...........................18
Focus...................................................21
Discourse Factors...................................... 26
Functionalism........................................... 31
Syntactic and Discourse Processes......................31
Performance..............................................34
Relative Weight........................................ 34
Defining the EIC principle.............................35
II. Competence/Performance................................................................................40
Changing Constituent Structure: Rightward vs. Leftward
Movement...................................................40
Rightward Movement ..................................... 40
Leftward Movement....................................... 41
Performance in Relation to Grammar......................44
Grammaticalization of the [V NP PP] Order.............. 47
Grammar and Performance in Heavy-NP Shift ............. 48
Summary.................................................................................................................................. 51
Chapter 3: A Psycholihguistic Perspective of Heavy-NP Shift.......... 56
I . Research Motivating Speech Production Models............................56
Speech error data......................................... 59
II. Heavy-NP Shift in Production.............................................................. 67
Semantic and Phonological Influences on Ordering......... 68
Animacy..................................................69
Thematic Role Structure ................................ 70
Length 72
Effects of Lexical and Referential Accessibility on Given-
New Ordering............................................ 75
III. Hearer versus Speaker Accommodation in Heavy-NP Shift 81
Evidence in Favor of Listener Accommodation.............. 81
Evidence Against the Model of the Listener............... 83
Speaker/Listener Predictions for Heavy-NP Shift ......... 92
Summary..................................................................................................................................94
iv
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Chapter 4: The Role or Verbs ahd Relative Height nr Heavy-NP Shift 97
The Role of Verbs in Heavy-NP Shift............................................................97
A Competitive Component ............................... 101
NP Length Manipulated.............................................................................................107
Experiment 1: Manipulating PP Length.....................................................110
Method................................................. Ill
Results and Discussion ................................ 114
Experiment 2: NP Length and PP Length Manipulations..................119
Method................................................. 120
Results and Discussion................................. 121
Summary...............................................................................................................................125
Chaster 5: Future Directions..................................... 129
Experiment Results....................................................................................................133
Sentential Complement Verbs and Heavy-NP Shift ......... 135
Calculating Phrase Length.............................. 136
Fundamental Production Processes ....................... 137
Multiple Interacting Constraints.................................................................138
V-PP Collocations...................................... 138
V-PP Collocations and Weight..........................141
V-PP Collocations and Verb Disposition................144
Additionai. Factors nr Production..................................145
The role of competition in production models.............147
Future Directions.......................................................................................................150
References.....................................................154
Appendix A .....................................................162
Appendix B .....................................................167
Appendix C .....................................................173
v
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission
L i s t o f T a b l e s
Table 1
Table 2
Table 3
Table 4
Table 5
Table 6
Table 7
: Sample items from Stallings, MacDonald, & O'Seaghdha
(in press) ....................................... 100
: Mean shifting percentages of NP—only and NP/S verbs
(with Standard Deviations) from Stallings, MacDonald,
& O'Seaghdha (in press)............................101
: Sample items in Experiment 1: PP length manipulated 111
: Mean number of words recalled for each phrase......117
: Experiment 1: Mean NP-PP length recall as a function
of structure choice* •*■**•»••■••••••■*■•**•••••••• 118
: Sample items in Experiment 2: NP and PP length
manipulated...................................... 121
: Experiment 2: Mean NP-PP length recall in 5-word
difference conditions..............................124
VI
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission
L i s t o f F ig u r e s
Figure 1
Figure 2
Figure 3
Figure 4
Figure 5
Figure 6
: Syntactic tree structure of a focused N P ..........23
: Tree structure of a heavy-NP shift sentence ...... 41
: A planning frame before open-class items are placed
into their slots at the positional level.......... 63
: Results from Stallings, MacDonald, & C'Seaghdha's
prompted recall task: Production of heavy-NP shifted
structures as a function of NP length............ 110
: Experiment 1 results: Production of heavy—NP shifted
structures in a prompted recall task as a function of
PP 1 e^c^th* #••••••••••••»■••••••••••••••••••■••*** 116
: Experiment 2 results: Production of heavy—NP shifted
structures in a prompted recall task as a function of
NP and PP length* 123
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Abstract
Heavy-NP shift is the tendency for speakers to place long or
"heavy" noun phrase direct objects at the end of a sentence
rather than in the canonical post-verbal position. The precise
nature of heaviness that motivates this tendency, however,
varies throughout the linguistic literature and depends more on
the nature of the theoretical assumptions underlying each
proposal. Evidence for a performance-motivated account of
heavy-NP shift is provided and a multi-factor account of the
spoken production of heavy-NP shift is considered. A single
factor account of weight is challenged by the influential role
of verbs on the production of heavy-NP shift structures.
Results from three production experiments using a constrained
production paradigm confirm that it is the relative weight
between the NP and its neighboring constituent, not the length,
complexity, or some property of the noun phrase alone, that
influences the ordering of the noun phrase and prepositional
phrase during spoken production.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Chapter 1: Heavy-NP Shift: An Overview
In the heavy-NP shift sentence 'Jake explained to Jim all of
the facts regarding the import and export taxes, ' a long, or
'heavy, ' direct object noun phrase (NP), appears in clause-
final position, with a prepositional phrase (PP) intervening
between it and the verb. ’Shifted* sentences such as this
contrast with * mishitted' or 'basic order' sentences, in which
the verb and NP direct object are adjacent, as in Jake
explained all of the facts regarding the import and export
taxes to Jim. Most speakers of English find both the shifted
and basic order structures to be acceptable when the direct
object NP is long, as it is in this sentence. When the direct
object NP is short, however, speakers have a strong preference
for the basic order (e.g., Jake explained the facts to Jim),
and shifted structures (e.g., Jake explained to Jim the facts)
are typically judged to be extremely awkward or ungrammatical.
The exact conditions under which intuitively short NPs are
deemed awkward or ungrammatical (and alternatively, when
intuitively long NPs are sufficiently long to be considered
acceptable) are difficult to capture, and hence, explanations
for word ordering in heavy-NP shift have proved elusive.
Several attempts to explain heavy-NP shift seem to derive
from at least some of Behaghel's (1932) principles of ordering.
Hawkins (1994) summarizes three of the Behaghel's (1932, p. 4)
principles as das oberste Gesetz, in which items that are
mentally close appear near one another; the Gesetz der
1
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
wachsenden Glieder, in which short items precede longer items;
and a pragmatic ordering principle (es stehen die alten
Begriffe vor den neuen) according to which given information
precedes new information. As Stockwell (1311) points out, the
primary determinant of ordering for Behaghel is the oberste
Gesetz principle (see Hawkins, 1994 for further review). As
Hawkins points out, this principle calls for continuous
syntactic structures so that an adjective, for example, should
be bound to the noun that it modifies. But, as is illustrated
in heavy—NP shift structures, the adjacency principle fails to
hold true at all times. stockwell (1977) indicates that the
additional short before long and given before new principles
(what he calls the heavier element and topicalization
principles, respectively) can actually override the adjacency
principle.
Various explanations for heavy—NP shift incorporate these
principles at least to some extent. Several syntactic
explanations, for instance, are offered in the spirit of the
short before long principle. But as chapter 2 will illustrate,
such explanations vary considerably and differ from a mere
preference to place short items before long ones. In the
spirit of the given before new ordering principle, Firbas
(1966) has proposed that heavy—NP shift is driven not by
weight, but by pragmatic factors such as the placement of given
before new information. As with the syntactic explanations,
the pragmatic proposals also vary in their accounts of heavy-NP
7
with permission of the copyright owner. Further reproduction prohibited without permission.
shift. Given the variety of proposals, it is curious which
factors are actually motivating heavy-NP shift.
The answer is not made any easier by the additional factors
that have been proposed to explain heavy-NP shift. In addition
to syntactic (Emonds, 1976; Erdmann, 1988; Kayne, 1994; Larson,
1988; Postal, 1974; Ross, 1967) and informational/pragmatic
(Dik, 1978; Firbas, 1966; Givon, 1988; Rochemont & Culicover,
1990; Siewierska, 1991) levels of linguistic representation,
prosodic (Zee & Inkelas, 1990) and lexical-semantic (Hawkins,
submitted; Stallings, MacDonald, & O'Seaghdha, in press; Wasow,
1997a; Wasow, 1997b) factors have also been proposed to account
for the placement of the NP in clause—final position.
Proposals for heavy-NP shift vary not only along the
dimension just mentioned. That is, whether shifting is
influenced by syntactic, pragmatic, prosodic, or lexical-
semantic factors. Within these various factors, the
explanations overlap across even further dimensions. Two
questions aid in identifying the additional overlap. Namely,
what are the defining characteristics of a given factor? And,
what are the assumptions underlying that factor? Research
addressing the defining characteristics of syntactic factors,
for example, seems to indicate that heaviness is not determined
at the syllable level (Rickford, Wasow, Mendoza-Denton, &
Espinoza, 1995; Stallings, 1994) nor does the length of the NP
alone seem to be sufficient. Beginning with Ross (1967),
several researchers propose that it is the complexity, not the
mere length, of the NP that motivates shifting. But among the
3
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
proposals that advocate NP complexity as the motivating factor,
even the definition of complexity varies. Moreover, syntactic
proposals are not limited to NP complexity. Hawkins (1994),
for instance, proposes that an explanation based on NP
complexity alone is also limited, arguing instead that it is
the weight of the NP relative to that of the PP that matters.
In addition to the variation in the nature of complexity being
proposed, the theoretical underpinnings of these proposals also
vary. Whereas the explanation forwarded by Ross is grounded in
Chomskyan assumptions about linguistic competence, for example,
Hawkins’ proposal is motivated more by performance factors.
The variations both between and within the types of factors
that motivate heavy-NP shift raise several questions about the
underlying nature of heavy-NP shift structures. The
fundamental question involves the underlying nature of
heaviness that induces shifting: Is heaviness defined by
properties of the NP alone, by the relative weight between the
adjacent NP and PP constituents, or some combination of the
two, since the length or complexity of one constituent can
feasibly contribute to the relative weight between two
constituents?
The answer to this question may depend partly on the
assumptions that underlie the explanations. If properties of
the NP alone influenced heavy-NP shift, for example, what is
the precise point at which rearrangement occurs? A grammatical
rearrangement rule could require that NPs with property X,
where X would be coherently defined, appear in clause-final
4
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
position. Relative weight, on the other hand, assumes more
gradient properties between constituents so that the NP may not
consistently appear in clause-final position at a determined
point (cf. short before long accounts which are more
grammatical in nature, e.g. Behaghel, 1932; and Stockwell,
1977) . Moreover, in chapter 2, it will become evident that a
relative weight account like the one proposed by Hawkins (1994)
questions the strictly generativist accounts that do not allow
for performance demands to impact ordering (e.g. Chomsky,
1965). Thus, whether grammatical or performance—based
explanations are better suited to adequately describe heavy-NP
shift data is another question that deserves attention.
The possibility that performance phenomena are capable of
driving heavy-NP shift requires that heavy-NP shift be
investigated with both competence and performance-relevant
measures. The linguistic literature reviewed in chapter 2
relies mostly on off-line acceptability ratings to reflect an
individual's competence best, but avoids on-line experiments
that actually test how people use their competence.
In a performance-motivated approach to ordering, Hawkins
(1994) analyzes text data to explore relative weight
predictions. Wasow (1997a; 1997b) approaches the task in a
similar manner, using the Brown corpus as a means for data.
But reliance on text data, or acceptability ratings for that
matter, limits conclusions about the effects of factors on
heavy-NP shift to comprehension-related processes without
5
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
really explaining what motivates the production of heavy-NP
shift structures in the first place.
In chapter 3, a production account of heavy-NP shift by
Wasow (1997a) is reviewed to consider whether heavy-NP shift is
induced more out of concern for the listener or as a result of
speaker-oriented processes. Because he argues that production-
based explanations can account for the data better than
parsing-based explanations, he extends his investigation of
text data to speech corpora, analyzing a transcribed corpus of
telephone conversations (the Switchboard corpus from Texas
Instruments).
An additional way to consider production effects on ordering
involves the use of experimental paradigms. Such paradigms
could conceivably tap into the real time course of production
processes, allowing research to move beyond the information
that has thus far been gained through analyses of
grammaticality judgments and text alone. Chapter 3 addresses
the question, how do psycholinguistic models actually account
for heavy-NP shift? That is, how would heavy-NP shift fit into
current production models?
Heavt-HP Shift nr Production
One potential limitation of investigating the spoken
production of heavy-NP shift is the relative infrequency of
shifting that occurs in natural spoken language. Given that,
we must question whether a constrained production paradigm is
even capable of inducing heavy—NP shifted structures. In
chapter 4, recent experiments by Stallings, MacDonald, &
6
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
O'Seaghdha (in press) are presented that indicate it is
possible to induce the spoken production of heavy-NP shift in a
constrained situation. In fact, in those experiments, three
variants of a sentence construction procedure used by (Dell &
O'Seaghdha, 1992) all elicited a significant percentage of
shifted structures. Essentially, participants read sentence
fragments on a computer screen, formed a plan for producing a
complete sentence, and then said the sentence aloud (from
memory) when cued by the subject-verb phrase of the sentence.
The primary goal of this research is to further consider the
specific nature of heaviness that influences the spoken
production of heavy-NP shift structures. Because testing
relative length differences at various levels in an
experimental paradigm could entail a somewhat monumental (if
not virtually impossible) task, effects of relative weight are
deduced by comparing the presence of NP length effects in
Stallings et al. to the presence of PP length effects in
Experiment 1 of chapter 4. Varying the length of the PP and
holding NP length constant should therefore provide further
information about the nature of length effects in production.
If relative weight, and not simply the mere length, or
complexity, of the NP, constrains the production of shifted
structures, then length effects will also obtain when PP length
is varied and NP length is held constant. Alternatively, we
will not find such effects if shifting is constrained only by
NP length.
7
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
An additional experiment that manipulates both the NP and PP
presents an attempt to fully address the extent to which
relative weight is confounded with or operates independently of
NP length in motivating shifting. A significant difference
between two conditions that are equal in relative weight, but
vary in terms of NP length would suggest that both relative
weight and NP length influence heavy-NP shift. Alternatively,
no difference between such conditions would argue for relative
weight effects only.
Implications for height bwsxcts
Answering questions about the nature of heaviness will
reduce the amount of variation that heavy-NP shift receives in
the linguistic literature. However, the answers will not only
add to our understanding of the linguistic literature. They
should also influence the psychoiinguistic literature. If as
Wasow (1997a; 1997b), Stallings, MacDonald, & O'Seaghdha (in
press), and Hawkins (submitted) suggest, a multifactor theory
of heavy-NP shift is warranted, then knowing the extent to
which NP length, relative weight, or some combination of the
two, induces heavy-NP shift will facilitate further exploration
of the constraints with which weight interacts.
Evidence that factors in addition to weight influence
shifting appears in Wasow (1997a; 1997b), Stallings et al. (in
press) , and Hawkins (submitted) . As mentioned earlier in the
chapter, pragmatics, prosody, and lexical—semantic factors are
potential motivators of heavy—NP shift, though Hawkins (1994)
argues that pragmatic factors account for no additional
8
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
variance once weight is considered. The small sample of 22
shifted structures in Hawkins' sample suggests, however, that a
larger sample must be considered before any conclusions about
the effects of pragmatic factors on heavy-NP shift can be
forwarded.
As chapter 3 will demonstrate the nature of pragmatic
factors is confounded with weight, as is prosody. The longer
the item, the more likely it is to contain newer information or
more prosodic structure. In accordance with Hawkins (1997),
therefore, pragmatic (or prosodic) factors should not be
assessed without also taking into account weight effects. To
do so effectively requires that the precise nature of weight
effects on heavy-NP shift be determined first. Whether
properties of the NP alone, relative weight, or both, motivate
shifting, uncovering the precise nature of heaviness is a
necessary preliminary step to further understanding the
interaction between it and additional factors.
In contrast to the traditional single-factor approach
assumed in much of the literature surrounding heavy-NP shift
(Chomsky, 1965; Postal, 1974; Rochemont & Culicover, 1990;
Ross, 1967), and in conjunction with evidence that appears in
chapters 2 and 3, I assume that multiple factors combine to
constrain the production of heavy-NP shift structures. This
assumption follows primarily from comprehension work by
MacDonald, Pearlmutter & Seidenber (1994) in which detailed
lexical representations are encoded by a wide variety of
frequency-sensitive information. As chapter 4 will show, a
9
with permission of the copyright owner. Further reproduction prohibited without permission.
multiple factor approach is further supported by production
work in Stallings, MacDonald, and O'Seaghdha (in press), in
which verbs that frequently appeared non-adjacent to their
direct object complements were most likely to influence the
production of shifted structures in an experimental task.
Stamasx
In summary, theories about heavy-NP shift vary and overlap
at many junctions. The complex and overlapping distinctions
among the various proposals raise several questions about the
nature of the processes that affect heavy-NP shift. Two main
points arise from the preceding discussion. First, the precise
nature of weight effects is as yet undetermined. Second, from
a performance-motivated perspective, a multiple factor theory
seems like the best way to fully account for the spoken
production of heavy-NP shift sentences. Without a complete
understanding of the nature of weight effects, however, a
detailed account of such a theory is difficult to obtain.
The focus in the following chapters, therefore, is on
investigating the psychological reality of the two syntactic
proposals mentioned above. Namely, whether properties of the
NP alone, the relative weight between it and other material in
the verb phrase, or some combination of the two, motivates
shifting in the spoken production of heavy-NP shift. Chapter 2
begins with a review of the linguistic proposals that have thus
far been offered to explain heavy—NP shift. It then offers
evidence that heavy-NP shift is possibly motivated by
performance processes alone. Following from that evidence,
10
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
chapter 3 therefore considers psycholinguistic evidence
pertaining to constituent ordering in production. Models of
speech production and their underlying motivations are
considered first, followed by a review of some production
studies and a discussion of possible factors that may
contribute to the shifting of an NP to clause-final position.
Chapter 4 presents data from some production experiments by
Stallings, MacDonald, & O'Seaghdha (in press) and then offers
two additional experiments that address the precise nature of
weight effects in the spoken production of heavy-NP shift. In
chapter 5, future research directions are outlined.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Chapter 2: A Linguistic Review of Heavy-NP Shift
References to shifting phenomena appear in the linguistics
literature at least as far back as the turn of the century
(Wasow, 1997b) . Since then, a wide variety of hypotheses have
been offered concerning the reasons why heavy-NP shift should
exist. One point of debate is whether shifting is or is not a
purely syntactic phenomenon. That is, whether it is triggered
by syntactic properties of the NP and other phrases or whether
other factors such as pragmatics and co-occurrence relations
influence shifting. A second dimension over which accounts of
heavy—NP shift vary is along a competence-performance
continuum: namely, to what extent are the miles of the
competence grammar generating shifted and unshifted sentences,
and to what extent is the variation the result of preferences
of a purely processing (i.e., performance) nature? This
chapter considers these dimensions in detail. In section I,
syntactic and non-syntactic properties that authors have
proposed as triggers for heavy-NP shift are reviewed,
accongianied by evidence that their ties to purely grammatical
(competence-driven) explanations limit some proposals. In
section II, evidence that heavy-NP shift lies along a specific
part of the competence-performance continuum is evaluated. In
both sections, I evaluate the criteria for considering
transformations as motivated by grammatical rearrangement rules
and provide support for a performance-motivated account.
12
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
I. Heavy-NP shift literature review
Traditionally, the grammatical literature has assumed that a
rearrangement rule identifying syntactic properties determines
the presence of the [V PP NP] shifted order- Criteria to
assess the status of a grammatical rearrangement rule are
listed below in (la-b). The applicability of the necessary
condition (la), that rearrangement rules refer to a
grammatically coherent category, is addressed throughout this
section. A sufficient condition (lb) for status as a
grammatical rearrangement rule, namely the extent to which
heavy-NP shift changes constituent structure is further
assessed in section II.
A rule or ordering must be formulated in terms of:
la. grammatically definable, or coherent, categories
A rule of ordering may introduce:
lb. grammatical changes in addition to the reordering
itself (e.g., a change in constituent structure)
(Hawkins, 1994, p. 91)
Heaviness Properties
While intuitive notions of heaviness are clear (Haegeman,
1991; Wasow, 1997b) and researchers agree that it induces
shifting, the precise characterization of heaviness is yet to
be determined. Haegeman (1991), for example, asserts that
rightward adjunction applies only when the NPs are heavy, but
only illustrates a heavy (e.g., 2a) and a non-heavy (e.g., 2b)
NP. Beyond the NPs provided in these examples, no definition
of a heavy NP is offered.
13
with permission of the copyright owner. Further reproduction prohibited without permission.
2a. Jeeves introduced to the guest the famous detective
from Belgium.
2b. *Poirot speaks badly English.
(Haegeman, 1991, p. 382).
A number of researchers have assumed that shifting is
triggered by a syntactic property of the heavy NP (Emonds,
1976; Postal, 1974; Ross, 1967). Some researchers have
proposed that a heavy NP is sufficiently defined in terms of
length (Postal, 1974) . Other researchers (Emonds, 1976;
Kimball, 1973; Quirk, 1972; Ross, 1967), observing that the
long/complex material of a sentence appears at the end, have
suggested that it is the syntactic complexity of the NP, not
its length, that is the crucial syntactic property governing
shifting. On this view, long NPs shift because they tend to be
complex. However, there has been no consensus on how exactly
to characterize syntactic complexity (for review, see Wasow,
1997b) .
Other attempts to define heaviness have included an
interaction of syntactic-prosodic processes (Zee & Inkelas,
1990), informational structure/focus (Firbas, 1966; Rochemont &
Culicover, 1990) and the relative weight of the NP and the
constituent over which it shifts, which extends beyond
properties of the NP alone (Behaghel, 1932; Bever, 1970; Dik,
1978; Hawkins, 1994; Stockwell, 1977). Plausible explanations
for weight clearly present a myriad of definitions. In this
section, I review the syntactic and non-syntactic explanations
for heavy-NP shift in detail. Some proposals are grammatical
14
with permission of the copyright owner. Further reproduction prohibited without permission.
in nature, while others are motivated by functional or
performance principles.
Grammatical Explanations
Syntactic Factors
Behaghel's (1932) definition of grammatical weight
emphasized the relative length between pairs, such that short
elements precede long items, but proceeding definitions focused
more on the syntactic properties of the NP alone to explain
shifting. In Ross (1967), for example, complex NP shift is
described as a rearrangement in which the NP appears in clause-
final position on the condition that it dominates a clause.
This structural definition adequately describes the
grammaticality of the basic order in (3a) and the shifted order
in (3c), as well as the ungrammaticality of the shifted order
in (3b) . Yet it fails to explain why NPs that do not dominate
S, such as those in examples (4) and (5) , can also be shifted.
3a. He attributed the fire to a short circuit.
3b. *He attributed to a short circuit the fire.
3c. He attributed to a short circuit the fire which
destroyed most of my factory.
(Ross, 1967, p. 29).
4a. They selected as revolutionary of the month the most
obnoxious member of Dominance Now.
4b. I showed to Jack all of the coffee ground and lettuce
cookies.
(Postal, 1974, p. 83).
15
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
5a. ...loose upon mankind such terrible and inhumane
weapons.
5b. ...identify independently a given value system or
behavior pattern.
6a. ...employ at all times the correct research methods.
6b. ...affied unto us his best endeavor.
(Brown Corpus).
In (4), the NP dominates PP and in (5) the NP consists of
two coordinated nominal items. The examples in (6), however,
contain no postmodifiers. As a result, theories stating that
complex NPs contain postmodified nouns (Emonds, 1976) describe
the grammaticality of examples (4) and (5), but cannot account
for the grammaticality of the examples in (6). Postal's (1974)
definition of a complex NP as one that it is long and/or
clause-containing accounts for the shifting of both non clause-
containing NPs (ex's. 4-6) and clause—containing NPs (ex. 3).
Importantly though, what constitutes a 'long NP' is left
unspecified.
Like Postal (1974), Quirk et al. (1972) define heaviness in
terms of both complexity and length, but they specify that the
length of an NP is defined by the number of syllables it
contains (for further review, see Leech & Svartvik, 1975).
Recent research seems to indicate that heaviness is not
determined at the syllable level (Rickford et al., 1995;
Stallings, 1994). The simplest option for defining a long NP
then is the length in number of words of the NP.
16
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Assuming length were measured by the number of words, one
could hypothesize that a 'long' NP must contain 5 or more
words, since the NPs in examples (4) and (5) contain at least
that many words. If we consider the less intuitively heavy NPs
in example (6), a 'long' NP should consist of three words. On
the same basis, according to the ungrammaticality of 3b, a two-
word NP should not be long. Yet, the extent to which length
alone adequately accounts for the data is challenged when we
compare (3b) to the examples illustrated in (7) . Whereas
shifting a two-word NP in (3b) is ungrammatical, doing the same
to a two (ex. 7a-b) or one-word (ex. 7c) NP in the latter
examples results in grammatical structures. These
counterexamples further illustrate both the difficulty in
assessing precisely what constitutes a heavy-NP and an inherent
problem in explaining heavy-NP shift via a grammatical
rearrangement rule. As we saw in (la), one of the necessary
criteria is that a rule must contain a grammatically
definable/coherent category. Whether the NP is defined as
long, complex, or long and/or complex, the inherent difficulty
in defining a heavy NP is apparent, precisely because ’heavy’
is only intuitively1, not grammatically, definable.
7a. ...declare publicly their faith.
7b. ...want from me the story.
7c. ...appeared to them libertines.
(Brown Corpus).
Writing a grammatical rearrangement rule using a
grammatically coherent category also fails to capture the
relevant data. Given the fact that an NP is a coherent
17
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
grammatical category and leaving the ungrammaticality of (3b)
aside for a moment, the examples in (3-7) might favor a rule
stating 'move NP.’ But such a rule results in the over
generation of shifted structures by necessitating the shifting
of all NPs: Both the ungrammaticality of (3b), and the presence
of any basic order sentence illustrates the problems with a
rule such as 'move NP. '
Additional proposals that are not strictly related to
syntactic weight also attempt to explain these data, but pose
similar problems because of the ways in which they also attempt
to define the properties of a heavy NP. The next section
considers the extent to which syntactic-prosodic factors
adequately capture the relevant data.
3yntactie-Prosodxc Processes
Zee & Inkelas (1990) argue that it is not the grammatical
component itself that is problematic, but rather the attempt: to
describe heaviness at the syntactic level of linguistic
reality: "syntax is simply not the right component in which to
look for generalizations about constituent weight" (p. 372).
Consequently, they define heaviness phonologically whereby the
reordering of constituents in heavy-NP shift yields better
prosodic contours. For example, (8a) presumably yields better
prosodic contours than does (8b) , which appears in the basic
order. That reordering yields better contours challenges
strict incremental production models (e.g., Levelt, 1989) that
assume a uni-directional relationship between syntax and
phonology. Zee & Inkelas (1990) therefore assume that the
18
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
interaction between syntax and phonology is bi-directional.
Thus, in addition to the constraints syntax exerts on
phonology, phonology also constrains syntax. Their view is
supported by evidence that ordering is sometimes better
predicted by prosodic groupings than syntactic structure (e.g.,
Zee & Inkelas, 1990) . Heaviness is thus realized not at a
syntactic, constituent-structure (herein C-structure) level,
but at a phonological, prosodic-structure (herein P-structure)
level.
8a. Mark showed to John some letters from his beloved
city.
8b. Mark showed some letters from his beloved city to
John.
(Zee & Inkelas, 1990, p. 377)
More specifically, at the phonological level, heaviness is
realized at the prosodic level due to evidence that it is not
the length of the words but the number of the words that
determines sentence production latency (Sternberg, Monsell,
Knoll, & Wright, 1978). If phonological rules, beyond the
prosodic level, influenced syntax directly, one would expect
the length of the words within a phrase to affect latencies.
Thus, the influence of phonology on syntactic processes fails
to occur beyond a prosodic level.
To satisfy the heaviness condition, Zek & Inkelas (1990)
state that the structure must branch. Accordingly, they claim
that heaviness is realized at P-structure, not C-structure,
given the plausibility that proper names comprising more than
19
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
one word can form one syntactic unit. The contrast between the
grammaticality of the Serbo-Croatian phrase 'u Rio de Zaneiru'
(in Rio de Janeiro) and the ungrammaticality of the phrase *u
Riju de Zaneiru', for example, illustrates that the phrase as a
whole derives in the lexicon. The first words (e.g., Rio) are
only case marked when they appear in isolation. The authors
(Zee & Inkelas, 1990) thus argue that the NP 'Rio de Zaneiro,'
which exemplifies only one category, forms only one syntactic
unit, so it is not heavy (in the branching sense) at C-
structure. On the other hand, it does satisfy the branching
requirement for heaviness at P-structure, because it contains
two phonological words (for further review, see Selkirk,
1980)2.
Moreover, Zee & Inkelas (1990) posit that, for heavy-NP
shift, it is the intonational phrase that must branch. This
claim derives from the fact that heavy NPs are typically
preceded by a pause and are associated with special intonation
effects (p. 377). Perlmutter, in Postal (1974), supports this
notion, arguing that a sustained intonation appears prior to a
shifted NP. Thus, an NP is heavy if it comprises an
intonational unit which contains 2 or more phonological
phrases, as in the phrase illustrated by Zee & Inkelas (1990),
'some letters from Paris'. The phrase 'some letters'
constitutes only one phonological phrase, so it is not
considered heavy.
Assuming the words 'Barry' and 'Bonds' each constitute a
phonological phrase, thus branching at the Intonational Phrase,
20
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
the shifting of the HP 'Barry Bonds' in example (9a), provided
by Wasow (1997a), can be explained. However, it does not seem
to explain the shifting of the proper name 'Bonds’ alone in
(9b), which comprises only one phonological phrase (and one
prosodic word for that matter) .
9a. That will bring to the plate Barry Bonds.
9b. That will bring to the plate Bonds.
(Wasow, 1997a).
As with the syntactic proposals, this phonological
explanation also fails to provide a grammatically coherent
definition of a heavy HP. It cannot explain the presence of
shifting in example (7c), where the HP contains only one word
(libertines) , nor would it explain the shifting of the one-word
proper name 'Bonds' in example (9b). Hor does Selkirk's (1980)
condition under which a one-word HP appears in shifted position
explain examples like those in (7c), since on her account
shifting occurs only if the one-word HP applies to subjects.
This condition does not apply to heavy-HP shift for the obvious
reason that the HP is a direct object. These examples that
shift one-word HPs to clause-final position3 therefore
demonstrate that the problems posed by purely prosodic
explanations are similar to those raised by purely syntactic
explanations; they both fail to provide a descriptively
adequate account of the data.
Focus
The difficulty in describing the shifting of one-word HPs
drives Rochemont & Culicover (1990) to argue that heaviness
21
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
judgments are not absolute, but instead are relative, noting
Bresnan's (1976) and Rochemont's (1978) observation that "even
proper names are marginally acceptable in cases of heavy NP
shift" (p. 107). Rochemont & Culicover (1990), like Zee &
Inkelas (1990), argue that proper names can appear in shifted
position when a heavy pause separates the name and its prior
material and a strong accent accompanies it (p. 107). Unlike
Zee & Inkelas (1990), Rochemont & Culicover (1990) argue that
it is not prosodic structure that motivates ordering. Nor is
it an interaction between prosody and length as Nespor & Vogel
(1986) might advocate, wherein longer items have a greater
likelihood of containing more intonational and phonological
phrases than shorter items. Rather, it is focus. In an
attempt to maintain motivation within the principles of UG,
their account of focus is syntactically rooted. They maintain
that a focused constituent displays "greater phonological
prominence" (p. 17) than an unfocused constituent in a given
intonational domain, but they argue that phonological
prominence is not due to prosodic factors. Instead, they argue
that it is displayed as accent, and they assume the accent
"correspond[s] always to a syntactic constituent" (p. 17).
Consequently, the basis of their analysis assumes a specific
structural relationship and prosody serves only as a marker
that an item is focused.
With respect to heavy-NP shift, NPs shift not on the basis
of heaviness, but on the structural configurations of focus In
their view, shifting is structurally, not stylistically,
22
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
motivated. The Focus Principle states that the NP is focused
if it meets the criteria in (10). Figure 1 illustrates a
focused NP: the verb canonically governs the NP, the NP is
neither case nor 0-marked by the verb, and the NP is not a
predicate that is 0-related to the verb.
10. The Focus Principle
a is a structural focus if
(i) there is a lexical head P(V) that canonically
governs a and a (NP)is neither Case-marked nor
0-marked by p.
(ii) a is not a predicate that is 9-related to p.
(Rochemont & Culicover, 1990, p.156).
Figure 1: Syntactic tree structure of a focused NP
vp
(Rochemont & Culicover, 1990, p. 152).
Due to their emphasis on the structural properties of focus,
Rochemont & Culivover (1990) do not acknowledge the role of
weight on any account in heavy-NP shift. Such emphasis
relegates heaviness to a secondary role where heaviness is
described only in relation to focus. In fact, they define a
heavy NP only as one that "appears in A-bar position and is
canonically governed by a lexical head" (Rochemont & Culicover,
1990, p. 157), as it does in Figure 1. The relationship
between the heaviness and focus principles assumes that if an
23
with permission of the copyright owner. Further reproduction prohibited without permission.
NP adheres to the criteria for the former, it will necessarily
satisfy the criteria for the latter (e.g., (10)). The converse
relationship is false: a focused item is not heavy. This claim
derives from the fact that not all structures need a heavy
constituent to appear in focused position. If they did.
Stylistic Inversion structures such as Directional/Locative
Inversion (e.g., Into the room walked John) and Preposing
around Be (e.g., Standing in front of her was Mary) would
involve heaviness; instead, they are not heavy, but merely
focused (Rochemont & Culicover, 1990).
This kind of account clearly provides the structural basis
by which heavy-NP shift is defined, but it does not explain the
underlying motivation for shifting an NP to clause-final
position. Instead it raises questions afcout how focus or
heaviness is determined. In addition, Hawkins (1994) observes
that this account does not differentiate between the basic and
shifted orders. It is unclear if only the NP is eligible to
appear in focused position, or if the PP can also be focused.
Rochemont & Culicover (1990) do not address the latter issues.
They do, however, present a rubric in order to assess focus.
Their rubric for identifying focused material illustrates
that just as heaviness is difficult to define, the same is true
for focus. Specifically, the authors suggest that information
that is accented in response to a question will appropriately
identify focused material. Thus, they employ a wh-
question/answer test, requiring that "all and only the
information requested in the question is focused in the
24
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
response" (p. 18). All and only the information is defined as
anything that "constitutes new information"4 (p. 20) in
relation to the context to which the information is being
added, so that newer information will constitute the focused
items.5 On this basis, the answer in (12), for example,
qualifies as an appropriate response to the question in (11a),
precisely because a brand new fur coat contains new information
in relation to the question that was posed.
11a. What did John purchase for his wife?
lib. For whom did John purchase a brand new fur coat?
12. John purchased for his wife a brand new fur coat.
(Rochemont 5 Culicover, 1990, p. 24).
The problem inherent in this rubric is clear in the
question/answer sequence in (13-14), which illustrates how
focus is difficult to define on the basis of new and given
information. The rubric only states that new information must
be focused. The converse relationship is not true, however.
That is, focused information can be either new or given. For
example, both of the answers in (14) qualify as focused
responses to the question in (13), even though Mary in (14b)
constitutes new information but JOHN/HIM in (14a) constitutes
previously mentioned (given) material. Determining the focus
status of an item based on its syntactic configuration is
appropriate, but this example illustrates the inherent problem
in defining focus based on whether or not it is construable by
the context at hand. In sum, it is clear that focus may be
25
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
explained at a syntactic level, but it is insufficient at an
interpretive level6.
13. Who does John's mother like?
14a. John's mother likes JOHN/HIM.
14b. John's mother likes MARY.
(Rochemont & Culicover, 1990, p. 21).
Although Rochemont & Culicover maintain that the structural
configuration of focus limits an item's distribution in
discourse such that focus is stylistic only as a consequence of
the required configuration, the emphasis on the wh- Q/A well-
formedness test misconstrues their theory as a discourse-
motivated one. Consequently, Rochemont & Culicover's original
intention to "reduc[e] the class of stylistic rules...to
independently motivated principlefs] of UG" (1990, p. 143) is
undermined. In addition, this account, like the syntactic ones
that attempted to define complexity, demonstrates the
difficulty in defining focus. It also fails to identify a
clear motivation for moving the NP to its specific structural
configuration.
Discourse Factors
The previous section highlighted the notion that
phonologically prominent material frequently appears at the end
of the sentence. This idea is not strictly tied to syntactic
or prosodic processes, however. Researchers have also offered
discourse theories to account for the presence of
phonologically prominent material at the end of the sentence.
In fact, the placement of such material facilitates the
26
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
successful communication of a message by emphasizing new
information (Clark, 1977; F. Ferreira, 1993; Firbas, 1966).
Prosodic factors therefore facilitate the extent to which a
message is successfully conveyed by a speaker, such that NPs
are more likely to convey new information when they appear in
shifted position.
Some theorists would argue that Rochemont & Culicover should
focus instead on the discourse-based properties of heavy—NP
shift, and reject the syntactically rooted hypothesis presented
above. They argue that shifting results neither from
syntactic, nor prosodic, properties of the NP, but only from
discourse-level information (Firbas, 1966 in Givon, 1988) .
That is, long NPs are shifted to the end of the sentence not as
a consequence of their length or phonological prominence, but
as a consequence of the information status they convey. One
significant problem within these proposals is that they
contradict one another. On the one hand, Givon (1988) argues
that new information should precede given information. On the
other hand, Firbas (1966 in Givon, 1988) reasons that because
long NPs tend to contain more new information whereas shorter
NPs contain more known information, shifting satisfies
pragmatic constraints favoring the given-new ordering of
information in English. On this view, an item might receive
stress merely as a function of discourse processes.
Of the examples illustrated above, only (3b) and (9)
illustrate how discourse factors such as the informational
status of constituents might constrain shifting. In (3b), the
27
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
indefinite article in the PP {to a short circuit) signals new
information that conveys the cause of the fire. In accordance
with Firbas' (1966) given-new theory of word ordering, the
ungrammaticality of (3b) is explained by the placement of a new
entity before a given one. Example (9) is a little more
difficult. Wasow (1997b) points out that the presence of the
NP in clause-final position actually provides a counterexample
to the given-new ordering theory in the following manner. The
announcer utters the statement throughout the course of the
entire game, and as the game goes into the final innings, the
information being conveyed becomes more known, especially to
the avid sports fan. Wasow (1997a; 1997b) actually offers an
alternative explanation for this example that involves the
relationship between the verb and the PP. In order to maintain
the discussion on discourse factors, however, that explanation
will be considered in more detail in chapter 3. An alternative
interpretation that considers the strength of the information
in the NP relative to that of the PP favors the given-new
ordering hypothesis. That is, in the context of a ballgame,
relative to the name of the batter, the act of a batter coming
to the plate is assumed to constitute more known (i.e., given)
information to any of the fans watching or listening to the
game. Shifting the batter's name to the end of the sentence
would therefore comply with pragmatic theories of ordering that
predict the placement of given information before new
information (see Firbas, 1966 in Givon, 1988).
28
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Obviously, these examples and the contrasting
interpretations fail to provide support for one theory over the
other. Unfortunately, it is impossible to assess whether the
NPs and PPs in the above examples reflect given or new
information, since the structures do not appear in the context
of the discourse. Consequently, it is not possible to compare
the strength of either pragmatic theory on these examples
alone. These theories are, however, compared in Hawkins (1994;
submitted) in an analysis of 69 [V PP1 PP2] structures.
Hawkins adopted a measure by Givon (1988) in which the
referential distance of an item is measured by analyzing the
preceding 20 clauses of the discourse. Items with lower
referential distances represent more known (i.e., given)
information and those with higher referential distances
represent less known (i.e., newer) information. On Firbas'
(1966) view, items with lower referential scores should appear
earlier in an utterance. On the other hand, Givon (1988) would
predict that higher referential scores should appear earlier.
Hawkins (submitted) demonstrates that when the referential
distances of the two PPs differ (25 of the 69 cases), the
given-new account is correct 58% of the time and new-given
orderings account for only 42% of the data.
One of the challenges to theories pertaining to discourse or
weight is to understand the extent to which information status
and length are related. In example (3b), for instance, the
ungrammaticality may not only be attributed to the placement of
new before given information. Note that the PP is longer than
29
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
the NP. Behaghel's account that short material should precede
longer material, therefore, could also explain the
ungrammaticality. In addition, the longer PP contains an
indefinite article whereas the shorter NP refers to a definite
item. Thus, the definite status of the item may also
contribute to the length of an item. The more definite an
item, the more known it is to the hearer. Hence, fewer words
are needed to describe it. On the other hand, an indefinite
item might suggest that the item is newer to the discourse and
therefore requires more details to describe it; hence, it
appears as part of a longer phrase.
Hawkins' analysis addresses the confounding nature of weight
and information status by comparing the differences between the
lengths of the PPs with the information status predictions.
Thirty-eight (79%) of the 69 structures appear in the order PP1
PP2 when PP2 contains at least one or more words than PP1. To
compare the effects of weight with discourse, Hawkins found
that the referential distances differed in 30 of the 38 cases.
In the other 8 sentences, the referential distances were equal
so that the discourse predictions could not be tested. Of
those 30 eligible cases, however, Hawkins found that 25
appeared in given-new orders. On the other hand, only 5
patterned as new-given. Perhaps more interestingly, Hawkins
demonstrated that as the difference in weight decreased or
reversed (i.e., PP1 contained more words than PP2), the
presence of new-given orders increased. Thus, in the 18 cases
where PP1=PP2 or PP1>PP2, 15 appeared in the new-given order.
30
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
These patterns suggest that when the weight of the first
constituent exceeds that of the second constituent, the
presence of new-given orderings is most prevalent. In fact, of
the four sentences that met this weight criteria, all 4 placed
new material ahead of given material. As the weight
differences become equal and the weight of PP2 begins to exceed
that of PP1, the number of new-given sentences decreases and
the number of given-new orders increases.
From these findings, Hawkins (1994; submitted) concluded
that the pragmatic predictions are less successful than his
predictions for weight, which are described in more detail in
the section on Relative Weight. But he also indicated that the
correlation between given-new orderings when short items
preceded long items was due to the fact that given information
is likely to contain fewer words than is new information.
Thus, the pattern of new-given ordering was most prevalent when
the length of PP1 exceeded that of PP2 (i.e., long items
preceded short items). In sum, Hawkins (submitted) argues that
weight and the correlation between length and information
status, not pragmatic factors, drive ordering. This claim
differs significantly from functionalist theories of ordering,
where pragmatic factors are argued to be the primary factor.
Functionalism
Syntactic said Discourse Processes
A functional grammar approach (Dik, 1989; Siewierska, 1991)
to ordering recognizes the influence of both pragmatic and
syntactic process, and the interaction between them, on
31
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
ordering preferences. Dik's (1989) Principle of Pragmatic
Highlighting, for instance, highlights the influence of
pragmatics on ordering. Easily identifiable referents appear
prior to those items that are more difficult to identify. In
addition, the length (or complexity) of an item is strongly
correlated with a referent's ease of identifiability. Placing
less complex before more complex material "is in part a
consequence of the overlap between identifiability and
topicality" (p. 212) . If an item can be easily identified,
then the amount of lexical material necessary will be shorter
or less complex for that item than one that is more difficult
to identify, where complexity is defined by the LIPOC
(Language—Independent Preferred Order of Constituents)
principle in the following ways:
15a. clitic < pronoun < noun phrase < adpositional phrase <
subordinate clause;
15b. for any category X:X < XcoX;
15c. for any categories X and Y: X < X [sub Y]
(Siewierska, 1991, p. 212).
When an NP contains strongly identifiable information and is
short, it will be more accessible than an item that is less
identifiable and more syntactically complex. Specifically, the
accessibility of an item is defined in terms of its
referential, not relational, accessibility. Chafe (1987)
argues that the referential accessibility of an item can be
activated at one of three levels: active, semi-active, and
inactive stages. These stages are represented in short-term
32
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
memory, "a person's peripheral consciousness," and in long-term
memory respectively (Sievrierska, 1991, p. 156) . The more
active the referent, the more prominent, and thus more
accessible, it will be in the mind of the speaker. On this
view, as opposed to a more relational perspective where the
status of an item is determined by information shared by both
the hearer and the speaker, the level of activation an item
receives is based on the speaker only. Thus, in the exchange
between two participants, "Who broke the window?" "I broke
it/I did." The answer 'I' is considered to be a given referent
because it refers to the speaker, who is assumed to be quite an
active referent in the conversation. From a relational
perspective (Gundel, Houlihan, & Sanders, 1988), however, the
answer 'I' represents new information, since the hearer does
not share the pertinent information regarding who it was that
broke that window.
Following this notion of referential accessibility, and in
accordance with criteria (15a-c), unless the NP contains a
subordinate clause the NP would always be less complex than the
PP. Consequently, LIPOC alone provides no motivation for
producing shifted orders such as those in examples (7) and (9),
where the PP is intuitively more complex than the NP. This
kind of view supports the influence of both form and function
on linear ordering, a view in direct contrast to Hawkins
(1994), who argues that "whatever pragmatic-semantic content
Heavy NP Shift structures have has no impact on their
distribution in performance. Form alone tells the whole story"
33
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
(p. 184). The next section considers Hawkins' (1994) proposal
in detail.
Performance
Relative Weight
The preceding section raised an important issue concerning
the relationship between form and content. To achieve a more
comprehensive understanding of such a relationship, the precise
nature of form must first be determined. Hawkins' proposal
that form alone motivates shifting is similar to earlier
syntactic explanations in that it both defines the syntactic
weight of constituents and advocates a single factor
explanation for shifting. Crucially, it differs from the
former syntactic proposals in that it rejects the underlying
grammatical assumptions connected with most syntactic
explanations, and instead assumes a performance-motivated
explanation. Furthermore, his approach does not rely on the
syntactic weight, or complexity, of the NP alone to motivate
shifting. As outlined above, syntactic definitions that have
done so fail to provide a clear definition of weight that
adequately captures the data. They also limit themselves to
one part of the sentence, hence ignoring the intuitions of
Miller & Chomsky (1963) and Frazier (1985) that the complexity
of a sentence is associated with the amount of tree structure.
Maintaining this intuition, Hawkins (1994) proposes that the
relative weight between an *(P and its adjacent constituents
determine heaviness.
34
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
This proposal of relative weight is not a new one.
Stockwell (1977), for example, asserts that "any NP in the
predicate which is somehow 'heavy', at least relative to what
follows it, can be moved to the end of that clause" (p. 155) .
This approach resembles the previous syntactic explanations in
that it provides no specific measure for determining whether
the NP is sufficiently heavier than its neighboring constituent
is. Assuming again that the appropriate measure occurs at the
word level, whether a five-word difference induces shifting, or
if a two-word difference would be adequate, remains
unspecified. Hawkins' (1994) account of relative weight, on
the other hand, provides a specific measure by which heaviness
is assessed.
Defining the EIC principle
The Early Immediate Constituency (EIC) principle, a
performance-driven approach to ordering, guides Hawkins' view
of relative weight. EIC relies on syntactic weight to explain
word order, such that certain orders are predicted over others
by the parser's preference to encounter "abstract syntactic
structure" (i.e., all of the nodes in a particular Constituent
Recognition Domain (CRD) ) as quickly and efficiently as
possible (Hawkins, 1994, p. 106) . In order for the NP and PP
constituents to be recognized as quickly and efficiently as
possible in relation to the verb, heaviness is based on the
weight (measured by the number of non—immediate constituents)
of the PP relative to that of the NP. An NP of eight words and
a PP of two words, as in example (1), is most efficient in the
35
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
shifted [V PP NP] order. It allows the verb, PP, and NP to be
recognized quickly over the course of four words, beginning
with the verb and ending with the first word of the NP.
More specifically, the parser prefers those structures that
have the most efficient IC (immediate constituent)-to-non-IC
(word) ratios within a CRD, where an IC is immediately
dominated by the mother nodes and only by the projections of
the head. With regard to heavy-NP shift, the CRD includes
three ICs: the Verb, the NP, and the PP. The heaviness or
complexity of a structure is thus measured by dividing the
"number of ICs in a CRD by the number of words required for
recognition" (Hawkins, 1994, p. 233) of that domain. For
example, in (16a) the IC—to-Word ratio of the VP CRD is three
Immediate Constituents (the Verb, NP, and PP) over four words
(from the first word of the first constituent VP [explained] to
the first word of the third constituent PP [ to] ) , an IC-word
ratio of 75%a.
16a .
[explained [the tax e s^ Jto Jiml
vp crd 1/ 1 2/2 2/3 3/4
100% 100% 67% 75%
16b.
J a k e [ expl ai ned [all o f the facts r egardi ng t he i mpor t and expor t taxes]] [ to
I w ______________________________ Z\
V P C R D ^ ^ =44%
100% 100% 25%
36
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
I6c.
Jake^ [ e x p l a i n e d ^ [ t o J i m ] ^ [ a l l of t h e f a c t s r e g a r d i n g t h e import and ex port t a x e s ) ]
VP CRD 1/1 2/2 2/3 3/4
100% 100% 67% 75% =8«%
Structures with a lower IC-word ratio are heavier, or more
complex, than those structures that obtain higher IC-word
ratios. Holding the PP constant and increasing the length of
the NP, for instance, results in a much more complex structure.
The structure in (16b) obtains only a 25% IC-word ratio, a much
more complex structure than the 75% in (10a). The NP in (16a),
relative to the weight of the PP, is significantly lighter,
since both the NP and PP contain only two words. In (16b), the
difference between the length of the NP and the PP is 8 words
and is therefore a much heavier structure than (16a). Hence,
shifting the NP of (16b) to clause—final position results in a
much less complex structure, obtaining a 75% ratio in (16c).
Directing attention to the relative weight of both the NP
and the PP, rather than to properties of the NP alone, the
shifting of NPs that are lighter than their adjacent PPs is not
so problematic. On this basis, one—word NPs such as those in
(7c) and (9b) are heavier than they would be if they were in a
structure that contained significantly more words in the PP.
EIC predicts, in this case, that shifting the NP of a structure
containing a one-word NP and a three-word PP (as in 9b), is
more or equally preferred to a structure containing a
difference of three or more words, since the ratios for shifted
orders become progressively worse as the length of the PP
increases. In the former case, when EIC ratios for both the
37
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
shifted and basic orders are close (e.g., just a two-word
difference), "there is little basis for choosing between them,
and not much is at stake if the slightly less-preferred order
is selected" (Hawkins, 1994, p. 85). Under these
circumstances, the cost for speakers to compute the difference
between close scores would be greater than if they were to
produce a slightly less optimal order.
Hawkins' SIC theory of relative weight differs in two ways
from the other theories presented thus far. First, it
considers the weight of the NP in relation to that of the PP
rather than concentrating solely on properties of the NP.
Second, performance principles, rather than UG principles,
motivate it. An obvious advantage of this approach is that it
does not have to subscribe to grammatical rearrangement rule
criteria. Moreover, it compares constituents, in this case the
NP and the PP, which comprise grammatically coherent
categories. This performance-motivated principle derives from
psycholinguistic evidence that language users are equipped to
comprehend and produce language on-line rapidly and efficiently
and from the notion that competence is not completely
independent of performance. In fact, Hawkins (1994) rejects
Chomsky's definition of the competence/performance dichotomy
(what he calls Chomsky's assumption of pure acceptability in
which performance has no impact on competence) on the grounds
that grammars have responded to performance processing
considerations. EIC is a performance principle that drives
38
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
linear ordering preferences and is therefore "not ultimately a
grammatical phenomenon" (Hawkins, 1994, p. 110) .
In svim, the proposals described above overlap at several
levels. Whereas some are grammatically motivated, others are
explained by functional or performance factors. In addition,
specific explanations (e.g., syntactic) vary depending on their
underlying theoretical assumptions (e.g., motivated by
grammatical, functional, or performance principles).
In reviewing the proposals that attempt to explain heavy—NP
shift, we have observed two important issues. First, we have
seen the difficulty that grammatically motivated proposals have
in meeting criterion (la) . A rearrangement rule must include a
grammatically coherent and definable category, and the fact
that a heavy-NP is not a grammatically coherent category raises
problems for grammatically motivated proposals. Second, and in
partial consequence to the first, several of the proposals have
alluded to the fact that one factor alone cannot adequately
describe all the data with respect to heavy—NP shift
structures. The literature therefore does not provide an
adequate account of the circumstances under which heavy-NP
shift occurs. On the one hand, it is unclear whether heaviness
is based on properties of the NP alone, though corpus data by
Hawkins (1994) seems to suggest it is not. On the other hand,
the precise interaction of factors such as weight, prosody, and
discourse factors is left unspecified. The next section
considers the detailed relationship between grammar
(competence) and performance-based assumptions.
39
with permission of the copyright owner. Further reproduction prohibited without permission.
II. COMPKttNCK/BEKBXmnUICte
Another dimension over which accounts of heavy-NP shift vary
involve the extent to which shifted and unshifted sentences are
generated by rules of the competence grammar or are merely the
result of processing (i.e., performance) preferences. As noted
above, most linguists have assumed that shifting phenomena are
grounded in the syntactic, phonological, or pragmatic component
of the grammar of the speaker. In contrast, psycholinguists
tend to place the locus of shifting in the nature of the speech
production process (Bock, 1982; Bock, 1987a) . The following
section considers in more detail the relationship between
competence and performance, suggesting that performance-related
explanations such as relative weight account for both the
rearrangement data as well as the grammatical!zed basic order.
Changing Constituent Structure: Rightward vs.
Leftward Movement
Rightward. Movement
Within the grammatical literature, two principal analyses
for changing constituent structure have been proposed for
heavy-NP shift; a rightward movement and a leftward movement
account. More commonly, heavy-NP shift is defined as a
rightward moving rule that "optionally moves a complex NP to
the end of the first sentence„.which contains it" (Ross, 1967;
see also Postal, 1974), as illustrated in Figure 2. NPi ('the
famous detective from Belgium') moves to the end of the VP from
its original V' position. Rochemont & Culicover (1990)
40
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
stipulate that a rightward moving phrase must be governed by
the same constituent as in its original site (i.e., the one
that governed the site of its trace)'.
Figure 2: Tree structure of a heavy-NP shift sentence
IromBelgnm.
tot h e g o e c t
(Haegeman, 1991, p. 383).
Leftward. Movement
An alternative analysis recognizes leftward movement of
phrases in which light predicates are raised to the left
(Kayne, 1994; Larson, 1988). Unlike the rightward accounts,
these proposals do explain the presence of both basic and
shifted orders. Larson (1988) proposes the leftward movement
account because it explains why 'heavy' NPs are restricted from
undergoing rightward movement in double object constructions as
in example (17). Larson (1988) asserts that either the verb
moves leftward by itself. Or, through reanalysis, additional
material such as the PP can accompany the verb as it moves
leftward. Reanalysis only applies under the condition that the
verb has unused arguments available. If it does, then V' is
reanalyzed and can be raised as V unit. In example (17) , the
verb 'send' has two available arguments. However, according to
Larson, the NP-trace and the NP 'a letter' use up the
arguments, so that reanalysis cannot apply to 'send e a
41
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
letter.' With respect to basic and heavy-NP shift, or more
appropriately, light predicate raised structures, leftward
movement without reanalysis results in the production of the
basic order. With reanalysis, however, the shifted order is
derived. For example, in the deep structure of example (18),
without reanalysis, the verb moves left into the empty verb
position, creating the basic order structure ’I gave everything
that he demanded to John.' In the phrase 'give to John,'
however, an argument is available. Thus, reanalysis can apply
and the entire phrase can undergo raising, resulting in the
shifted structure 'I gave to John everything that he demanded.'
Although the presence of both structures is accounted for, the
motivating circumstances under which reanalysis should apply is
lacking. Like the rightward movement accounts, this account
suggests what would happen for the shifted order to appear, and
it would be clear that the shifted order could not apply if
sufficient arguments were unavailable. However, in the case
that sufficient arguments are available, it is not the case
that shifting always and/or necessarily results. The question
remains as to the conditions under which shifting might
actually apply.
17. *John sent a letter every musician in the orchestra.
(Larson, 1988, p. 355) .
18. I e everything that he demanded give to John.
(Larson, 1988, p. 348) .
In another leftward movement account, Kayne (1994) agrees
that heavy-NP shift is not the appropriate name for this
42
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
structure, but instead is a case of light predicate raising.
However, he argues against Larson's view of leftward movement
because it violates the antisymmetry10 requirement of the Linear
Correspondence Axiom (herein LCA) . It does so by placing the
predicate phrase and NP in a sisterhood relationship that would
be acceptable if it were a relationship of adjunction. This
relationship is not, hence, violating the requirement. Unlike
Larson's theory, in which both the verb and, through
reanalysis, the PP undergo movement, Kayne (1994) argues for
more of a scrambling approach in which leftward movement
applies to the PP without the verb.l*
Kayne indicates three advantages to his proposal. The first
is an advantage of leftward accounts in general which account
for the ungrammaticality of sentences like (19), which are
typically explained via Ross' right roof constraint.
Additional advantages to Kayne's proposal, in particular, are
that it can explain why rightward adjunction does not apply in
Dutch and German. Leftward movement also restricts a
constituent from intervening between a preposition and its
complement, hence the ungrammaticality of shifting objects of
prepositions, as in (20).
19. *The fact that John gave to Bill is irrelevant all his
old linguistics books.
20. *John was talking to about linguistics one of my very
oldest friends.
(Kayne, 1994, pp. 72-73) .
43
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
In sum, both rightward and leftward accounts assume a change
in constituent structure, a sufficient condition for a
grammatical rearrangement rule. Following the assumptions that
syntactic structure underlies both of these movement accounts,
Heavy-NP shift is necessarily a grammatical rearrangement rule.
Under different theoretical assumptions, heavy-NP shift may or
may not be considered a rule. Assuming flat or discontinuous
structures, for example, does not entail the presence of a
grammatical rearrangement rule (see Hawkins, 1994 for a review
of McCawley, 1982; McCawley, 1987; and O'Grady, 1987), and
leaves the possibility of a pure performance account available.
In addition, if we deviate from strictly grammatical
assumptions, we find strong evidence for a performance-
motivated explanation of heavy-NP shift. The next section
explores the plausibility of a performance approach in relation
to grammar.
Performance in Relation to Grammar
Hawkins' (1994) account of relative weight considers the
likelihood that performance processes could drive ordering
rearrangements like heavy-NP shift, a possibility not allowed
by Chomsky's (1965) strict competence/performance dichotomy.
In Chomsky's view, the grammar determines all orderings and the
performance module is responsible for judging only the
acceptability of the orders delivered to it. The notion that
linear orders are potentially induced by performance factors,
but not grammatical conventions, in effect, questions the
44
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
extent to which heavy-NP shift is motivated by grammatical
rules in addition to performance.
A performance-motivated approach is plausible if we reject
Chomsky's (1965) standard theory of generative grammar. In
this standard theory, the grammar serves a prominent role in
determining ordering, but the performance module is responsible
only for ranking the preferences (see 21a). Directly related
to heavy-NP shift phenomena, it is likely, as stated in (21a),
that the grammar delivers the constituents V, NP, and PP as
ordered representations. For example, it could deliver the
basic order and the rearranged [V PP NP] order to the
performance module, which would then rank the preferences. On
this view, if the NP were relatively long in comparison to the
PP, the performance module would rank the shifted order higher
than it would the basic order.
21a. The grammar orders (i.e., both arranges and
rearranges) the constituents before delivering them to
the performance module for ranking preferences only.
Hawkins (1994) mentions two additional ways that performance
can interact with the grammar.
21b. The grammar is responsible only for delivering the
constituents to the performance module, which then
orders and ranks them.
21c. The grammar arranges the constituents, but the
performance module can "potentially rearrange and
convert a linearly ordered structural representation
45
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
provided by the grammar into one with a different
linear ordering" (p. 89) .
Example (21b) indicates that it is plausible to treat the basic
[V NP PP] order as a free order, such that the grammar only
delivers the constituents V, NP, and PP. At that point, the
performance module would select the logically possible orders
(which would all be grammatical) , and then rank them
accordingly. Alternatively, Hawkins argues that if performance
can arrange free orders as in (21b), then it should also be
able to rearrange grammaticalized basic orders (21c) . The
grammar could thus feasibly deliver the [V NP PP] basic order
to the performance module for rearrangement. If the NP were
significantly longer than the PP, for example, it would be the
responsibility of the performance module to rearrange the
orders in accordance with performance ordering predictions set
forth by Hawkins' EIC principle.
This third possibility logically derives from the fact that
in free word orders performance factors are already responsible
for both arranging (i.e., selecting) constituent orders and
assigning the preference rankings for those orderings.
Moreover, the performance module is responsible for arranging
and selecting the "IC orderings that are demonstrably free"
(Hawkins, 1994, p. 89) in English (e.g., PP1 PP2 orderings)12, a
language with highly fixed word orders. Consequently, Hawkins
argues that performance may also be able to rearrange the basic
orders assigned by the grammar, thus providing the performance
module with a much more extensive role than it would have in
4 6
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
(21a) . But if it is indeed plausible for a performance module
to rearrange a grammatically conventionalized order such as [V
NP PP], we must consider the interaction between grammar and
performance for basic and shifted orders. Distinguishing the
influence of the grammar and performance modules on ordering is
a difficult task for both shifted [V PP NP] and basic [V NP PP]
orders because EIC ultimately influences ordering in both the
grammar and the performance modules (Hawkins, 1994, p. 89) .
(Z-r-anwna i-i 7 i xa iri n n o f the [V NP PP] Order
Hawkins (1994) argues that the basic [V NP PP] order, for
example, has been grammaticalized in response to performance
data because it is in general the most optimal order. First,
in English, "direct object NPs—are significantly shorter than
PPs" (p. 89). Second, in order to recognize a CRD (the
immediate constituents necessary to identify the phrasal mother
node) quickly and efficiently, the shorter constituent should
appear closest to the verb. It logically follows that the most
optimal order is the one in which the NP precedes the PP.
Hawkins' (1994) text counts, in fact, illustrate this pattern:
The NPs averaged 2.5 words, the PPs 4.6 words, and 458 of 480
orders appeared in the [V NP PP] order (p. 89) . [V NP PP] is
not only the most optimal order, but it is also more frequent
than is [V PP NP] .
Although it is reasonable to argue that grammar and
performance interact in such a way that grammars adopt "basic
orders as conventionalizations of the most optimal and most
frequent orders in performance" (Hawkins, 1994, p. 92), these
47
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
data alone do not prove that [V NP PP] has been
grammaticalized. In free orders, for example, performance
would result in the same data. Hawkins' performance ordering
theory would make similar predictions: Orders with the most
optimal ratios will appear more frequently than orders with
less optimal ratios in the unmarked case (in the marked case,
orders will be selected/preferred "in direct proportion to the
magnitude of their ratios" (p. 84)). Initially, the data
follow such a pattern, but closer analysis reveals that a
significant amount of shifting in Hawkins' text counts does not
occur until the NP is greater than the PP by four or more
words, hence, arguing for the grammaticalization of the [V NP
PP] order. The performance ordering theory predicts that in
order to maintain optimal ordering preferences, the shifted
order should occur in performance data (as represented by text
frequency counts, on-line psycholinguistic experiments, and
perhaps, acceptability judgments) as soon as the NP is greater
than the PP by just one word. Consequently, [V NP PP] is not
merely the most frequent arrangement of free word orders;
rather, it must be a grammaticalized order: "if there were no
grammaticalization of [V NP PP], we would not expect this
delayed activation of EIC in performance" (Hawkins, 1994, p.
90) .
Grammar- and. Performance in Hca-ry-HP Shift
If we consider that grammars are shaped not only by "highly
general grammatical principles," but also by "highly general
principles of language performance" (Hawkins, 1994, p. 17), as
48
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
we have just illustrated, it follows that a basic order such as
[V NP PP] may not always be most optimal in language use. In
the 10—15% of the cases where the NP is actually greater than
the PP, it is often beneficial to rearrange the orders so the
EIC advantages that motivated the basic order in the first
place can be maintained. Again, both grammar and performance-
motivated rearrangements can serve the purpose of improving EIC
ratios when the lengths of the PP and NP "[go] against the
aggregate for the great majority that has led to the
grammaticalized basic order" (p. 91) . Compare, for example,
the EIC ratios for basic and shifted order structures for
heavy-NP shift with those for sentential extraposition. The
difference between the basic order of (16b), and the shifted
order in (16c), repeated below for convenience, illustrates
that shifting a basic [V NP PP] order with a long NP results in
better EIC ratios. The aggregate ratio for the basic order
structure in (16b) with a 10-word NP and a 2-word PP is only
44%; but when the NP is shifted in (16c), the ratio
significantly improves to 86%. Importantly, it is not always
the case that shifting must occur when the length of the NP
exceeds that of the PP. As discussed above, heavy-NP shifted
structures appear in Hawkins' data, for the most part, when the
length of the NP is four words greater than the length of the
PP, resulting only in a 5% shifting rate and maintenance of the
basic order in 5-10% of the cases. In other words, despite EIC
predictions, the processing load must not be so great when the
NP is only one or two words greater than the PP, that the
49
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission
grammaticalized order cannot be maintained until the difference
is significantly larger.
EIC ratios are also improved when sentential extraposition
is applied. In example (22), extraposing S' (see 22b) results
in a much improved EIC aggregate ratio of 100% over its basic
order counterpart in (22a), which has only an aggregate ratio
of 70% (Hawkins, 1994).
16b.
__ [all of the facts for the import and export taxes]] [to Jim I
I ^ P P
vp crd ^ 2/2 3/12 =*|4%
100% 100% 25%
16c.
Jake^ [explained w [to Jim] ^ [all of the facts for the import and export taxes]]
Vp CRD 1/ 1 2/2 2/3 3/4
100% 100% 67% 75% =M%
22a.
[■That Bill was frightened] [surprised [Maiy]]]
s 5 I _______________________ r I I n p [
S CRD 2 / f / o VP CRD ^ =70%
22b.
s [ft ,p [surprised ^ [Mary] ^ [that Bill was frightened]]]
s c 8D 5S « , V P C R D
It is noteworthy that as the CRD increases, the extraposed
structure will not be as likely. For example, if the NP 'Mary'
were extended to a longer phrase such as 'the woman who loved
him' the IC-to—Word ratio would be 70% for (23a), as it is in
(22a). In (23b), on the other hand, the VP constituent
recognition domain would attain a much lower IC-to-Word (3/7)
ratio than in (22b), resulting in an aggregate ratio of only
50
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
71.5%, a ratio that is only slightly better than that of the
basic order in (23a). Erdmann's (1988) sentential
extraposition data, reported in Hawkins (1994), support this
observation. Of 385 sentences, a structure with a sentential
subject and an adjective without postmodification appears as an
extraposed structure in 94.8% of the cases. But when the
adjective is postmodified, the extraposed structure occurs in
only 76.1% (51 of 67) of the cases. EIC ratios are typically
improved by extraposition, but as the preceding examples and
Erdmann's data illustrate, the rearranged structure does not
achieve as strong a preference from EIC as the length of the NP
increases relative to the length of S', because the overall CRD
takes longer to identify. These examples therefore illustrate
that performance is responsible for motivating both
grammaticalized and free order structures.
23a.
sty [That B ill was fti ^flenadj^surprised^the woman who low ed him[ U.
. 2/5 *' 2/2 *
40% vpa(D 100% =70 %
23b.
s[It surprised^the woman who loved Mm]g.[that Bill was frightened]].
S C R D 1U K VPCRD 43% =7LS %
Smaaiar
In this chapter, I reviewed several proposals that attempted
to define the properties of heaviness that motivate heavy-NP
shift. In doing so, we observed that the proposals overlap on
at least two levels. Not only do they vary with respect to
51
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
whether the factors are syntactic or non-syntactic in nature,
but they also varied in terms of their underlying theoretical
assumptions. Thus, some were motivated by purely grammatical
principles, whereas others were grounded in more functional or
performance-based beliefs. At least one limitation of the
grammatical theories included the inability to provide a
descriptively adequate account of the available data. I
demonstrated, following Hawkins (1994), that this limitation
was due primarily to the fact that a heavy NP is not a
grammatically coherent/definable category. In addition, I
provided evidence that, depending on one's assumption (e.g.,
that structures are flat or discontinuous), heavy-NP shift is
not necessarily subject to a change in constituent structure.
Moreover, it was shown that performance principles, such as
EIC, can explain the existence of both basic grammaticalized
orders and shifted orders. This possibility and Hawkins'
(1994) support for a performance-motivated approach to ordering
warrant the investigation of heavy-NP shift factors from a
performance perspective.
One particular advantage of the performance approach is that
it is compatible with current psycholinguistic principles,
which place the locus of shifting in speech production
processes rather than in the syntactic, phonological, or
pragmatic component of the speaker's grammar. This approach
discourages the notion that grammar and performance comprise
distinct modules, and instead encourages an interaction between
the two. Furthermore, multiple factors combine to affect
52
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
language processing and production. This notion resonates with
the data presented earlier and Wasow's (1997b) claim, both of
which suggest that multiple factors can motivate heavy-NP
shift. The grammatical explanations summarized above fail to
account for the possibility that more than a single factor is
responsible for motivating the production of heavy—NP shift
structures. Hawkins (1994), for example, argues that form
alone explains heavy-NP shift, while Firbas (1966 in Givon,
1988) and Givon (1988) propose that discourse factors alone are
sufficient predictors of constituent ordering preferences.
Although each carries different underlying assumptions, they
may not be so incompatible. The functionalist account of
ordering (Dik, 1989; Siewierska, 1991), to the extent that it
acknowledges a relationship between an item's identifiability
(i.e., information status) and its syntactic complexity,
recognizes that some factors are closely related. For example,
newer information typically requires more words to communicate
it, so that identifiability and length are interacting.
Likewise, the longer an item, the more prosodic structure it
will contain.
The next chapter provides a framework in which discourse and
other factors, not syntactic complexity alone, contribute to
the efficiency of syntactic processing. These factors are
discussed further in light of current production models,
whereby semantic, syntactic, lexical, and phonological factors
all interact to influence the production of rearranged
structures such as heavy—NP shift.
53
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
The relatively short NPs in examples (6) and (7) illustrate
that the difficulty in defining heaviness is not limited to
grammatical principles, but to intuitive notions as well.
2 Further evidence that heaviness is characterized in
prosodic, not syntactic, terms is illustrated in Serbo-
Croatian. The PPs 'to Peter' and 'to that man' both branch
at c-structure, but only the latter branches at p-structure,
since prepositions combine with the first word of the NP to
form a phonological phrase (Zee & Inkelas, 1990, p. 374) .
J Wasow (1997b) provides further counter-evidence to prosodic
proposals arguing that phonological criteria are difficult
to measure since sentences that were never intended to be
read aloud appear in shifted structures: e.g., "send to
LISTSERV0DHDURZ1 the command GET TEX-PUB FILELIST". In
addition, Wasow points out noun phrases such as 'free will*,
which comprise only one lexical item, and hence, only one
phonological phrase. Consequently, the sentence, "ignore
completely free will" is an exception to the prosodic
account
4 According to Grice's cooperation principle, it is logical
that the focus would represent new information. If it
didn't, there would be no reason to actually utter it
(Grice, 1975 in Siewierska, 1991).
5 It is also a property that this information must be accented
(though accent alone is an insufficient means of identifying
focus; see Rochemont & Culicover for review) . For example,
(ii) is an appropriate response to the question in (i) ,
whereas (iii) accents more information than is actually
required, and so, does not appropriately answer the question
posed in (i).
i. Who likes Mary?
ii. JOHN likes Mary.
iii. JOHN likes MARY.
6 In (15a), the item is already introduced into the context,
but in (15b) , the item is not construable by the context at
hand. In cases like these, a further distinction between
contrastive and presentational focus is necessary.
' Rather than calculating non-ICs, Hawkins (1994) argues that
the number of words can be applied in the interests of
practical expediency as a shorthand for measuring the amount
of abstract structure present. Thus, more non-terminal
nodes in a structure also warrants more dominating nodes, c-
commanding structure, and sister nodes. In other words, as
the number of words increases so does the amount of abstract
(non-IC) structure.
54
with permission of the copyright owner. Further reproduction prohibited without permission.
8 The aggregate ratio of 86% obtains by taking the average of
each of the IC-word ratios that comprise the given CRD in
question.
9 This stipulation explains why sentences such as those in (i)
are ungrammatical.
i. Mary put the money on yesterday a table that was
sitting at the entrance to the hall.
(Rochemont & Culicover, 1990, p. 135)
10 According to Kayne's antisymmetry requirement, constituents
must asymmetrically c-command one another, such that c-
coramand is defined by him as the "first node up " (p. 7).
lx Kayne does not commit to whether or not the verb operates as
it does in Larson's work, or whether it patterns with the
complement as a nonverbal head as advocated in Pesetsky
(1995).
‘ z Hawkins notes however that the PPs are grammatically free in
English only if the preposition is not embedded within an NP
of the first PP as in the following phrase, '[through the
bam [of the farm]]np-'
55
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Chapter 3: A Psycholingotstxc Perspective of
Heavy-NP Shift
The heavy-NP shift phenomenon poses interesting problems for
researchers in language production. If speakers choose shifted
vs. basic order structures based on the relative length of the
phrases in the verb phrase (Hawkins, 1994), then they would
appear to have rather detailed quantitative information
concerning the phrasal contents of to-be-uttered sentences
during the stages of production in which syntactic structures
are chosen. Investigations of heavy-NP shift during on-line
speech production could therefore be very revealing about the
nature and time course of production operations. The first
part of this chapter reviews speech error data, which has
greatly influenced our knowledge of production processes.
Plausible factors that influence the spoken production of
heavy-NP shift are then applied to current production models.
Finally, the extent to which hearer and/or speaker needs are
accommodated in the guise of such models is evaluated.
X. R e s e a r c h M o m r a - T n ts S p e e c h P p o d v c t x o n M o d e l s
Efforts to study speech production processes include the
investigation of aphasic speech (Schwartz, Saffran, Bloch, &
Dell, 1994; see also Caplan, 1992, for further review) and the
use of computational models (Kempen & Hoenkamp, 1987; Steedman
& Johnson-Laird, 1980). Most important has been the link
between speech error studies and normal production processes.
Garrett's (1975; 1976; 1980; 1982) analysis of speech errors,
56
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
for example, greatly influenced work on production processes.
This influence is evidenced by much of the subsequent
production literature that followed from, or at least commented
on, his 1975 work (Bock, 1982; Bock, 1987a; Bock & Levelt,
1994; Dell, 1986; Dell & O'Seaghdha, 1992; F. Ferreira, 1993;
F. Ferreira, 1994; F. Ferreira & Clifton, 1986; Garrett, 1976;
Garrett, 1980; Garrett, 1982; Kelly, Bock, & Keil, 1986; Levelt
& Maassen, 1981; Levelt, 1989; MacDonald, 1993; Stemberger,
1985).
Most of this work has been conducted within the framework of
production models that identify three major stages in sentence
production: conception, formulation, and articulation. A
message (conception stage) is grammatically encoded at the
formulation stage, comprised of both functional and positional
levels of processing. At the functional level, lexical
selection and function assignment occur. Lexical selection
refers to the selection of the lemma that underlies the
conceptual item, not to the phonetic realization of these
items, and function assignment maps the message level
representations onto grammatical roles such as subject,
predicate, and object. These processes entail the retrieval of
major content words such as nouns and verbs (Bock, 1987a; Bock,
1990; Bock, Loebell, & Morey, 1992; Levelt, 1989). Positional
processes, on the other hand, associate grammatical roles with
particular syntactic structures and flesh out the phonology of
lexical items in preparation for articulation, such that the
lexical content along with its function assignments are ordered
57
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
into a set of word and morphological slots. Heavy-NP shift is
not directly addressed in these models, but as we shall see, it
makes sense to conceive of it as a late functional or early
positional process.
In these models of language production, the order of major
syntactic constituents may often be determined in the mapping
from conceptual to functional representation in an incremental
fashion (Bock, 1982; Bock, 1987a; Bock & Levelt, 1994; De
Smedt, 1994b; De Smedt, 1990; Kempen & Hoenkamp, 1987; Levelt,
1989). That is, words that are more accessible during the
utterance—planning stage by virtue of their semantics,
frequency, or some other factor, will be the first to be
incorporated into the speech plan and will therefore tend to
appear earlier in the sentence. For example, in the case of a
simple declarative sentence, assignment of the most accessible
noun or noun phrase to subject position determines the choice
of active or passive sentence structure and thus the locations
of the other sentence constituents (e.g., Bock, 1987b; see Bock
& Levelt, 1994; Kempen & Hoenkamp, 1987; Levelt, 1989, for
extensive discussion of such Incremental sentence
construction). The ordering of phrases constituting heavy-NP
shift structures, which place the [PP] or an adverb, rather
than the [NP], adjacent to the verb, however, seems to be more
related to positional level processes.
Dative alternation provides another much studied case in
which speakers must choose between two feasible orders; namely,
double object and prepositional dative structures, e.g., give
58
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
the dog a bone vs. give a bone to the dog. Bock and Warren
(1985) found that the more accessible of the two post-verbal
nouns tended to be placed in direct object position in recall
of dative sentences, indicating that the syntactic choice in
this case is semantically conditioned. De Smedt (1994b) has
applied this approach to heavy-NP shift, hypothesizing that
shorter phrases tend to become available for sequencing before
longer phrases during the production process, such that longer
NPs will tend to appear after more readily accessible PPs (De
Smedt, 1994b). This approach defines accessibility in terms of
number of words in the phrases, but an incremental
accessibility account could also be developed for some of the
other approaches discussed above. For example, more complex
syntactic structures may be less accessible than simpler
structures, prosodically complex structures may be less
accessible than prosodically simpler structures, and new
information may be less accessible than old information.
Speech error data
A review of Garrett's speech error data allows for a more
careful examination of the distinction between these two
levels, and the extent to which they constrain the production
of heavy-NP shift. Using speech errors that occur in a fluent
stream of speech, Garrett (1975; 1976; 1980) analyzed items
that were either "partly missing, ... added unintentionally,
[or appeared] in the wrong order” (p. 138). Following the
analysis, Garrett claimed that clear patterns of word,
morpheme, and sound exchanges support distinct levels of
59
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
processing at the formulation stage and ultimately provide
insight into the speech production processes that speakers tap
during the grammatical encoding process. In general, errors
involving whole word and phrase exchanges are ascribed to the
functional substage: They arise as a result of misassigning
entities such as nouns and noun phrases to equivalent syntactic
roles. For example, two nouns are exchanged in the sentence I
left the briefcase in my cigar (Garrett, 1980), and two NPs are
exchanged in the sentence I got into this guy with a discussion
(Garrett, 1980).
More specifically, the distinct patterns that obtain for
sound and word exchange errors provide helpful clues about the
processes that are accessed during the course of production.
First, the ratio of between—clause errors to within-clause
errors is smaller for sound exchanges (7%) than for word
exchanges (15%). Second, whereas sound exchanges and shifts
are similar both phonetically (i.e., they contain similar
phonetic features, stress exchanges with stress etc.) and
environmentally (word initial with word initial etc.), the word
form exchanges are not similar. In the sound exchange error,
1sbummer sipends,' intended to be 'summer stipends’ (Garrett,
1980, p. 183), for example, the exchange is phonetically and
environmentally similar. In the word exchange error, ' I have
to fill up the gas with oaf* (Garrett, 1975, p. 155) , on the
other hand, the words, gas and car, illustrate clear phonetic
differences despite the fact that they appear in a similar
environment. In addition, shifts do not preserve stress,
60
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
whereas word exchange errors do (e.g., 'Breakfast; is the
ltoeati.es of Champions' for the intended ’ Wheaties is the
breakfast of Champions'), suggesting that stress is assigned to
a position, not to a particular lexical item. These sound and
word exchange errors are indicative of a third difference in
terms of surface adjacency: Sound exchange errors are more
likely to appear closer together than word exchange errors.
A fourth difference involves the 'same form class'
constraint in which two words of the same grammatical category
(verbs and prepositions respectively) are exchanged: ’Older
men choose to tend younger wives’ and 'How much can I buy it
for you from.?' (Garrett, 1980, p. 191) . In between-clause
exchanges, sounds do not adhere to this constraint, but words
do. A further analysis of the 'same form class' constraint for
within-clause speech errors illustrates that violations are not
restricted to sound exchanges. Of the within-clause class of
errors, the following patterns obtained. Eighty-four percent
of the independent word exchanges (e.g., ’I have to fill up the
gas with car' ) conform to the 'same form class' constraint.
The combined form word exchanges, also known as stranding
errors (e.g., 'You have to square it facely'), however,
violated the 'same form class' constraint 70% of the time
(Garrett, 1975, p. 151), corresponding more closely to the
patterns of sound exchange errors.
The high occurrence of the 'same form class' constraint
violations among sound exchange errors may contribute to a
fifth difference between sound and word exchange patterns.
61
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Specifically, both sound and word exchange errors are highly
likely to occur within clauses, but sound exchange errors are
more likely to be limited to a phrasal boundary than are word
exchange errors. For example, the sound exchange error 'pons
and pats’ occurs within a phrase, but the word exchange error,
'This spring’ has a seat in it' occurs between phrases (Garrett,
1980, pp. 183 & 188 respectively) . The difference between the
pattern of sound exchange errors, shift errors (e.g., 'band
hadwriting' and 'They're only the ones that get...' Garrett,
1980, p. 203), and combined form word exchange (stranding)
errors (e.g., 'I went to get a cash checked'; 'You have to
square it £acely') for which parts of the words are exchanged
but the inflection (e.g., ed and ly) is left behind, and word
exchange error patterns suggests that two separate processes
(functional and positional) are involved in the mapping of a
conceptual message to phonetic form.
Additional support for Garrett's (1975; 1980) distinction of
levels is supported by the behavior of open and closed class
items. Stemberger (1985) highlights three of Garrett's
observations. First, closed-class items are less likely to
strand morphemes than open-class items. Second, closed-class
items do not participate in phonological errors. Third, stress
moves more with closed-class items than with open-class items
because the former participate more frequently in shifts (where
stress is shown to move with the item) than the latter.
On the basis of these error patterns and those described
above, Garrett argues for a model in which open-class lexical
62
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
items (lemmas) and their grammatical roles are first assigned
at the functional level. These roles are then translated to
mappings, by way of planning frames at the positional level
which are "enriched. . .with closed-class items" (Kempen &
Hoenkamp, 1987, p. 208), as illustrated in Figure 3.
Figure 3: A planning frame before open-class items are placed
into their slots at the positional level.
NP VP
DET N
THE
[ N ] [vpast 1 ON THE
(Bock, 1986a, p. 353)
In other words, closed class vocabulary items facilitate form
construction, mediating the representation of grammatical roles
at the functional level to the actual form at the positional
level. The patterns of closed-class versus open-class items in
errors and the preservation of stress in word exchanges, but
not shifts, for example, argues for the separation between the
lexical instantiation of an item at one level (functional
processes) and the serial ordering (positional processes) of
that item on another level.
In summary, three key differences (for further discussion,
see Dell & Brown, 1991; Kempen & Hoenkamp, 1987; Stemberger,
63
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
1985) between functional and positional processes arise from
the speech error patterns discussed above. First, phonological
form is pertinent to the ordering at the positional level,
whereas lexical instantiations of an item (lemmas) are accessed
at the functional level. Second, function words and
inflections are included in the planning frames at the
positional level. At the functional level, words are assigned
grammatical roles and phrasal membership. Third, positional
processes involve phrase—level ordering and functional
processes operate on a clausal level.
In contrast to Garrett's (1975; 1976; 1980) strict top-down
production model in which the functional and positional levels
are independent, Stemberger (1985) and Bock (1982; 1986c;
1987a; 1987b) advocate models that do not confine lexical
access, syntactic structure, or phonological form to processes
that must occur independently of one another. Instead, they
argue that these processes interact with one another, though
their underlying assumptions differ. Whereas Bock (1987b)
preserves the distinction between functional and positional
levels, Stemberger (1985) argues that separate processing
levels are unnecessary. In Stemberger's model, for example,
lexical items are activated in parallel with syntactic
structures and phonological form, with those receiving the
highest levels of activation winning out over alternative
lexical structures. Unlike Garrett's model, Stemberger
attributes speech errors to the multiple activation of certain
features or words at any given time. Specifically, the model
64
with permission of the copyright owner. Further reproduction prohibited without permission.
assumes that various slots containing certain syntactic
features exist for each sentence. Any given slot is already
coded for particular features: A slot for a noun will be more
strongly activated when a noun is accessed than when a verb is
considered for that slot. If an incorrect noun and a verb are
for some reason more highly activated than an intended noun,
the incorrect noun is more likely to appear in the target
noun's slot than is the verb (which is also incorrect). On
this basis, morpheme stranding errors occur, Stemberger (1985)
argues, because both the lexical items and morphemes are
"accessed separately and in parallel" (p. 163). The failure to
access either the lexical item or the morpheme does not
necessarily result in the failure to access the other. Thus,
blends, such as 'most disappointing out here is £r±ab., ' when
'fresh fish’ was intended, occur because two competing words
are accessed at presumably similar levels of activation. When
two items are similarly activated in this kind of model, both
’win out’ and an error occurs because neither has enough
activation to inhibit the other (Stemberger, 1985, p. 165).
Similarly, Stemberger attributes open and closed—class
errors to word frequency effects such that higher frequency
words are less likely to appear as errors because they
correlate strongly with higher activation levels. Thus,
closed-class items are less likely to strand morphemes because
they are higher in frequency, fewer in number, and more
restricted syntactically than their open-class counterparts (p.
166). For the same reason, closed-class items are more highly
65
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
activated for production and hence, are more likely than open-
class items to appear in shifts where one unit becomes
activated faster than a unit that should have appeared earlier
in the utterance. In addition, findings that high frequency
open-class items contain fewer phonological errors than low
frequency open-class items (see Stemberger and MacWhinney 1984,
as cited in Stemberger, 1985) led Stemberger to argue that the
lack of phonological errors in closed-class words is due not to
a difference between open and closed-class items. Instead, he
attributes it to a difference in the frequency of occurrence of
such words (Stemberger, 1985, p.167). Namely, the lower error
rate in higher frequency words is an effect of higher
activation rates (Stemberger & MacWhinney, 1984 and Miller &
Ellis (in press), as cited in Stemberger, 1985). Thus,
Stemberger (1985) argues that distinct processing levels are
unnecessary to account for speech error data.
Bock (1987a; 1987b), on the other hand, maintains that the
distinctions between functional and positional levels of
production are necessary. She argues that separate processing
levels can account for the fact that lexical substitution
errors are rarely phonologically related but are frequently
semantically related to the intended word, as well as the fact
that sound versus word exchange errors occur within, versus
between, phrases. In addition, experimental evidence presented
in Bock (1987a; 1987b) supports the separate processing level
assumption. In this experiment, she demonstrated that
phonological primes closely related to their targets displayed
66
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
an inhibitory effect on syntactic structures (e.g., actives vs.
passives). Unprimed words appeared earlier in structures than
did primed words. These findings contrast with those in Bock
(1986b) where only semantic, not phonological, priming
influenced syntactic structure, but Bock attributes the
difference between the two studies to the relationship between
the phonological primes and their targets. In the Bock (1987b)
study, the primes were more closely related to their targets
than they had been in the Bock (1986b) study. As a result,
Bock concluded that phonological accessibility and hence,
positional level processes do interact with and influence
processes at the functional level. The next section
demonstrates that this interactive component of a model can
more adequately represent the factors that constrain the
production of heavy-NP shift than a strict top down model like
that of Garrett (1975; 1980).
II. Heavy-NP Shift nr Pbcooctzch
Until recently (see Stallings et al., in press), there have
been no experimental studies of heavy-NP shift, so that the
most relevant production research was provided by studies that
have examined the ordering of words rather than whole phrases
(e.g., Bock, 1986c; Bock, 1987b; Bock & Warren, 1985; Kelly,
Bock, & Keil, 1986) . These studies of word order laid the
foundation for Stallings, MacDonald, & O'Seaghdha (in press) to
form specific hypotheses about the factors that constrain
heavy-NP shift in spoken language production.
67
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
To understand the hypotheses tested in Stallings et al. (in
press), this section begins with a review of some production
work by McDonald, Bock, £ Kelly (1993) that is relevant to
ordering. Next, the plausibility that animacy and thematic
role structure (both factors that typically constrain word
order in production experiments and are strongly associated
with grammatical role assignment) influences the production of
heavy-NP shift is considered. Then, we consider the
implications of the more intuitive length constraint in
production. Fourth, we examine evidence for the influence of
information status in production.
Semantic and Phonological Influences on Ordering
In one particular production study, McDonald, Bock £ Kelly
(1993) investigated both semantic and phonological influences
on word order in several constructions. The authors
manipulated the animacy, length, and stress patterns of the
nouns in active and passive sentences such as A farmer
purchased a refrigera tor/A refrigerator was purchased by a
farmer, in conjunctive phrases within sentences, The key and
the manager were nowhere to be found/The manager and the key
were nowhere to be found, as well as in isolated phrases such
as manager and key/key and manager. They examined the order of
the two nouns in recall of the sentences or phrases and found
robust effects of animacy, especially in the active/passive
sentences where the choice of phrase order affected the
assignment of grammatical roles. That is, participants tended
to place animate before inanimate nouns in recall. However,
68
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
animacy had little effect on the ordering of nouns within a
conjunctive phrases when the phrase was part of a sentence.
McDonald et al. concluded that animacy had little or no effect
on word order choices that did not entail grammatical role
assignment.
In contrast to animacy, the length and stress patterns of
the nouns had very little effect on the order of recall in any
of the conditions, but McDonald et al. did find some
sensitivity to length effects in an acceptability rating task.
For example, participants judged conjoined NPs to be more
acceptable when a shorter noun preceded a longer noun (e.g.,
book and refrigerator) than when this order was reversed,
consistent with other length effects found in acceptability
judgments (Cooper & Ross, 1975; Pinker & Birdsong, 1979). In
production tasks, however, semantic effects such as animacy
appear to play a role in the choice of word order, especially
if order entails assignment of grammatical roles, but there
seems to be only a very restricted role for phonological
effects such as word length. Though word and phrase ordering
are not necessarily guided by the same production mechanisms,
these studies of word ordering yield two important implications
for heavy-NP shift; one regarding animacy and the other having
to do wich length.
Animacy
First, there are clear effects of noun accessibility,
including accessibility modulated by animacy, on word and
phrase ordering choice that entail grammatical role assignment,
69
with permission of the copyright owner. Further reproduction prohibited without permission.
but not on word ordering choices that do not affect role
assignment. As heavy-NP shift is the rare case of phrase
ordering that does not affect grammatical role assignment
(i.e., the direct object NP retains this role whether shifted
or not), Stallings et al. (in press) predicted that shifting
would not be sensitive to animacy effects. In fact, in an
experiment that manipulated PP noun animacy (e.g., to John vs.
to work) and NP length on the spoken production of heavy-NP
shift, no effects of animacy obtained. These results confirmed
that the conceptual accessibility of factors (e.g., focus,
prototypicality, and animacy in De Smedt, 1994b; Bock & Warren,
1985; and McDonald et al., 1993, respectively) seems to be a
central component in determining word order only if it is tied
to grammatical role assignment. Findings that animacy did not
influence the production of sentence-internal conjunctive noun
phrases (McDonald et al., 1993, Experiment 2), but did affect
the production of active and passive sentences (F. Ferreira,
1994; McDonald et al., 1993) further support this view. Thus,
it seems that animacy only plays a role in assigning thematic
roles such as Agent, but not in the assignment of order if it
is divorced from such assignment (see McDonald et al., 1993).
Thematic Role Structure
On this view, the influence of thematic roles on the
production of heavy—NP shift is also expected to be weak.
Although production experiments show clear effects of thematic
role structure on ordering, the effects appear only in
structures that affect grammatical role assignment. F.
70
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Ferreira (1994), for example, showed that choice of active vs.
passive structures is modulated by participants' preferences to
place stronger thematic roles such as agents and experiencers
(as opposed to time, location or manner) in more grammatically
prominent positions (e.g., subject). These findings fellow
from the fact that thematic roles are represented
hierarchically and are related to grammatical role assignment
(see Primus, 1994).
These ties to grammatical role assignment strongly suggest
that heavy-NP shift is not constrained by thematic roles.
Further evidence for this hypothesis is provided in Stallings,
MacDonald, & O'Seaghdha's (in press) production task. In that
study, animacy was confounded with thematic roles: the animate
PPs consistently expressed a Goal role, whereas the inanimate
PPs contained more adjunctive (and less prominent) roles of
Time, Location, and Manner. Considered alone, the thematic
role hierarchy predicts that PPs expressing more prominent
roles such as Goal would be more likely to appear in shifted
structures than PPs expressing less prominent roles. If
thematic roles were constraining heavy-NP shift, effects of
animacy should have obtained. To be clear about these effects,
the authors compared the argument/adjunct status of the PPs
with respect to shifting rates. All of the inanimate items
were adjuncts, and of the inanimate items, approximately 75%
were arguments. No effect of argument status was found. In
fact, argument and adjunct PPs showed very similar shifting
rates— 15.4% and 17.5%, respectively. This argument status
71
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
factor in combination with the null result of animacy suggests
that thematic roles are also unaffected by phrasal ordering in
heavy-NP shi ft-
Length
The second important implication for heavy-NP shift is that
word length effects are weak or nonexistent in studies of word
ordering. The evidence that word length has little effect on
word order in sentences and phrases seems on the face of it to
be inconsistent with heavy-NP shift, which by definition
involves a substantial role for phrase length or relative
phrase length in the choice of shifted vs. basic syntactic
structure (Hawkins, 1994; Kimball, 1973; Wasow, 1997b).
However, the processes involved in weighing the 'heaviness' of
phrases need not include an assessment of the length of
individual words. Thus, Zee & Inkelas (1990) stipulated that
full phonological information is not available to the decision
making process. Likewise, Sternberg, Monsell, Knoll, & Wright
(1978; see also F. Ferreira, 1991; Wright, 1990) found that the
number of words in a speech plan, rather than the length of the
words, determined production latency. In addition, Wheeldon
and Lahiri (1997) showed that initiation time for short
prepared sentences of fixed length is determined by the number
of phonological words1 (see Nespor & Vogel, 1986) . In other
words, the ordering of phrases within a sentence may be
sensitive to phrase length defined in word or phonological word
units, even though the order of words within phrases is not
sensitive to the length of those units.
72
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Given this clear need to relate previous phrase length and
word order production studies to phrase ordering processes in
heavy-NP shift, Stallings et al. (in press) manipulated NP
length and held the PP constant. In doing so, they were also
testing one of 3 logically possible ways to consider Hawkins'
relative weight proposal. They predicted that if weight were
influencing phrasal ordering processes, the percentage of-
shifting should increase significantly in the long NP
condition. The details of the experiment appear in chapter 4,
but in summary, participants did shift long items significantly
more often than short items. Such results would provide clear
evidence that 1) the task was capable of adequately measuring
shifting, and 2) the choice of the heavy-NP shifted vs. basic
order (NP-PP) structure during speech production is
constrained, at the very least, by the length of the NP in the
verb phrase.
In two—stage interactive incremental models of sentence
production, phrase ordering should occur at an early stage,
before the phonological properties of words become fully
available, so that it should be sensitive to aspects of
syntactic—prosodic planning that precede phonological
retrieval. This implies that the "heaviness" of a long noun
phrase must be assessed in terms of conceptual, syntactic, or
prosodic complexity rather than in terms of the length in
phonological units of the words that comprises it. In support
of this conclusion, initiation times of prepared sentences are
sensitive to syntactic complexity as indexed by number of
73
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
words, but not to word length per se (see Sternberg et al.,
1978).
The placement of length effects in production is also
compatible with the kind of parallel incremental model
described by De Smedt (1994b), in which the conceptualizer,
formulator, and articulator "operate independently and in
parallel on different fragments of an utterance” (p. 3) . He
hypothesizes that shorter phrases as well as more conceptually
accessible" ones tend to become available for sequencing before
longer phrases during the production process. On this view, a
large HP, placed adjacent to a small PP, would be less
accessible than the PP, thus resulting in the greater
probability of a shifted order being produced.
An NP of similar, or equal, length to that of a PP, on the
other hand, would cause the production mechanism to consider 1)
the accessibility of factors other than length (e.g., verb
type, adjacency, pragmatics, prosody), and/or 2) rely less on
accessibility, focusing instead on other factors that
”determin[e] the order in which conceptual fragments are passed
to the formulator" (De Smedt, 1994b, p. 12) . The possibility
of it doing both 1 and 2 is reasonable given that the second
option may be confounded with the first. That is, in focusing
on other factors to motivate ordering, as De Smedt argues, it
is difficult to know whether or not the production mechanism is
considering the accessibility of those constraints. Whether
accessibility or the influence of other factors, independent of
accessibility, regulates order, the key point to be made is
74
with permission of the copyright owner. Further reproduction prohibited without permission
that additional factors can constrain the production. The next
section, in fact, demonstrates how additional factors such as
lexical and referential accessibility constrain the placement
of given-new ordering.
Effects of Lexical and Referential Accessibility on
Given-New Ordering
Unlike length, pragmatic factors are based less on the
actual form of a structure than on the content (or discourse
level information) that might influence the form of a sentence.
Bock & Irwin (1980) demonstrated this pattern in a production
study that showed the influence of both referential and lexical
accessibility on given-new ordering, where givenness was
defined as "a definite referring expression or a proper name
with a coreferential antecedent in an immediately preceding
sentence" (p. 467). Thus, information that had been recently
mentioned prior to the target utterance counted as given
information.
In a question-answer recall task, participants were more
likely to recall answers that placed items conveying given
information (when applicable) earlier in a sentence than
constituents that conveyed new information. For example,
participants were asked to recall the sentence, The xanchex
sold the hoxse to the cowboy, following the statement and
question sequence in (24a-24d) . The frequency of correct
recall for the sentence decreased in descending order, with
(24a) receiving the most number of correct responses, and (24d)
the least: The material in (24a) provides a clear referential
75
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
antecedent that is also lexically identical, but (24d)
contained neither.
24a. A rancher had a horse that kept running away. What
did the rancher do?
24b. A rancher had a stallion that kept running away. What
did the rancher do?
24c. A rancher received an inquiry from a cowboy about
something he needed for his act. What did the rancher
do?
24d. A rancher received an inquiry from Roy Rogers about
something he needed for his act. What did the rancher
do?
The placement of referentially more accessible items before
less referentially accessible items (and the strengthened
effect when the item was lexically identical) suggests that
”[t]he conceptualization level thus provides the intended
referents with access to the linguistic system, while the
assembly level retrieves and arranges the lexical items chosen
to convey the intended referents within the context of the
discourse” (Bock & Irwin, 1980, p. 468). In a different
sentence recall task, prompted by a noun instead, participants
were more likely to place the prompted word (tree or
lumberjack) earlier in the utterance when attempting to recall
the sentence: 'The falling tree crushed the lumberjack.' These
findings provide evidence that both referential and lexical
accessibility contribute to the effects of given-new ordering
(Bock & Irwin, 1980, p. 480).
76
with permission of the copyright owner. Further reproduction prohibited without permission.
These results also correspond with Clark & Haviland's (1977)
argument that the "given-new contract" fulfills the function of
efficient communication processes. Items that are
referentially and/or lexically accessible (because they have
previously been produced, and hence activated) will appear
earlier in an utterance, and as a result will leave more room
in memory for additional processing of later material. Due to
the accessibility and activation of lexical items and
referents, the presence of conversational context should
increase the extent to which speakers produce given before new
information. Speakers frequently begin an utterance with given
information, and then continue their conversational turn with
an assertion (i.e., new or important information), as the form
of the message they want to convey becomes more readily
available and hence, accessible. On this view, the early
production of given information is, in general, more accessible
to the speaker and therefore speeds up the processing of the
items. In the absence of context, however, lexical
accessibility is bound to be weak.
The early reference to given information, therefore, allows
the speaker to free up memory for processing the new
information to follow, but it also allows the listener to more
quickly attach new information to referents stored in memory.
Without the reference to the given information, for instance,
Haviland & Clark (1974) suggest that the listener must utilize
several additional processes in order to comprehend the
utterance. The listener must either find a way to link the
77
with permission of the copyright owner. Further reproduction prohibited without permission.
utterance to previous input, start over as if all information
was new, or reassess what in the utterance was given and what
was new (p. 518) . Consequently, the given-new contract
suggests that both speakers and listeners would benefit from
given-new ordering. It also suggests that discourse processes,
like weight, can affect the processing efficiency of heavy-NP
shift orders.
Two potential critiques of the given-new contract include
the following. First, it is not entirely consistent with the
data that showed new—given ordering preferences in 48% of the
sentences that Hawkins (submitted) analyzed. Second, placing
given information early in an utterance contrasts with the
argument that salient information defined only as important and
at the forefront of a speaker's attention appears early in a
sentence (Halliday, 1967, cited in Bock, 1982). If an item is
salient in the mind of the speaker, it does not necessarily
follow that it is given information, especially from the
perspective of each of the conversational participants.
On similar grounds, it initially contradicts the reason that
Johnson-Laird (1968, cited in Bock, 1982) provides for the
finding that participants consistently chose to describe a
display including two colors next to one another, with the
sentence 'After the red, there is some blue, ' rather than with
the sentence 'After the blue, there is some red. ' Johnson-
Laird argued that the former sentence was used to describe the
display because the larger red item was more perceptually
prominent in the mind of the speaker and therefore the more
78
with permission of the copyright owner. Further reproduction prohibited without permission.
important item. If perceptual prominence defines givenness,
however, Johnson-Laird's proposal might actually conform to the
given-new contract, such that the speaker refers to the
perceptually prominent material in order to create a link to
the upcoming material, and hence, facilitate processing.
Interestingly, Costermans & Hupet (1977, cited in Bock,
1982) interpret the same Johnson-Laird findings much
differently. They attribute the results instead to the idea
that the assertion, there is some blue is the more important
piece of information. The task at hand is thus one of defining
importance. Following Hornby's (1972, cited in Bock, 1982)
conclusion that assertions (new information) typically follow
presuppositions (given information), Costermans & Hupet define
assertions, not presuppositions, as the important material. On
that view, the smaller item required emphasis and so it was
more accepted in the assertion (i.e., later) position.
A principal difference between these two interpretations is
that Johnson-Laird explains the problem as a salience (or focus
of attention) issue, whereas Costermans & Hupet propose that it
follows from a preference to place given information prior to
new information. Bock (1982) attributes the main conflict
between the two interpretations to the fact that ordering
cannot be explained simultaneously by both information status
(i.e., givenness) and salience. If it could, new information
would be much less likely to attract a speaker’s attention.
In addition, part of the problem inherent in this conflict
seems to revolve around assumptions about the speaker and/or
79
with permission of the copyright owner. Further reproduction prohibited without permission
listener. From Johnson-Laird*s perspective, it is unclear
whether the speaker mentions the perceptually prominent
material first to benefit the speaker alone or the listener as
well. As stated above, the ordering could be attributed simply
to the salience of the item in the speaker's mind, or to the
fact that the speaker is using the perceptually prominent item
as a link to further material. From Costerman & Hupet*s point
of view, however, the underlying motivation seems to derive out
of concern for the listener. Emphasis of an assertion would
only be necessary if the speaker were attempting to convey
something to a listener.
The contrasts between these two theories raise questions
about the speaker and listener processes that may be involved
in heavy-NP shift. We saw that the positioning of lexically
and referentially accessible items early in an utterance can
ease the pressure being placed on the processor. Thus, in
addition to the dimensions on which heavy-NP shift varied in
chapter 2 (e.g., syntactic vs. non-syntactic factors and
competence versus performance-based explanations), this
evidence suggests that within the class of performance-based
explanations, accounts vary on yet another dimension. Namely,
whether the lightened processing load is more generally
motivated by accommodation to the needs of the hearer or by
inherent constraints on speaker performance (Wasow, 1997b).
That is, in choosing between a basic [V NP PP] and a shifted [V
PP NP] order, to what degree, if any, is the speaker
80
with permission of the copyright owner. Further reproduction prohibited without permission.
accommodating the listener? The next section considers this
question.
I I I . He a r e r v e r s u s S s e m x b i A c c o m o d a x io n i n H e a v y - N P S h i f t
This section reviews both comprehension and production
research pertaining to listener as well as speaker
accommodation. Ultimately, evidence in support of the premise
that heavy-NP shift is motivated more by the processes inherent
in production than by listener accommodation is offered.
Evidence in Favor of Listener Accommodation
In the previous section, Bock (1982) asserts that given-new
ordering occurs because speakers sometimes have not planned out
what they intend to say. This may explain why information that
should be salient (Bock, 1982; Osgood & Bock, 1977) (and would
therefore appear early in an utterance) actually appears later
in an utterance, following Bock & Irwin's (1980) evidence that
given information precedes new information. That communication
requires interaction, and in accordance with Grice's (1975)
principle that it should also be cooperative, some authors
(e.g., Clark, 1977) have proposed that speakers actually
consult a 'model of the listener' in their efforts to produce
comprehensible sentences. Doing so requires that speakers not
only consider, but also incorporate, the knowledge they share
with their listeners when formulating utterances.
Production research, in fact, asserts that a speaker's words
must be uttered in such a way that they are "digestible for the
listener" (Kempen & Hoenkamp, 1987, p.202) . They must also
81
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
serve to "persuade, enthrall, outrage, entertain, or otherwise
affect the response to communication apart from its literal
content" (Garrett, 1988, p. 89). In addition, Tanenhaus (1989)
reports findings by Clark (1979) demonstrating that speakers
choose utterances that are influenced by the context shared
with the listener, and Steedman and Johnson-Laird (1980) argue
that speakers only utter statements, questions, or requests
that will be heeded, or at least considered. Similarly, these
kinds of proposals and findings suggest that the speaker must
not only attract the listener's attention with an utterance,
but must also create a comprehensible string of words that will
presumably benefit the listener. A speaker's knowledge of the
listener is therefore crucial.
MacWhinney (1977) also argues that placing shared (or given)
information at the beginning of an utterance serves the purpose
of attracting the listener's attention. Once the listener's
attention is focused, the speaker can emphasize the new
information. This approach also facilitates processing for the
listener in the following way. Because the shared information
was previously raised, the second mention of that information
enables the listener to quickly identify it in memory, so that
the listener can then rapidly attach the new information to the
previously-known material (Haviland, 1974). The sequence of
sentences in (25) illustrates how processing is facilitated.
In (25a), listeners indicated via button press that they
understood the second (target) sentence 'The beer was warm'
much more quickly than they did in (25b) . The faster response
82
with permission of the copyright owner. Further reproduction prohibited without permission
times were attributed to the fact that, the NP ’the beer' had
been previously mentioned in the former, but not in the latter
(Haviland, 1974).
25a. We got some beer out of the trunk. The beer was warm.
25b. We checked the picnic supplies. The beer was warm.
(Haviland, 1974, pp. 514-515).
To ensure that this effect was due to a given-new ordering
strategy and not to a repetition facilitation effect, Haviland
4 Clark (1974) manipulated (25b) in a further experiment so
that the word 'beer' was repeated in both the context and
target sentence (see 25c). Once again, the time to understand
the target sentence was still quicker in what they called the
Direct Antecedent condition (e.g., 25a) than in the Indirect
Antecedent condition (e.g., 25c) even with the repeated noun.
Thus, they argue that the quicker processing in the Direct
Antecedent condition was not an effect of repeating the
critical noun (cf. work on repeated name penalties as discussed
in Gordon, Grosz, & Gilliom, 1993).
25c. Andrew was especially fond of beer. The beer was warm.
(Haviland, 1974, p. 516).
Evidence Against the Model of the Listener
Additional research showing that active sentences are more
easily understood than passive structures (Bever, 1970; F.
Ferreira, 1994) implies that the listener’s needs are
accommodated when such structures are produced, simply because
the memory load placed on the listener is less than it would be
for the latter. On this basis, listeners also benefit more
83
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
when structures appear in the order [less complexNP complexNP]
than when they appear in the order [complexNP less complexNP] .
Similarly, in example (26), (26a) is more difficult to process
than (26b). In (26a), the use of the reduced relative hinders
comprehension because the initial verb sequence is not the main
clause. Unlike the unambiguous version in (26b), (26a)
interferes with the listener's initial perception of the
sentence (Bever, 1970, p. 317).
26a. The horse raced past the bam fell.
26b. The horse that was raced past the bam fell.
Research on ambiguous structures like that in (26a) has
illustrated that processing is facilitated when either a) a
clausal structure matches the expectations developed by the
listener at the beginning (Tyler & Marslen-Wilson, 1977), b)
words that would otherwise be ambiguous follow a disambiguating
region (Frazier & Rayner, 1987), or c) the lexical
representation of a verb is biased more strongly towards one
interpretation over another in an ambiguous structure
(Trueswell, Tanenhaus, & Kello, 1993). Sentences that
incorporate these strategies appear to accommodate the
listener, since they result in less complex structures, which
in turn, facilitate comprehension.
Heavy—NP shift structures serve the same purpose. Several
authors (Bever, 1970; Frazier, 1985; Hawkins, 1994; Kimball,
1973; Miller & Chomsky, 1963) in fact, have argued (at least
indirectly) that the reduced complexity of heavy-NP shift
structures eases the processing load for the listener. In the
8 4
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
unshifted [V long-NP PP] order, a long NP can create a
substantial distance between the preposition and the verb, a
less than desirable situation for the comprehender. The
shifted order, on the other hand, creates only a short distance
between the verb and the head of the long NP (the head noun or
perhaps the determiner, in some syntactic analyses), hence,
easing the processing load for the listener. That the more
complex structures exist at all (and are produced on a not-so-
infrequent basis), however, casts doubt on a strictly
interpreted model of the listener. The existence of difficult
structures suggests that perhaps the speaker is not
accommodating the listener's needs at all.
Perhaps these complex sentences, though they make
comprehension more difficult, may somehow facilitate speaker
processes. Although evidence for the processes that would
allow such a conclusion is not easily transparent, it is clear,
that the lack of ambiguity in a language would neither
accommodate the listener nor the speaker. Eliminating
ambiguity would not only require an unrealistic number of
lexical and structural additions to the language but would be
very costly to both speakers and listeners: A unique meaning
for every word would require knowledge of an extraordinary
number of words, making language inherently more difficult.
Consequently, listeners and speakers are compelled to utilize a
variety of linguistic representations (e.g., the type of verb,
clarifying clauses, background context etc.) to facilitate
their processing of such difficult sentences. On this basis,
85
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
hearers who assume that speakers are following Grice's
cooperative principle can make inferences about the meaning of
a structure based on the knowledge shared with the speaker.
Dell & Brown (1991) question the extent to which speakers
access the model of the listener in producing comprehensible
utterances, arguing that speakers do not always and/or
necessarily consult the model of the listener. Although
speakers do accommodate the listener, in what Dell & Brown call
particular-listener adaptations, speakers do not consider
listener needs in generic-listener adaptations. Particular-
listener adaptations are instances in which speakers will alter
their speech by talking louder or slower, for example, due to
their beliefs about a listener's abilities to hear or
understand the speaker. This type of adaptation resonates with
register acquisition studies that have investigated both
babytalk (Ferguson, 1977) and foreignertalk (Andersen, 1990).
Generic-listener adaptations, on the other hand, are based more
on "production decisions [that] exhibit a bias toward content,
structure, or a manner of production that is, in general, easy
to comprehend" (p. 106). In the latter adaptation, a speaker's
decision to produce a particular structure is not motivated by
the needs of the listener, even though the listener's ability
to comprehend the structure may be facilitated by that
structure. The benefit to the listener is a consequence only
of a production system that "exhibits biases to produce
messages that are adapted to the average listener" (p. 107) .
86
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
When speakers choose to produce an easier word, it is the
frequency with which a word appears in the language that drives
the speaker’s choice, even though the word facilitates listener
comprehension. Because the word is also more easily
comprehended, the authors argue that this kind of model assumes
similar processes for production and comprehension systems (for
which both lexical selection and recognition are facilitated).
More importantly, the production system utilizes resources that
may seem to consult a model of the listener but that are
actually quite separate from them.
Additional evidence that speakers are motivated to produce
structures based on production processes, not on a model of the
listener, appears mostly in the production literature. For
example, a review of several production studies illustrates
that prototypical items appear earlier in phrasal conjuncts
than non-prototypical items (Kelly et al., 1986). Animate
items are more likely to be placed in subject position in the
production of transitive sentences than are inanimate items
(McDonald et al., 1993). And, the more accessible of two post
verbal nouns tends to be placed in direct object position in
the recall of dative sentences (Bock & Warren, 1985). The
accessibility of these items corresponds with the notion of
referential accessibility advocated in Chafe (1977) . The more
active an item is in the mind of the speaker, the more easily
participants are able to recall it. Thus, the results showed
that participants, when asked to recall items in a film, had
more difficulty labeling playground equipment (less typical)
87
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission
than they did identifying a piece of fruit (more typical). The
ease of identification for semantically accessible items
indicates that these items are more active in the speaker's
mind, thus they appear earlier. As a result, producing
semantically accessible items earlier in a sentence reduces the
communicative burden for the speaker, and hence, facilitates
the task of conveying the message at hand for the speaker as
well (see Bock, 1982) . Activating more accessible items
requires less attention than having to activate less accessible
information, allowing the speaker to focus attention on the
more difficult material (i.e., less accessible information)
that is to follow.
In a production study comparing actives and passives, F.
Ferreira (1994) also found that the form of the utterance
(i.e., active vs. passive) varied depending on the prominence
(defined by a thematic role hierarchy) of items. Speakers
placed more prominent thematic roles in more prominent
syntactic positions (i.e., subject positions). Thus
participants were most likely to produce passive sentences with
theme-experiencer verbs (e.g., challenged) because it placed
the experiencer role which is higher on the thematic role
hierarchy than in the more prominent syntactic position (i.e.,
subject). Similarly, agent-theme verbs (e.g., avoid) were more
likely to appear in active sentences because the agent appears
in subject position and the theme in object position. Ferreira
concluded that a speaker’s choice of active and passive
sentences "must appeal to the way in which the animacy of
88
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
arguments and the meaning of verbs (whether in the form of
thematic roles or some other type of semantic information)
interact during sentence formulation" (p. 25) . No mention was
made, however, of whether speakers consult a model of the
listener in determining which form of structure to produce.
One can arguably assume that by not addressing the benefits of
the listener, a speaker's production processes such as lexical
retrieval are paramount in a speaker's choice of alternative
structures, and that, following Dell & Brown, any facilitation
for the listener is simply a positive consequence of these
processes.
Similar conclusions follow from evidence that previous
exposure to a particular syntactic form influences production
and comprehension. Thus, speakers were more likely to describe
a picture using an active structure after reading aloud an
unrelated active sentence than after an unrelated passive
sentence (Bock, 1986c). In addition, comprehension of reduced
relatives was facilitated after being presented with other
reduced relative sentences (Bever, 1970). Again, Dell & Brown
(1991) argue that these results were not motivated by a
speaker's desire to produce what they believed would most
benefit the listener, but instead, were a reflection of
processes inherent in the production system.
Further evidence that shifting is a byproduct of production
constraints, comes from work on Dative Alternation structures
by Arnold, Losongco, Ginstrom, Brynolfson and Wasow (1997) .
The authors found that constituent order was influenced by
89
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
production difficulty as measured by the presence of speech
disfluencies. This suggested that choices in constituent
ordering are at least in part a strategy that is invoked when
the production process is particularly difficult or the shifted
item is less accessible.
The applicability of these results to heavy-NP shift is
further illuminated in a comparison between the model of the
listener hypothesis and what Dell & Brown (1991) call the
modularity hypothesis in which the production system relies on
its own processes. In examining the placement of typical vs.
atypical items in story retelling, Dell & Brown found that
speakers did not use information about the listener in within-
clause categories (e.g., 'The robber stabbed the man with a
knife. The robber used a knife to stab the man. . .The robber
knifed the man.'). However, in separate clause categories
(e.g.. The robber stabbed the man. He used an ice pick... The
robber had a knife. He stabbed the man.), the model of the
listener was consulted (p. 112). The findings for within-
clause categories not only argues for a modular system of
production in which structures are facilitating listener
processing only as a result of the production processes, it
also makes clear predictions about listener/speaker
accommodation in heavy-NP shift.
Dell & Brown's findings hold further implications about the
role of discourse information in motivating heavy-NP shift
structures, suggesting that within-clause items should not be
sensitive to pragmatic factors. This hypothesis clearly
90
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
contradicts the results presented earlier, whereby pragmatic
factors influence given-new ordering preferences for heavy-NP
shift. It also contrasts with findings by Bock and Irwin
(1980) that given information was more likely to precede new
information in a written recall task where the answers were
within-clause structures (e.g., The rancher sold the cowboy the
horse, p. 471). At least two explanations are available for
Bock & Irwin’s (1980) results. First, the presence of given
before new information may have been influenced by the between-
clause status of the question that induced recall (e.g., A
rancher received an inquiry from a cowboy about something he
needed for his act, p. 471) . Secondly, the placement of given
before new information may have occurred during an editing or
monitoring stage. While the first alternative is difficult to
assess, the latter is not inconceivable, since the task
required a written recall in which participants were allotted
sufficient time to engage in editing processes while writing
their answers (Bock & Irwin, 1980). This explanation aligns
itself with Dell & Brown's (1991) argument that a speaker's
model of the listener would only be relevant during editing
stages, and that pragmatic factors would only influence
separate clauses.
That the necessary constituents of heavy-NP shift are
contained within a clausal unit suggests that speaker, not
listener accommodation, should prevail following application of
Dell & Brown's within-clause results to heavy-NP shift.
Whether a person is speaking or listening, heavy-NP shifted
91
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
structures are produced in order to lighten the load of the
processor at the beginning of the sentence. This process frees
up more attention to process the more complex (i.e., longer,
more prosodic, newer etc.), and less accessible, elements at
the end of the sentence. If the more accessible information is
presented earlier than the complex material, then processing is
less likely to be interrupted later in the sentence when
heavier, or more complex, information is presented, "since
there is normally a pause break during which there is no new
influx of sentence material to be processed" (Cooper & Ross,
1975, p. 89).
Speaker/Listener Predictions for Heavy-NP Shift
In accordance with the comprehension and production research
presented above, it is not unreasonable to predict a speaker
accommodation view of heavy-NP shift. Thus far, the evidence
encourages a view in which heavy-NP shift structures are
motivated by production processes, rather than pure listener
accommodation alone. The information status of the NP and PP
should also be tested in order to confirm (or disconfirm) the
modularity hypothesis advocated by Dell & Brown (1991). Given
the relationship between weight and information status, we must
first understand the precise nature of weight effects before
such a task can be successfully pursued.
If findings by Wasow (1997a; 1997b) are any indication,
support for the modularity hypothesis should be confirmed.
Wasow (1997a; 1997b) argues that heavy-NP shift does not
facilitate comprehension, but instead "facilitates utterance
92
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
planning and production" (p. 22). In example (9b), repeated
here for the sake of convenience, Hasow argues that this kind
of statement provides the announcer with additional time to
check the scorecard for the batter's name, ultimately
benefiting the announcer (i.e., the speaker). In terms of
accessibility, it is feasible that the listener simply benefits
from production processes that place more accessible
information (the act of coming to the plate) before less
accessible information (the name of the next batter), thus
supporting the speaker-accommodation view of Dell & Brown.
9b. That will bring to the plate Bonds.
Moreover, based on the assumption that the listener benefits
the earlier a commitment to a structure is made, whereas the
speaker benefits when options are allowed, Wasow compares the
frequency of heavy-NP shift sentences in transitive (e.g.,
brought) and prepositional (e.g., wrote) verbs. He predicts
that transitive verbs, which require only an object, should
facilitate processing for the listener by allowing early
commitment to a structure. On the other hand, prepositional
verbs, which allow an object and PP, or a PP alone, should
provide more options for the speaker (e.g., late commitment)
than for the listener. In other words, Wasow predicts that if
the subject transitive matrix 'Pat brought' is encountered,
listeners benefit because they expect an NP; thus, if the verb
appears in the basic order, the listener is satisfied.
Likewise, in the shifted order, the listener knows to wait for
the NP (in the shifted order, the presence of 'to' also alerts
93
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
the listener that there is both a PP and an NP, since the verb
still requires an object). If heavy-NP shift is produced to
accommodate listener needs, the frequency of heavy-NP shift
orders should be greater with transitive verbs than with
prepositional verbs, which do not allow for early commitment.
Since the NP is optional in the latter class of verbs, the
listener is left wondering whether or not the [V PP] sequence
constituted the entire utterance, or if an NP would appear. In
fact, transitive verbs were less likely to appear in heavy-NP
shift structures than were prepositional verbs (5.6 to 9.3% in
the written corpus and 1.45 to 3.82 % in the spoken corpus) .
Both the written and spoken corpora data support a speaker
accommodation view, though Wasow (1997a; 1997b) cautions that a
listener perspective is not entirely irrelevant.
Satmaxa
This chapter began with a review of speech error data and
the ways in which these data support current production models.
Evidence against strict top-down incremental models and in
favor of more interactive incremental models was provided.
The data presented in this chapter extend beyond mere
consideration of heaviness, the intuitively obvious driving
force behind heavy—NP shift. Two factors that show effects on
the production of active and passive structures include animacy
and thematic role structures. Findings in Stallings,
MacDonald, & O’Seaghdha (in press), reported that these factors
do not significantly influence the production of heavy-NP shift
structure. The authors attribute this to the lack of
94
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
grammatical role assignment that is effected in these
structures (i.e., assignment of the direct object occurs
independent of whether it appears in a shifted or basic order
structure).
Implications for length effects in the guise of current
incremental models of productions were outlined. In addition,
production evidence for the effects of lexical and referential
accessibility on given-new ordering structures was considered.
One of the difficulties inherent in given-new ordering is that
it is not easily compatible with focus of attention
explanations. Accessibility is therefore defined in terms of
the level of activation within a speaker's mind, not merely the
importance of an item. The definition of accessibility ties in
with whether heavy-NP shift benefits the speaker or the
listener as well. That is, the presence of context will
influence a speaker to produce shared material early on in an
utterance. Whereas some researchers argue that structures such
as heavy—NP shift are produced to accommodate the needs of the
listener, others (Brown, 1987; Horton & Keysar, 1996; V.
Ferreira & Dell, 1996) argue that speakers do not take listener
needs into account during the initial planning of an utterance.
Given the clausal nature of heavy—NP shift and evidence by
Wasow (1997a), it is likely that heavy-NP shift is motivated by
speaker needs and the listener simply benefits as a consequence
of the speaker processes. Wasow's emphasis on the influence of
different verb types on speaker versus listener accommodation
resonates with an additional factor that has not yet been
95
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
introduced in this chapter. It involves the role of verbs as
well and has been shown to influence the spoken production of
heavy-NP shift in Stallings, MacDonald, & O'Seaghdha (in
press) . This factor is reviewed in detail at the beginning of
chapter 4.
Furthermore, due to the interactive nature between weight
and information status, it is difficult to set out predictions
about the effects of pragmatic factors on heavy—NP shift
without a clear understanding of the weight effects that
influence the placement of the NP in clause—final position in
the first place. The next chapter therefore presents one
experiment from Stallings, MacDonald, & O'Seaghdha (in press)
and two additional experiments that combine to specify the
nature of weight effects that influence heavy-NP shift. The
first manipulates NP length and holds the PP constant. The
next manipulates PP length and holds the NP constant, and the
last manipulates both NP length and PP length simultaneously.
*Nespor & Vogel (1986) define a phonological word as one that
contains "syllables...joined into n-ary branching feet" (p.
143), but from a psycholinguistic perspective, the evidence
for whether the number of syllable affects initiation times
for phonological words is still undecided.
^Accessibility in an incremental model of production is similar
to the referential accessibility described in chapter 2.
96
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Chapter 4: The Role of VErbs and Relative Weight in
He a v y - N P S h i f t
The Soix or VXrbs or Heavy-NP S a m
The plausible constraints on heavy-NP shift considered so
far have focused on global syntactic and pragmatic properties
of ordering. Beyond counting the number of words in a phrase,
none have considered the lexical features of individual words.
Aside from Wasow's (1997a; 1997b) analysis of transitive and
prepositional verbs, the influence of verbs on ordering has
been largely ignored. To address the possible influence of
verbs on heavy-NP shift ordering, Stallings, MacDonald, &
O'Seaghdha (in press) introduced a factor that arises from
recent work in syntactic comprehension in which argument
structure properties of verbs have been accorded increasing
importance in ambiguity resolution and sentence interpretation
(F. Ferreira & McClure, 1997; Garasey, Pearlmutter, Myers, &
Lotocky, 1997; MacDonald, 1994; Trueswell et al., 1993) . The
constraint-based framework of lexical and syntactic processing
in comprehension (e.g., F. Ferreira & McClure, 1997; Gamsey et
al., 1997; Kawamoto, 1993; MacDonald, 1994; MacDonald et al.,
1994; Tanenhaus & Trueswell, 1995; Trueswell et al., 1993)
argues for detailed lexical representations that encode a wide
variety of frequency-sensitive information whereby the
frequency with which a word appears in a particular syntactic
context is encoded. For example, MacDonald et al. (1994) have
97
with permission of the copyright owner. Further reproduction prohibited without permission
suggested that the representations of verbs include information
about how frequently a verb has appeared in different
environments, including alternative tenses, active vs. passive
voice, and alternative argument structures such as taking an NP
complement vs. an S-complement.
More specifically, this framework assumes that the frequency
information is represented via weighted links to
representations in the lexicon. For example, a verb that is
typically used in the active voice would have a strong link to
the 'active' representation and a weak link to the 'passive'
representation, whereas a verb that is used more often in both
voices would have equally strong links to each representation
of voice. The weights on these links determine the extent to
which alternative interpretations, such as 'basic' or 'shifted'
orders will be activated when a verb is encountered.
If we assume, in keeping with lexically driven production
models (e.g., Bock, 1987a; Bock & Levelt, 1994; Bock et al.,
1992; Levelt, 1989) that these same representations play a
prominent role in production, then phenomena such as heavy-NP
shift are likely to be influenced by some of the same processes
that have been identified in studies of comprehension. This
account can therefore be extended to suggest that when a verb
is selected during the production process, frequency-weighted
lexical information becomes activated and could provide an
additional constraint on the choice of a syntactic structure.
On this basis, Stallings, MacDonald, & O'Seaghdha (in press)
developed a verb disposition hypothesis, which states that
98
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
individual verbs carry with them information on the history of
their participation in shifted structures, and that this
history influences the likelihood of their allowing heavy-NP
shift. Thus, for heavy-NP shift sentences such as (16c), Jake
explained to Jim all of the facts regarding the import and
export taxes, a speaker's activation of the verb explained
should partially activate both the shifted and the basic
syntactic ordering options for the direct object and goal
arguments of the verb. Furthermore, the degree of activation
of each option should be a function of the frequency with which
explained has participated in the alternative structures, the
unshifted structure in which the complement of the verb (the
NP) is adjacent to the verb, and the shifted structure in which
the verb and complement are not adjacent.
As illustrated in Table 1, the authors compared verbs on two
levels. A complete list of items appears in Appendix A. The
first, 'NP/S verbs' (e.g., explained, demonstrated), are those
that permit both NP direct objects and sentential complement
constructions, sometimes known as 'that-complements' or 'S-
complements' (e.g., Jack explained that the book was missing) .
The second, 'NP-only verbs' (e.g., donated), on the other hand,
permit only direct object NPs (e.g., Jack donated the money) .
Heavy-NP shift is clearly an example of a structure in which
the verb and its complement are non-adjacent, but it is not the
only one. Sentential complements, for example, typically do
not appear adjacent to the verb when other material (adverbs or
PPs) is in the verb phrase, as in explained [immediately] [that
99
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
the book was missing] or revealed [to Jerry] [that the jewels
had been stolen]. Relative to verbs that do not take S-
complements, NP/S verbs promote shifting because they often
appear in "non-adjacent structures," those in which a
complement of the verb is non-adjacent to the verb. The source
of the previous history of non-adjacency may be quite varied.
In addition to a verb's ability to take S-complements, some
verbs may tend to take long NP complements, which would often
be shifted. Likewise, they may tend to take prosodically
prominent complements or complements expressing new
information, again promoting a past history of non-adjacent
usage and thus greater activation of this structure during
production.
Table 1: Sample Items in Stallings, MacDonald, & O'Seaghdha (in press)
Condition NP-only & NP/S Verb PP NP
Short NP introduced/ announced on Monday the captain
Long NP introduced/announced on Monday the new vice chair of
the first corporate
finance group
Stallings et al. (in press) tested the hypothesis that verbs
which have frequently appeared non-adjacent to their
complements, whether S or NP, should promote shifting more than
verbs that rarely appear non-adjacent to their complements. As
illustrated in Table 2, the findings illustrated that the
'shifting disposition' of individual verbs influences the
production of heavy-NP shift structures. Participants uttered
shifted structures more than twice as often when the sentence
contained an NP/S verb than with an NP-only verb. Verbs that
100
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
did not hold a strict adjacency requirement on their NP
complements were more likely to appear in shifted structures
than the verbs that required their complements to appear in an
adjacent position. These effects suggest that the relevant
difference between the two kinds of verbs is that the NP/S
verbs previously appeared non-adjacent to their complements
more often than the NP-only verbs. The authors concluded that
the propensity to shift, determined as a function of the
frequency of usage in various non-adjacent structures, is
therefore presented with each verb. Lexical representations of
verbs therefore encode a wide variety of frequency-sensitive
information, including the frequency with which a word appears
in a particular syntactic context.
Table 2: Mean shifting percentages o f NP-only and NP/S verbs (with Standard
Deviations). From Stallings, MacDonald, & O'Seaghdha (in press) .
Verb Type % Shifted. Order
NP-only 5.10 (11.88)
NP/S 11.13 (18.24)
A Competitive Component
While constraints such as length, prosody, and informational
status illustrate the important role of incremental!ty in that
accessible constituents such as NPs influence syntactic
structure choice, the verb disposition data point to a clear
competitive component in the process as well. This account is
related to an earlier constraint satisfaction account of
competitive production processes (Bates & Devescovi, 1989;
MacWhinney & Bates, 1989), which emphasizes the simultaneous
101
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
activation of multiple syntactic alternatives. It differs from
the earlier model, however, in that the basis of competition is
lexically rather than conceptually mediated (see for review De
Smedt, 1994b).
Applied to the verb disposition constraint, the competition
is tied to individual verbs, specifically the frequency with
which each verb participates in the adjacent-complement and
non-adjacent complement structures. On this view, verbs that
rarely appear in non-adjacent structures engender little
competition between alternatives, whereas verbs that more
frequently appear in non-adjacent structures should engender
partial activation of the alternative structures and more
competition. If evidence is found for a competitive component,
as there appears to be, the verb—based activation levels of the
shifted and unshifted orders should modulate the extent to
which length, prosody, pragmatics, or other factors promote
shifting. For verbs that have rarely appeared in shifted
structures, the shifting alternative would have little
activation, and the presence of another strong determinant
would be necessary for shifting to occur. Conversely, the
presence of a verb that has frequently been used in shifted
structures would activate the shifted option, and thus would
need only weak prompting from other factors to induce shifting.
Additional competition is evidenced in two ways: a) greater
sensitivity to contextual factors that might promote one or the
other structure and b) effects on decision times. Greater
sensitivity to context was demonstrated in F. Ferreira's (1994)
102
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission
data. Theme-experiencer verbs had a higher frequency usage in
the passive voice than did simple transitive verbs (by virtue
of the thematic hierarchy, as Ferreira suggests, or because of
some other factor that affects the usage of these verbs). The
manipulation of another factor for choice of active vs. passive
structure, subject NP animacy, had a greater effect on
syntactic structure for the theme-experiencer verbs than for
the simple transitive verbs. The theme-experiencer verbs'
sensitivity to animacy can be attributed to the partial
activation of the active and passive voice for these verbs.
The simple transitive verbs, by contrast, only slightly
activated the passive voice and therefore the animacy
manipulation did little to promote the use of passive
structures with these verbs.
Response time data elicited from Stallings, MacDonald, and
O'Seaghdha’s (in press) experiments provide evidence for (b),
effects on decision times. The competition hypothesis makes
predictions concerning decision times conditionalized on choice
of the basic vs. the shifted structure. When the basic order
is chosen, decision times should be shortest when there is
little activation of the shifted structure. Thus, decision
times for the basic order should be shorter for sentences with
rarely shifted verbs than for frequently shifted verbs. This
is the pattern observed in the analyses: Frequently-shifted
items yielded longer decision times in favor of the basic order
and shorter decision times in favor of the shifted order,
compared to rarely shifted items. These results suggest that
103
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
the effect may lie in the decision process. That is, decisions
in favor of the shifted structure should be made more quickly
when there is substantial activation of the shifted structure,
as with frequently-shifted verbs, than when there is little
activation of this structure, as with rarely-shifted verbs.1
Inspection of the experimental items from Stallings et al.'s
stimuli revealed that most, but not all, frequently-shifted
items contained NP/S verbs, and that the items that shifted
with moderate frequency (i.e., 4-5 times) contain about equal
numbers of NP-only and NP/S verbs. This pattern is expected on
the view that it is a verb's previous frequency in non-adjacent
structures, and not its NP-only vs. NP/S argument structure per
se, that determines its current disposition to shift. Note
that this competition could be realized as rivalry between the
orders themselves (NP-PP vs. PP-NP) and/or in terms of
consequent competition between the NP and PP for insertion in
the first postverbal phrasal slot.
Further support for competition processes was found by
correlating the shifting frequencies with decision times in the
long NP condition for the basic order and for the shifted
order. The more frequently shifted an item, the longer the
decision time in favor of the unshifted, basic order and the
shorter the decision times for the shifted order. For each
experimental item with each verb (each item appeared with two
different verbs) , we counted the number of times the item was
uttered in shifted form. The range of shifting frequencies was
0-11 (3.9 mean; see Appendix A), out of a total of 24
1 0 4
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
presentations (12 of which were in the short condition, for
which shifting was very rare) . Most, but not all, frequently-
shifted items contained NP/S verbs, and the items that shifted
with moderate frequency (i.e., 4-5 times) contained about equal
numbers of NP-only and NP/S verbs. Overall, the correlational
analyses support the competition hypothesis and suggest that
both the shifted and unshifted orders were partially activated
in the production task."
Such evidence for competition calls into question strict
incrementalist accounts of production (e.g., Garrett, 1980), in
which the sequential activation of constituents in the sentence
(such as NPs) plays a major role in determining the choice of
syntactic structure. It is unclear how frequency—weighted
lexical information would activate a structure in the context
of functional and positional processes. Conceivably, lemmas
could be assigned to different phrases at the functional level,
after which the phrases would be serially ordered during
positional processes. But considering the fact that lemma
assignment occurs during functional processes, the process
still leaves unspecified how a verb's lexical representations
would affect the positional ordering of the structure.
Likewise, a strict top-down model of production may limit other
possible constraints on heavy—NP shift, since the source of the
previous history of non—adjacency may be quite varied across
verbs.
It is possible that heavy-NP shift, by virtue of its
independence from syntactic role assignment, is especially
1 0 5
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
sensitive to competition between relatively free syntactic
phrases for insertion into syntactic plans. If so, competition
may only apply to word order preferences that involve phrasal
ordering. Thus, other syntactic choices may not engender
competition to the extent that heavy-NP shift appears to do.
Focus on the properties of individual verbs, rather than, or in
addition to, properties of entire verb classes, will be
essential in further refining our understanding of the extent
to which competitive mechanisms contribute to syntactic
processes in production. Either way, if results such as those
shown here generalize to spontaneous production, then strictly
incrementalist accounts of sentence construction (Bock &
Levelt, 1994; Kempen & Hoenkamp, 1987; Levelt, 1989; V.
Ferreira, 1996) must be tempered by recognition of the role of
competition at the level of phrase ordering in sentence
planning.
In sum, a lexical-semantic factor such as the previous
shifting disposition of verbs clearly influences the spoken
production of heavy-NP shift. The challenge to current
production models was described in detail, but it is not the
only challenge that these results pose. In addition, these
verb type data provide strong evidence against single factor
explanations of heavy-NP shift, hence challenging purely
grammatical explanations. As stated earlier, a relative weight
account of heaviness assumes a performance-motivated
explanation, but clear evidence about the role of weight in the
spoken production of heavy-NP shift has not yet been obtained.
106
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
In order to consider the extent to which additional factors
might also influence the spoken production of heavy-NP shift, a
firm understanding of the nature of weight effects is
necessary. Is it relative weight that motivates the production
of heavy-NP shift, as Hawkins (1994) argues, or is heaviness
based on some property of the NP alone? Three logically
possible alternatives are available to test the hypothesis that
relative weight drives heavy-NP shift ordering. All are
examined in the following experiments. First, data from an
experiment by Stallings, MacDonald, & O'Seaghdha (in press) in
which NP length was manipulated and PP length was held constant
is presented. Next, two new experiments are presented to more
fully determine the precise nature of weight effects. The
first manipulates PP length and holds NP length constant and
the second manipulates both NP and PP length simultaneously in
order to consider what, if any, relationship, between NP length
and relative weight exists.
NP Length M utxpviated
In addition to verb type, Stallings et al. also manipulated
NP length (either 10 or 2 words) and held the PP constant at 2
words. This manipulation resulted in either an 8 or 0-word
difference between the length of the NP and that of the PP. In
a production task that asked participants to assemble sentences
from phrases presented on a computer screen, Figure 4
illustrates that the number of shifted sentences was
significantly more frequent in the long (10-word NP) condition
than in the short (2-word NP) condition. Participants shifted
1 0 7
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
about 15% of the time in the long NP condition as opposed to
just over 1% of the time in the short NP condition, F\ (1, 23)
= 25.3, p < .001; F2 (1, 39) = 29.56, p < .001.
These results contrast with the findings of McDonald, Bock,
and Kelly (1993) who found no effects of length on the
production of word order. Stallings et al. attribute the
different effects to the distinction between phrase and word
length arising at different stages of production. Word length
may have little effect on the ordering of words within a phrase
(McDonald et al.), but the results indicate that phrase length
measured by the number of words constrains the ordering of
phrases within a syntactic structure (Stallings et al.). Thus,
production models must account for the influence of length
effects on phrasal ordering processes.
Two important issues arise from these data. The first
concerns the extent to which NP length and/or relative weight
drive heavy-NP shift. In this experiment, NP length is
confounded with relative weight in that the relative weight
between the NP and the PP increases as che length of the NP
increases and the PP length remains constant. Consequently,
whether shifting was motivated by NP length alone (the 10-word
condition) or the relative weight of the two constituents (the
8-word difference condition) is unclear, since the larger
relative weight difference condition contained the long NP.
One way to tease apart the difference is to conduct another
experiment in which PP length, not NP length, is manipulated
and then compare whether the effects of NP length are
108
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
maintained. To aid in the determination of whether production
models should account for NP length alone or relative weight,
the next experiment will do just that.
Before continuing, however, the second issue that derives
from these data warrants further exploration. In particular,
as we attempt to untangle the difference between NP length and
relative weight, it is important to clarify the extent to which
relative weight, as described in Hawkins (1994), is actually
instantiated in these experiments. In chapter 2, relative
weight was defined in terms of EIC ratios, such that structures
with efficient immediate constituent (IC) to word ratios are
most preferred in performance. As the ratios of the basic V NP
PP structures decrease, the likelihood that a shifted order
structure would appear increases. Relative weight is not
tested here in the strict sense of Hawkins' EIC principle, in
that the IC to word ratios are not compared across conditions.
Rather, the manipulation of relative weight that was used in
these experiments emerged from thinking about weight within the
framework of incremental production. Within incremental
models, ratios are not ascribed any particular status in
production as with EIC. Instead, the relative lengths of items
are taken into account. As the phrases are being readied for
production, shorter ones will tend to be ready before longer
ones such that weight effects will become more prominent the
larger the difference between the two phrases. In the
following sections, therefore, relative weight refers to the
difference between the number of words in the NP and PP.
109
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Figure 4: Results from Stallings, MacDonald, & O'Seagbdha's (in press) prompted
recall task: Production o f heavy-NP shifted structures as a function of NP
length.
[ j O-Wd. D iffr. (2NP)
I 8-Wd. D iffr. ( 10NP)
a U]
Relative Weight
E S xberxmbkt 1 : M uttpuz& m x; P P L e n g t h
This experiment was designed to investigate the effects of
relative weight on choice of basic [NP-PP] or shifted [PP-NP]
order structures. Participants constructed sentences from
phrases appearing on a computer screen, choosing either the
basic order in which a verb and direct object NP were adjacent
and a PP followed the NP [S-V—NP-PP] , or the shifted order in
which the PP intervened between the verb and the direct object
NP [S-V-PP-NP] . The primary goal was to assess the effect of a
manipulation of PP length (2, 5, & 7 words) in this task. If
NP length alone influences the production of heavy-NP shift
orders, we expect participants to choose the shifted structure
equally often in each of the 3 PP conditions since NP length is
held constant. However, if relative weight is driving the
production of the shifted order, then we expect participants to
shift more in the short PP conditions, which would result in
110
with permission of the copyright owner. Further reproduction prohibited without permission.
longer relative weight differences than in the long PP
condition.
Method
Participants. Thirty-six University of Southern
California undergraduates were either paid or received extra
credit in psychology courses for their participation. All were
native speakers of English.
Materials. We constructed thirty experimental items, each
consisting of three components: a subject and verb, a noun
phrase, and a prepositional phrase. An example is shown in
Table 3, and a full list of items appears in Appendix B.
Table 3: Sample items in Experiment 1: PP length manipulated
Condition Subject-Verb PP NP (10 words)
3-word
difference
(7 wd. PP)
The mayor
reported
with some regret
and a little
hesitation
the hard facts about
seven bank thefts and
five murders
5-word
difference
(5 wd. PP)
The mayor
reported
with some regret
and sorrow
the hard facts about
seven bank thefts and
five murders
8-word
difference
(2 wd. PP)
The mayor
reported
with regret the hard facts about
seven bank thefts and
five murders
Each of the 30 subject-verb phrases was 3-4 words in length
(e.g., The dancer realized, The radio listeners accepted). The
set of verbs did not include any alternating datives such as
give, because these verbs can participate in phrase orders
other than those that were under investigation here. A subject
NP, one direct object NP, and three PPs manipulating relative
weight were written. Short PPs contained 2 words and 2-5
syllables. Mid length PPs (5 words and 7-8 syllables) and long
PPs (7 words and 10-12 syllables) were also constructed. The
111
with permission of the copyright owner. Further reproduction prohibited without permission.
NPs contained only prenominal adjectives and prepositional
phrases.
The PPs, including as, at, by, during, from, in, on, with
without, expressed Time, Location and Manner roles, as in
during class, at a busy national park, with some regret and a
little hesitation, etc. Moreover, all contained inanimate
nouns (e.g., at lunch) and all could be syntactically analyzed
as adjuncts of the main verb.
Design and Procedure. Both pp length (2 vs. 5 vs. 7
words) and screen position (NP top vs. NP bottom) were
manipulated within participants.
The task employed a constrained production paradigm in which
participants prepared and then uttered sentences using phrases
presented on a computer screen (Dell & O'Seaghdha, 1992; F.
Ferreira, 1994; Stallings et al., in press; V. Ferreira, 1996).
We instructed participants to read the phrases, arrange them in
a sensible order, and prepare to produce the resulting sentence
from memory when cued to speak. We informed them that some
sentences would make more sense with the phrase order MIDDLE-
TOP-BOTTOM and others would work better with the order MIDDLE-
BOTTOM-TOP. With the exception of possessive apostrophes,
there was no punctuation. The subject-verb phrase always
appeared underlined in the center position to indicate that it
constituted the beginning of the sentence. The position of the
other two phrases was counterbalanced across participants and
items, so that the NP was on top and the PP on the bottom on
112
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
half of the trials and the reverse was true for the other half
of the trials.
The 10 practice and 20 of the 30 filler items were
constructed so that only one ordering of the top and bottom
phrases yielded a grammatical sentence. The remaining 10
filler items included a two-word NP and a two-word PP. In 6 of
these 10 items, the NP appeared at the top of the screen in
order to counteract any tendency, observed in pilot work, to
favor the phrase order MIDDLE-BOTTOM-TOP. Following the same
reasoning, 13 (65%) of the remaining 20 filler items, were
grammatical only in the phrase order MIDDLE—TOP—BOTTOM.
At the start of each trial, three left-justified phrases
appeared at top, center, and bottom screen locations.
Participants pressed a key to indicate when they were ready to
begin speaking. This keypress was followed by a one-second
blank interval, which was in turn followed by a prompt to begin
speaking. The speaking prompt was the Subject-Verb phrase
(e.g., The dancer realized) of the sentence, which reappeared
in its original location in the middle of the screen. When
participants saw this cue, they were to say the entire sentence
from memory as accurately as possible. As soon as participants
began speaking, the Subject-Verb phrase disappeared and they
completed the sentence as best they could. None of the phrases
reappeared on the screen once the prompt disappeared, and the
screen remained blank until the experimenter pressed a key to
record the order (basic or shifted) in which the sentence had
been uttered. Equipment errors were recorded by hand.
113
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Following the 10 practice items, the experimental and filler
items were presented in random order. The sessions lasted
between 30 and 45 minutes.
Results and Discussion
This recall task is quite difficult, especially in the long
NP conditions, and we did not expect error-free recall.
Rather, we were primarily interested in whether the NP or PP
immediately followed the SV prompt. Because it is feasible
that the heavier memory demands could have reduced the length
of the NPs produced and thus the overall incidence of shifting,
all sessions were tape-recorded for later transcription. Given
Stallings, MacDonald, & O'Seaghdha’s (in press) findings of
reliable shifting differences, we do not expect the task to
erase effects of shifting. However, the possibility that
results are influenced by the nature of the recall task, such
that the amount of words recalled actually changed the
proportion of words across conditions, would call into question
the results of primary interest (i.e., whether participants
produced the shifted or basic across varying conditions).
Consequently, two dependent variables were analyzed: (1) the
percentage of shifted [PP NP] orders that participants chose;
and (2) the relative weight between the number of words that
were actually uttered in the NP and the PP. These data were
analyzed as a function of PP length and screen position (NP top
vs. NP bottom). In the following analyses, 1.85% of the trials
were excluded due to short ready response times (< 500 ms).
114
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
which indicated too little time for the participant to see the
phrases, let alone prepare an utterance.
We also analyzed the participants' utterances for number of
words produced across conditions, based on transcripts of the
recorded experimental sessions. Due to the nature of the task,
participants sometimes changed, deleted, or added words to the
original stimulus items. In an additional 6.67% of the trials,
participants failed to utter either a [V NP PPI or [V PP NP]
structure. In 53 of these trials, participants recalled words
from only one of the constituents (either the PP or the NP).
These trials as well as an additional 19 trials in which
participants produced sentential complements [V S'],
infinitivals [V to], [V], or [PP V NP] constructions were
excluded from the analyses. Moreover, due to failure of the
tape recorder, some trials were not recorded during the testing
of some participants, resulting in the loss of two additional
trials in the analyses that were based on the transcripts.
Phxra.se order choice. As shown in Figure 5, there were
clear effects of PP length. As the PP increased in length,
participants shifted less frequently than when the PP contained
fewer words. Participants shifted most in the 8 word-
difference condition (38.7%), as opposed to 37% and 29.7% in
the 5 and 3 word-difference conditions, respectively. Although
no significant length differences obtained between the 5 and 8
word difference conditions, the difference between the 3 word
and 8 word difference conditions was significant in both
analyses; Fi (1, 35) = 4.215, p < .05; F2 29) = 4.559, p
1 1 5
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission
< .05. The difference between the 3 word-cLifference and the 5
word-difference condition was significant in the subjects
analysis only Fi (1, 35) = 4.599, p < .05; Fz < 1.
A position effect also obtained in both subject and item
analyses for this condition. Participants shifted more
frequently when the NP appeared at the bottom of the screen
(36.99%) than when it appeared at the top of the screen
(29.68%); Fi (1, 35) = 4.756, p < .05; Fz d, 29) = 6.479, p
< .05. Thus, it seems that participants were reading from top
to bottom, despite previous pilot work in which the tendency
was to read in the order Middle—Bottom—Top. Furthermore, since
the NP was held constant in this experiment, the findings
clearly favor the hypothesis that relative weight, not NP
length alone, motivates the spoken production of heavy-NP shift
sentences.
Figure 5: Experiment 1 results: Production of heavy-NP shifted structures in a
prompted recall task as a function of PP length.
50 --------------
□ 3 - W d . D i f f r . ( 7 P P )
■ 5 - W d . D i f f r . ( 5 P P )
40" ■ 8 - W d . D i f f r . ( 2 P P )
Relative Weight
I fords uttered.. From the transcripts, we calculated the
number of words produced in each condition. Corrections,
116
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
hesitation words, and repetitions of words were not included in
word counts. As shifting is strongly influenced by length, we
were less concerned with whether the phrases were recalled
perfectly accurately than with the number of words that were
uttered. For example, for the NP thirty paintings, accurate
reproduction of this NP would be counted as a two-word NP
utterance, as would the slightly inaccurate twenty paintings,
and about thirty paintings was scored as a three-word NP. The
number of words produced in the each of the conditions is
presented in Table 4. The table shows that while the short PP
condition yielded fairly accurate recall of the two-word PP
(2.08 words), participants tended to omit some words from the
longer PP conditions and the long NP. Thus, they produced a
mean of only 4.33 and 5.14 words in the 5 and 7-word PPs
respectively. Moreover, across all three conditions, they
produced a mean of only 8.18 words for the intended 10-word
long NP.
Table 4: Mean number of words (and Standard Deviations) recalled for each phrase
across each condition.
Phrase Length NP Recall PP Recall
7-Word PP 8.05 (1.5) 5.14 (1.33)
5-Word PP 8.00 (1.48) 4.33 (.76)
2-Word PP 8.45 (1.36) 2.08 (.23)
Due to the nature of our paradigm whereby the relative
weights are strongly constrained, participants are unable to
respond to the stimuli with much flexibility. As shown above,
recall varies little from the pre-determined conditions.
Because there is so little difference in recall, we predict
that any effect these small differences could have on shifting
1 1 7
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
would be undetectable from the relatively small number of
subjects in this study. In fact, as Table 5 illustrates for
the mid PP and long PP conditions, the relative weight
differences that participants uttered changed in direct
proportion to the original conditions, so that the long PP
condition which constituted the largest weight difference also
reflected the largest difference condition. In addition, the
relative weight differences did not differ significantly based
on whether the shifted or basic order was produced, though the
means in the table for the shifted conditions represent very
few observations and should be viewed with caution. In the
shortest PP (8-wd. Difference) condition, for example,
participants uttered a mean difference of 6.61 words when they
produced the shifted order, and a mean difference of 6.35 words
when they produced the basic order; Fi (1, 30) = 1.201, p > 1;
£ * 2 (1, 28) = .635, p >1. Due to empty cells in the mid PP
(5-wd. Difference) condition, F tables were not available,
though the mean differences were again similar: a 3.5 mean
difference in the shifted order and a 3.7 mean difference in
the basic order.
Table 5: Experiment 1: Mean NP-PP length recall (and standard deviations) as a
function of structure choice
Basic Order Shifted. Order
5-Wd. Difference 3.7 (1.5) 3.5 (1.3)
8-Wd. Difference 6.35 (1.28) 6.61 (1.25)
Conclusions. These results indicate that participants
produced heavy-NP shift orders significantly more in the two
small PP conditions (the larger difference conditions) than in
118
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
the 3-word difference condition. Given that NP length was held
constant, the effects of PP length obtained in this experiment
in combination with the NP length effects presented in
Stallings, MacDonald, & O'Seaghdha (in press) provide evidence
that the spoken production of heavy-NP shift is motivated by
properties of both the NP and PP, rather than the NP alone.
Importantly, the recall analysis further confirms that
participants recalled words in proportion to the actual
conditions considered in the choice analysis. However, whether
NP length influences the production of heavy-NP shift
structures above and beyond the relative weight effects from
this experiment cannot be determined from these data. To
explore this question, both NP and PP length are manipulated in
Experiment 2. The fillers are slightly different, so that
fewer MIDDLE-TOP-BOTTOM orders appear in order to more
carefully consider the position effect obtained in this
experiment.
ExBBRzm sx 2 : HP .Length and PP L e n g t h M a h i f u l a tl o n s
To further consider the nature of weight effects in the
production of heavy-NP shift and to ensure that some task
peculiarity was not responsible for the conflicting NP length
and PP length effects in the previous experiments, this
experiment manipulates both NP and PP length. We compare two
conditions in which the relative weight difference is the same
but the NP length varies, with a third condition where there is
no difference between the lengths of the NP and PP. We predict
that if relative weight is the driving force of heavy-NP shift,
119
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
then the percentage of shifted orders that are produced in the
larger relative weight difference conditions should exceed the
percentage of shifted orders produced in the much smaller
relative weight difference condition. If, on the other hand,
NP length motivates the production of heavy-NP shift structures
above and beyond the effects of relative weight, then within
the two equal difference conditions, shifting should be greater
in the longer NP condition.
Method
Participants. Thirty-six University of Southern
California undergraduates were either paid or received extra
credit in psychology courses for their participation. All were
native speakers of English.
Ma.teria.ls and. Procedure, a sample of the materials is
illustrated in Table 6. They are similar to those from
Experiment 1 (a complete list appears in Appendix C) , except
that two additional levels of NP length were used (10, 7, or 2
word NPs) and only two levels of PP length (5 or 2 word PP)
were used. Again, screen position (NP top vs. NP bottom) was
manipulated. Due to the position effect in Experiment 1, the
filler items were changed slightly. Since the experimental
conditions include short (2 word NP, 2 word PP) items in this
Experiment, we excluded the short filler items that were used
in Experiment 1. We included 37 fillers that were grammatical
in only the Middle-Bottom-Top or Middle-Top-Bottom order. To
counterbalance the tendency in the last experiment for people
120
with permission of the copyright owner. Further reproduction prohibited without permission
to read the Middle-Top-Bottom order, we presented 18 of the
filler items in the order Middle-Bottom-Top. The remainder
appeared in the Middle-Top-Bottom order. Again, each session
was tape recorded for later transcription and lasted about 25-
40 minutes.
Table 6: Sample items in Experiment 2: NP and PP length manipulated.
Condition Subj ect-Verb PP NP
0-word
difference
The mayor
reported
with regret the crimes
5-word
difference
(7-wd.NP)
The mayor
reported
with regret the recent bank thefts
and strange murders
5-word
difference
(10-wd. NP)
The mayor
reported
with some
regret and
sorrow
the hard facts about
seven bank thefts and
five murders
Results and Discussion
As in Experiment 1, we measured the same dependent variables
as a function of NP/PP weight and screen position: phrase order
choice and the relative weight differences that were actually
uttered. 1.67% of the trials were excluded because time to
read the phrases and press the key was less than 500 ms.. We
again analyzed the participants' utterances for number of words
produced across conditions, based on the transcripts of the
recorded experimental sessions, according to the same coding
criteria applied in Experiment 1. In an additional 3.24% of
the trials, participants failed to utter either a [V NP PP] or
[V PP NP] structure. Of these excluded trials, 27 failed to
include either the PP or the NP and 8 were produced as
sentential complement or infinitival structures. Moreover, due
to failure of the tape recorder, one trial was not recorded
during the testing of one participant, resulting in the loss of
121
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
an additional trial in the analyses that were based on the
transcripts.
Phrase order choice. As shown in Figure 6, no
significant effects of NP length above and beyond the influence
of relative weight obtained on participants' choice of phrase
orders. Thus, there were again clear effects of relative
weight alone. Participants uttered sentences with a heavy-NP
shift structure about 7-9 times more often in the 5-word
difference conditions than they did in the 0-word difference
condition, Fi (2, 70) = 50.339, p < .001; F2 (2, 58) = 51.360,
p < .001. The planned comparisons showed no significant
difference between the two 5-word difference conditions (Fs >
1). However, length effects did obtain in the analyses between
the 0-word difference condition and each of the 5-word
difference conditions (7-word NP: Fi (1, 35) = 66.749, p <
.001; F2 (1, 29) = 72.110, p < .001; 10-word NP: Fi (1, 35) =
75.079, p < .001; F2 (1, 29) = 73.034, p < .001). These
results suggest that shifting is more likely to occur when the
difference between the NP and PP is great than when the lengths
are equal. They also illustrate that, above and beyond the
relative weight effects, NP length does not motivate the spoken
production of heavy-NP shift structures. Moreover, no effects
of screen position obtained in this experiment, suggesting that
the position effect in Experiment 1 was a by-product of the
unbalanced filler items.
122
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Figure 6: Experiment 2 results: Production o f heavy-NP shifted structures in a
prompted recall task as a inaction of NPand PP length.
□ O-WA Dif iV. (ZNP)
■ 5-WA Dif fr. (7NP)
£ 40 ' ■ 5-WA Diffr. (I0NP)
Relative Weight
Words littered.. These data were analyzed In the same way
as in Experiment 1. The mean length of the NP and PP across
each of the conditions is presented in Table 7. As in the
first experiment, recall was fairly accurate, with almost
perfect recall in the 0-word difference condition (2.0 and 2.07
in the 2-word NP and PP condition) and very similar recall in
the longer conditions. As a result, the number of words
actually uttered was in direct proportion to the original
conditions. In the 5-word difference, 7-word NP condition,
participants produced an average of 6.55 words in the NP and
2.06 words in the PP. Similarly, in the 5-word difference, 10-
word NP condition, participants uttered an NP of approximately
8.44 words and a PP of 4.34 words. In addition, the relative
weight difference actually produced in each of the conditions
was very similar (approximately 4.5 words).
Furthermore, no significant relative weight difference
obtained between the basic and shifted orders in the 5-word
difference 10-word NP condition. However, in the 5-word
1 2 3
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
difference 7-word NP condition, an effect in the anticipated
direction obtained in the participants' analysis only. The
relative weight actually uttered was greater in the shifted
(4.71 mean difference) than in the basic (4.4 mean difference)
order; Fi (1, 28) = 4.883, p < .05; F2 (1, 27) = .867, p < 1.
Table 7: Experiment 2: Mean NP-PP length recall (and standard deviations) in 5-word
difference conditions.
Condition NP PP
7-Wd. NP; 2-Wd. PP 6.55 (.79) 2.06 (.21)
10-Wd. NP; 5-Wd. PP 8.44 (1.32) 4.34 (.83)
Conclusions. The structure data in Experiment 2
confirmed that relative weight is the driving force behind the
spoken production of heavy-NP shift structures, and that NP
length does not exert any additional influence on the
production of these structures. Participants produced the
shifted order significantly more in the two 5-word difference
conditions than they did in the 0-word difference condition.
The percentage of heavy-NP shift orders that were produced did
not differ significantly in the two 5—word difference
conditions, hence, providing further evidence that relative
weight motivates heavy-NP shift and showing that NP length
alone does not exert any additional influence on shifting.
That the number of words recalled in the NP of each of the 5-
word difference conditions differed further corroborates these
conclusions. With the exception of the small effect of choice
in the participants analysis of the 5—word difference, 7-word
NP condition, accurate recall in this experiment resulted in
124
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
undetectable differences of relative weight on shifting as in
Experiment 1.
Smaasr
In sum, the experimental data presented in this chapter
suggest that lexical factors such as verb disposition effects
as well as weight effects both influence the spoken production
of heavy-NP shift. Furthermore, the presence of NP length
effects in Stallings et al. coupled with PP length effects in
Experiment 1 suggest that properties of both the NP and PP are
relevant factors in the production of heavy-NP shift. In
particular, as the difference between the length of the NP and
PP increases, shifting becomes more prevalent. The results in
Experiment 2 further illustrated that NP length alone exerted
no additional influence on shifting rates. If it had,
participants would have shifted significantly more frequently
in the 5-word difference, 10-word NP condition than in the 5-
word difference, 7-word NP condition. The fact that
participants' recall of the items was in direct proportion to
the designed conditions further corroborate these findings,
suggesting that the additional memory load required by this
kind of task did not significantly alter the results.
In each of the experiments, relative weight was measured by
the difference between the lengths of the NP and PP, not in
terms of EIC ratios as Hawkins describes. Consequently, the
evidence does not speak directly to EIC effects in the spoken
production of heavy-NP shift and the interpretation of these
effects is limited to the fact that properties of the NP and PP
1 2 5
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
combine to exert an influence on heavy-NP shift, not properties
of the NP alone.
The experimental data in this chapter call into question
current linguistic and psycholinguistic theories. Evidence of
verb disposition effects and relative weight effects challenges
a common tenet within grammatically motivated explanations of
heavy-NP shift. Namely, that a single-factor explanation fails
to adequately account for the existing data. Evidence of these
factors, thus, supports a multi-factor explanation of heavy-NP
shift ordering.
In addition, the evidence for verb disposition strongly
suggests that a competitive component, not only an incremental
one, is involved in the production of heavy-NP shift
structures. A plausible account for these data appears in
(Bock, 1986a) whereby alternative verb forms are activated at
functional levels of production which, depending on the
strength of activation, can influence the production of a given
structure. Initially, this might correspond with the verb
disposition evidence. However, the underlying assumption of
such theories is that processes assigned at the functional
level affect the grammatical role assignment of the items
involved. The effects of verb disposition in heavy-NP shift,
therefore, pose a challenge to these models, since the
assignment of the direct object to the NP is unaffected by the
order in which the NP appears.
Moreover, the finding that relative weight influences the
production of heavy-NP shift independent of properties of the
126
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
NP supports Hawkins’ (1994) initial text count findings in
which the percentage of shifted structures increased as the
relative weight between the NP and the PP increased. Following
De Smedt (1994a) these findings would challenge traditional
grammar-based generation systems including De Smedt's (1994b)
own IPF (Incremental Parallel Formulator) model. De Smedt
argues that such a model does not allow for surface features
like weight to influence production on the basis that
constituents are assembled into the syntactic structure long
before the length has been assessed.
In a functional integration model (e.g., Bock, 1986a; Bock,
1987a), however, factors like length can influence structure.
On this view, syntactic frames are processed at the functional
level of processing and phonologically-related effects, like
weight, are processed at the positional level of processing.
Moreover, relative weight effects might also lend support to
the two processing levels assumption described in chapter 3 (as
opposed to the one-stage model proposed by Stemberger (1985)).
Suppose, for example, that the basic order structure is
accessible at the functional level so that the syntactic frame
being passed to the positional level is the V NP PP structure.
However, as the weight of the constituents is processed, a
heavy NP is sufficiently longer than the PP, so that the PP is
already accessible. While the NP is being readied for
production, the information is fed back to the functional level
for reassignment. This kind of approach, of course, is
reasonable only if the lack of grammatical role assignment
127
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
effected by heavy-NP shift does not restrict heavy-NP shift
ordering to positional level processes. These issues are
considered in more detail in the Future Directions section of
chapter 5.
1 Another interpretation of the effect of verb type in the
basic order choice time analysis is that the longer decision
times for NP/S verbs simply reflect longer reading times for
these sentences. This pattern could be due to the added
complexity of having to resolve the argument structure
ambiguity inherent in NP/S verbs, in contrast to the
unambiguous NP-only verbs.
The competition model also makes predictions for at least
one well-known class of speech error, whole word exchanges.
Analyses of speech errors (e.g., Garrett 1975; 1988) have
generally operated within the framework of canonical phrase
orders, and so provide little guidance in theorizing about
the specific case of heavy-NP shift. In addition, because
the phenomenon itself concerns speaker preference, there is
no level of speech error that directly addresses it. The
verb disposition findings, however, do make some predictions
for whole word exchanges: Exchanges between same category
words (e.g., nouns, prepositions, etc.) should be
proportionate to the degree of competition between the
phrases to which they belong in syntactic planning. For
example, exchanges between NP and PP nouns should be more
frequent following the frequently-shifted NP/S verbs than
following NP-only verbs. A similar suggestion has recently
been proposed by V. Ferreira (1996) who analyzed
experimental errors predicted by incremental and competitive
processes. Although Ferreira concluded that the error
patterns did not favor the competition view, we note that
there were very few errors in the experiments. In addition,
Ferreira's experiments involved syntactic choices that
involved role assignment, not the relatively free phrase
ordering that is the hallmark of heavy-NP shift. Therefore,
it remains to be seen whether our conjecture finds support
in existing speech error corpora and in future experiments.
128
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission
Chapter 5: Fgtore Directions
The fundamental question addressed in the previous chapters
concerns the precise nature of weight effects that motivate
heavy-NP shift. This question assumes of course that syntactic
factors such as weight drive heavy-NP shift, but the question
itself raises subsequent issues that must also be addressed.
The preceding chapters, therefore, addressed the following
questions in pursuit of the fundamental question regarding the
nature of weight effects:
1. Are grammatical versus performance-based explanations
better suited to adequately describing heavy-NP shift
data (ch. 2) ?
2. If performance—based explanations are plausible, then...
a) How would psycholinguistic models account for heavy-NP
shift (ch. 3.2)?
b) Would the speaker or listener benefit from such an
account (ch. 3.3)?
c) Is a multiple factor approach to heavy-NP shift
plausible (ch. 3.2)? If s o . . .
1) What are the multiple factors (ch. 3.2 & ch. 4)?
2) How do multiple factors interact to motivate heavy-
NP shift?
3) Is it possible to induce shifting in a constrained
production paradigm (ch. 4)?
In the preceding chapters, I have presented evidence that a
grammatical account is not necessarily the most appropriate
129
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
mechanism by which one should explore heavy-NP shift. In the
spirit of Hawkins' (1994) ordering explanation, the account of
heavy-NP shift presented here deviates from the generative
grammar accounts that have so far been forwarded to explain
heavy-NP shift. It does so by maintaining that performance
principles need not be excluded from the grammar, but can, on
their own, adequately account for the complexity of factors
that combine to motivate the spoken production of heavy-NP
shift. In that light, a variety of factors ranging from those
that are intuitively relevant (e.g., weight) to those that are
less intuitively obvious (e.g., verb disposition) were
examined.
In chapter 1, I introduced three of Behaghel's (1932)
principles that determine ordering. According to Stockwell
(1977), the most important principle, for Behaghel, is that
items that belong together should appear together. This
principle, however, is sometimes overridden by one of the
following principles: short material precedes long material, or
given information should precede new information. These
principles laid the foundation for subsequent ordering work
that represented varied explanations. Chapter 1 pointed out
that despite several grammatical proposals, explanations for
heavy-NP shift overlapped on several dimensions including the
linguistic representations and the underlying theoretical
assumptions that influence heavy-NP shift. Consequently,
capturing the exact conditions under which an NP appears in
clause—final position has proved difficult.
1 3 0
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
In chapter 2, I summarized the nature of the problem in more
detail. Within grammatical proposals, for example, researchers
have proposed both syntactic and non-syntactic factors. In
addition, proposals that were syntactic in nature varied in at
least a few ways. Some researchers ignored Behaghel's short
before long principle and proposed that properties of the NP
alone are sufficient to induce shifting, while others extended
it, arguing that it is the relative weight between the NP and
its neighboring constituents that matters. Research advocating
that properties of the NP alone motivate shifting varied
further. Some researchers propose that the mere length (i.e.,
number of words) influenced heavy-NP shift, but others argue
that the syntactic complexity of the NP (i.e., an NP dominated
by S) defines the heaviness. Furthermore, we examined
proposals that posit the same motivating factor (e.g., weight),
yet are driven by different underlying theoretical assumptions
(e.g., competence vs. performance theories).
Despite the variance across these proposals, no one theory
provided a descriptively adequate account of the existing data.
Examples were illustrated to reveal that heavy-NP shift
structures often contradicted the theories: some NPs were not
syntactically or phonologically complex, some did not appear to
contain new information, and some were not longer than their
neighboring PP. The lack of descriptive adequacy, particularly
within the grammatical theories, increased the plausibility of
a performance-based explanation of the data. Substantial data
from Hawkins (1994) was presented in chapter 2 to support the
1 3 1
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
plausibility of a performance—based explanation, and motivate
the consideration of heavy-NP shift using performance-relevant
measures.
At least one specific advantage to a performance—based
approach is that it is not limited to single—factor theories.
In accordance with the data presented here and Wasow's
conclusion that shifting is not triggered by a single factor,
we will best be able to account for the factors that motivate
shifting if the underlying theoretical model allows for a
multifactor theory.
In chapter 3, I therefore evaluated the extent to which
heavy-NP shift fit in a model that placed the locus of shifting
in production processes, and not in the syntactic,
phonological, or pragmatic component of the grammar of the
speaker. In the course of that chapter, section 1 included a
review of speech error data in order to establish and justify
the framework of current incremental production models. In
section 2, the applicability of factors such as animacy,
thematic role structure, prosody, and pragmatics in current
production models was considered. Unfortunately, no direct
production evidence is currently available for the role prosody
and pragmatic factors might play in the production of heavy-NP
shift structures. The implications of these factors are
further addressed later in this chapter. In section 3, I
summarized comprehension and production evidence for and
against listener versus speaker accommodation, advocating the
likelihood of a speaker-based account of heavy-NP shift.
1 3 2
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission
EzBBznm sT R e s u l t s
Significant results across three variations of a production
Cask in Stallings, MacDonald, & O'Seaghdha (in press)
established the viability of inducing shifting in a constrained
production paradigm. Chapter 3 presented evidence from
Stallings et al. that despite effects of animacy and thematic
roles on word ordering in other production studies, neither
factor influenced the production of heavy-NP shifted
structures. The authors attributed these findings to the fact
that animacy and thematic roles are typically processed at the
functional level in incremental models of production where
grammatical role assignment occurs. In the case of heavy-NP
shift, grammatical roles are unaffected: the NP is always the
direct object regardless of its verb-adjacent or clause-final
position.
Chapter 4 presented additional results from Stallings,
MacDonald, & O'Seaghdha's production experiments. In doing so,
evidence for verb disposition and NP length effects were
established. The first, verb disposition, provides evidence
that heavy-NP shift is conditioned not by properties of weight
alone, but by the 'shifting disposition' of verbs as well. The
second factor, NP length, constituted one of the three
logically possible ways to test relative weight, but failed to
distinguish the extent to which weight is defined by heaviness
of the NP alone versus the relative weight of the adjacent
constituents. To unveil the precise nature of weight effects,
Chapter 4 included two additional experiments that considered
1 3 3
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
alternative measures of relative weight. The first manipulated
P? length and held NP length constant and the second
manipulated both NP length and PP length. The transcript
analyses revealed that the number of words participants
recalled were proportionate with the original conditions. All
three experiments confirmed that the choice of the heavy-NP
shifted vs. basic order (NP-PP) structures during speech
production is constrained not by NP length alone, but by the
weight of the NP relative to that of the PP in the verb phrase.
In these experiments, relative weight was measured by the word
differences between the NP and PP, not in terms of ratios as
advocated by Hawkins (1994), in conjunction with incremental
production models where shorter items are more quickly accessed
and readied for production than longer ones. The results
nevertheless correspond with the corpus analysis of Hawkins
(1994) and not with the grammatically motivated theories that
rely on properties of the NP alone to motivate shifting.
The experimental findings raise at least four questions that
require some additional attention. First, with respect to the
verb disposition findings, how is it that verbs that typically
take sentential complements influence the production of a
structure such as heavy-NP shift, which includes no sentential
complement? Second, is the number of words an accurate
reflection of phrase weight? Third, given the nature of the
task in which participants were provided with the exact forms
of the phrases, did the results reflect fundamental production
processes, or did they merely reflect a post-syntactic editing
134
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
process? Fourth, given the justification for a multi-factor
theory of heavy-NP shift, what is the nature of the interaction
between the multiple factors in the guise of current production
models? Next, I address the first three questions briefly and
then consider the fourth question in a bit more detail.
Sentential Complement Verbs and Heavy-NP Shift
The verb disposition findings discussed in chapter 4 suggest
that NP/S verbs, which take both NP and sentential complements,
appear in heavy-NP shifted structures significantly more
frequently than NP-only verbs, which take only direct object NP
complements. These results are particularly interesting given
that verbs which allow sentential complements influence heavy-
NP shift structures, which contain only a subject-verb phrase,
an NP and a PP, not a sentential complement. Following recent
comprehension work that argues for detailed lexical
representations that encode a wide variety of frequency-
sensitive information, Stallings, MacDonald, & O’Seaghdha (in
press) suggest that NP/S verbs carry with them information
about the structures in which they appear. On this view, these
verbs' frequent appearances in structures non-adj acent to their
complements (e.g., Mary said yesterday that she went to the
party) can influence the extent to which they engage in other
non-adjacent structures like heavy-NP shift. In other words,
the propensity to shift is determined as a function of the
frequency of usage in various non-adjacent structures, such
that the non-adjacency history is represented with individual
verbs. Thus, the key point here is not that the verb contains
1 3 5
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
a sentential complement, but instead that it has a history in
which it frequently appears non-adjacent to its complement.
This history, in turn, influences its appearance in other
structures such as heavy-NP shift, where the NP-complement is
shifted away from its canonical verb adjacent position. The
fact that participants, who were provided with sentence
ingredients and therefore knew they were dealing with an NP
structure, not a sentential complement, still shifted the NP to
clause-final position in these experiments appears to be rather
strong evidence of the force of the verb effects. It therefore
seems that knowledge of what constituents may be deferred,
regardless of the reason, is of vital interest to speakers.
The evidence here suggests that the frequency representations
that play a role in comprehension are also operating in
lexically driven production models, wherein frequency-weighted
lexical information becomes activated and could provide an
additional constraint on the choice of syntactic structure.
Clearly, this view of lexical representations differs from the
way linguists conceive of the lexicon. Just as further
research is necessary to consider what it is about these verbs
that influences shifting (i.e., is it that NP/S verbs are
merely more prone to taking long, new, or prosodically
prominent NPs, or it some other property), the relationship
between these distinct approaches also warrants investigation.
Calculating Phrase Length
In these experiments, phrase length was calculated by
counting the number of words in a given phrase. Whether the
1 3 6
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission
number of words is the appropriate measure for the determining
the effects of relative weight might be questioned, since
weight is often defined by the amount of syntactic structure an
item contains. Hawkins (1994), however, claimed that counting
the number of words resulted in the same pattern of efficiency
ratios, as did calculating non-IC structure. In addition,
Wasow's (1997b) data suggests that this calculation method is a
reasonable one. He compared the rate of heavy-NP shift across
three methods of calculating length: number of words, number of
nodes, and number of phrasal nodes. The percentage of heavy-NP
shift structures heeding the generalization that constituents
occur in order of increasing weight (what he terms, weight
monotonicity) were consistent across each of the three methods
(ranging from 88.4% to 91.4%). He concluded that the method of
calculation did not alter the results in any significant way,
and, as a result, any of the three methods constituted an
appropriate measure for weight.
Fundamental Production Processes
Given the nature of these tasks in which the participants
were provided with the complete form of each phrase, one might
question whether the experimental results are truly indicative
of pure production processes that give rise to heavy-NP shift
in spontaneous production, or if they merely reflect a late
"post-syntactic" editing process of a kind that is likely to be
exercised in deliberate writing more than in spontaneous
speech. To the extent that the verb disposition hypothesis is
supported, the pure production process view should gamer
137
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
support. This is because there is no obvious mechanism in the
post-syntactic editing account that could explain the kind of
detailed sensitivity to the history of prior use that is
implied by the verb disposition hypothesis. Given that the
findings of relative weight and verb disposition reflect the
nature of spontaneous speech, the next question concerns the
extent to which these factors interact to motivate the
production of heavy-NP shift structures.
M d z t x p o c I n t e r a c t i n g c o n s t r a i n t s
The results presented in the preceding chapters extend
Hawkins' (1994) EIC account, suggesting that factors in
addition to relative weight (e.g., verb disposition) also
influence the production of heavy-NP shift structures. This
section considers the impact of another factor, V-PP
collocations, on heavy-NP shift, and then considers both the
possible nature of its interaction with weight as well as its
relationship to the verb disposition hypothesis. Finally, the
viability of current production models given the findings
outlined above is assessed.
V-PP Collocations
Additional verb-related explanations were briefly introduced
in chapters 2 and 3 and appear in Wasow (1997a; 1997b) to
support the notion that a speaker-oriented view of heavy-NP
shift must be acknowledged. In chapter 3, for instance, data
were presented that showed the percentage of shifted structures
1 3 8
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
produced in speech and text was greater for prepositional than
for transitive verbs.
In chapter 2, another factor was proposed briefly to account
for the shifted order that appeared in example (9), That will
bring to the plate Bonds, even though the weight of the PP
clearly exceeds that of the NP. Wasow argues that shifting is
also influenced in part by the frequency of adjacent V PP
collocations such as bring to the plater keep in mind and take
into account. Like the argument surrounding prepositional and
transitive verbs, Wasow argues that the heavy—NP shift
structure in example (9) is produced to give the announcer more
time to look at the scorecard to see who is coming to the plate
next. But, the phrase also appears because it has become a
conventionalized collocation for the speaker. That is, the V
and PP in the phrase, bring to the plate, frequently occur
adjacent to one another. Furthermore, Wasow argues that
because this V PP combination is semantically transparent to
the extent that a listener would have no trouble understanding
the intended meaning of the verb if it appeared independent of
the PP, the production of the shifted structure benefits only
the speaker. If the verb were dependent on the PP for its
meaning, however, the listener would be at more of a
disadvantage if a basic order structure were produced.
Wasow (1997a; 1997b) compares the number of heavy-NP shift
structures that appear in semantically transparent collocations
and semantically more opaque collocations. Because the opaque
collocations require the PP before the correct interpretation
139
with permission of the copyright owner. Further reproduction prohibited without permission.
of the verb can be properly assigned, the listener benefits
when heavy-NP shift structures are produced. The speaker also
benefits from the production of a heavy-NP shift structure with
an opaque collocation. Once the V—PP collocation is produced,
the speaker can then exert extra attention to producing the
remainder of the utterance. For the same reason, speakers
benefit when they produce heavy-NP shift structures with
transparent collocations. The main difference between opaque
and transparent collocations then is that the listener does not
benefit from the production of heavy-NP shift structures with
transparent collocations, since the verb's meaning is assigned
independent of the PP.
Wasow argues that the presence of heavy-NP shift in
transparent collocations should help to ascertain the
difference between a speaker or listener-based account of
weight effects. Specifically, if a speaker-based account is to
be supported, then the number of heavy-NP shift structures
appearing with transparent collocations should be similar to
the number appearing in opaque collocations. On the other
hand, if a listener-based account is more appropriate, then the
transparent collocations should pattern more like their non-
collocational counterparts (i.e., appear in few shifted
structures).
Of 827 sentences from the Aligned-Hansard corpus, Wasow
coded 439 (53%) as collocations. Of these collocations,
semantically transparent collocations were less likely to
appear in heavy-NP shift structures (47%) than were
1 4 0
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
semantically opaque collocations (60%), but much more likely
than were non-collocations (only 15% shifting). Consequently,
Wasow claims that the effects were more intermediate. The
difference between transparent and opaque collocations was
significant, but so was the difference between transparent and
non-collocations. These findings suggest that the listener's
perspective cannot be entirely dismissed in an account of
ordering, but, at the same time, psycholinguistic explanations
need to focus on more than the listener alone. They also raise
at least two further questions. First, how do V PP
collocations, as a factor that motivates heavy-NP shift,
interact with weight? Second, how does Wasow's proposal
resemble or differ from the verb disposition hypothesis set
forth by Stallings, MacDonald, & O'Seaghdha (in press) in
chapter 4?
V—PP Collocations and. Weight
Wasow's analysis of V-PP collocations considers evidence for
a speaker versus listener-based account of weight effects, but
other than comparing the number of heavy-NP shift structures
with the type of collocation, the impact of weight is never
directly assessed. To fully consider the interaction between
weight and V PP collocations it is important to know what the
length differences between the NP and PP are when the V PP
collocations appear in heavy-NP shift structures.
In Hawkins (submitted) , for example, it becomes clear how
predictions about dependencies and EIC might interact in the
analysis of 125 verb-particle (Pt) structures. First, he sets
1 4 1
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
out the EIC predictions for [V Pt NP] versus [V NP Pt]
ordering. When the length of the NP and particle are equal,
80% of the structures appear in the [V NP Pt] order, but as the
length of the NP increases, [V Pt NP] becomes more strongly
realized. In addition to testing EIC predictions, Hawkins
compared the kinds of dependencies that certain particles hold
with their verbs and compared the extent to which the particles
prefer to appear in verb-adjacent position.
Particles were divided into two categories. The first
category is illustrated in (27), where the particle is assigned
the same meaning as if it were not in that context. The
dependency between the meaning of the verb and particle in this
case is weak and so the particle is assigned independent status
(PTi). On the other hand, the particle in (28) is dependent on
the verb for its meaning, so that the dependency relationship
is a bit stronger (PTd). Thus, in the PTd case, lexical
dependency defines an adjacency preference for V-PTd-NP
structures, such that PTd particles will be more likely no
appear adjacent to the verb than PTi particles. In fact, PTd
particles appeared adjacent to the verb in 94% (59) of the
cases, whereas PTi particles appeared adjacent to the verb in
only 45% of the cases.
Rather than just comparing weight and adjacency predictions
separately, Hawkins compared the two against one another. In
doing so, he demonstrated that the pattern of the PTd cases was
highly consistent with EIC. In the four PTd cases that
appeared non-adjacent to the verb, the NPs contained only one
142
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
word, so that the NP and particle length were equal. The fact
that the V-NP-PTd structure is attested in 80% of the cases
when the NP and particle lengths are equal is due to a
combination of EIC and lexical dependency factors. For PTd
particles, EIC defines an equal preference for both V-NP-PTd
and V-PTd-NP structures when the lengths are equal. As stated
above, lexical dependency defines an adjacency preference for
V-PTd-NP. Consequently, V-NP-PTd is only attested when the
lengths are equal, not when the length of the NP is greater
than the PTd. Similarly, Hawkins attributes the preference for
V-NP-PTi structures when the NP and particle lengths are equal
to another factor. Namely, a preference to place the topic
before the predicate. The comparison of EIC predictions with
PTi particles illustrates that many of the structures contain
only short NPs, precisely when the preference for [V Pt NP] is
not terribly strong (i.e., when EIC is weak).
27a. He lifted up the child.
27b. He lifted the child up.
28a. He looked up the word.
28b. He looked the word up.
(From Hawkins, submitted, p. 34)
Unfortunately, none of Wasow's actual stimuli are provided,
so there is no way to test whether weight and the influence of
V PP collocations interact in a similar manner. Following
Hawkins' dependency strength predictions, we might predict that
the opaque collocations are more dependent on their verb for
meaning than are the transparent collocations and thus would
1 4 3
with permission of the copyright owner. Further reproduction prohibited without permission
necessitate smaller weight differences to induce shifting.
That is, the more dependent the PP on the verb for meaning, the
more likely it will be to appear in the shifted order, even if
the weight differentials are not significantly large. Future
research could explore the extent to which these predictions
are actually realized.
V-PP Collocations and. Verb Disposition
The V-PP collocation work by Wasow (1997a; 1997b) and
dependency relation predictions by Hawkins (submitted) both
assume the importance of adjacency between the verb and the PP.
These accounts, therefore, differ from the verb disposition
hypothesis reported by Stallings, MacDonald & O'Seaghdha (in
press) which stresses the role of non-adjacency between a verb
and its complement. If we consider these proposals from an
incremental view of production, the frequency of co-occurrence
of adjacent phrases undoubtedly plays an important role. For
example, the frequent production of collocations such as keep
in mind could make the PP in mind quite accessible when the
verb keep is readied for production. Similarly, particles that
are dependent on the verb for meaning become more accessible
once the verb is selected than do PTi particles.
It is not as clear how an incremental model would directly
support verb disposition findings unless a competitive
component were also involved. The plausibility of such a
component is explored in the next section, Multiple Factors in
Production. But first, it is important to provide additional
support for the non-adj acency hypothesis.
144
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission
Forthcoming corpus analyses (MacDonald, Stallings, &
O'Seaghdha, 1998) indicate a definite role for non-adjacency in
shifting as well. A simple adjacency approach predicts that
optionally intransitive verbs, such as run, move, walk, drive,
etc., which frequently occur adjacent to PPs such as to the
store, should have a disposition to shift by virtue of the
frequent V-PP co-occurrence. The non-adjacency account does
not make this prediction, because the V-PP sequences in
intransitive sentences such as Mary walked to the store do not
contain the requisite non—adjacent verbal complement. In fact,
MacDonald et al. (1998) found that optionally intransitive
verbs appear in shifted structures extremely rarely, suggesting
that the mere co-occurrence of verbs and adjacent PPs does not
fully determine a verb's shifting disposition. Collocations
and dependencies undoubtedly exert an influence on the
production of heavy-NP shift structures, but they fail to
explain the different shifting patterns between NP/S and NP-
only verbs and the rare appearance of optionally intransitive
verbs in shifted structures. At the very least, then, a four-
factor approach to heavy-NP shift is reasonable.
Additional factors nr Bbodoctlos
As argued above, the effects of relative weight described in
chapter 4 are compatible with incremental models of production,
as are proposals like those by Wasow and Hawkins. Evidence
that weight effects influence heavy-NP shift ordering during
positional level processes is substantiated by the similarities
between the intuitive driving force behind heavy-NP shift
1 4 5
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
(i.e., weight) and the defining features of processes at the
positional level. Based on the error patterns noted at the
beginning of chapter 3, constituents are ordered and placed
into word and morphological slots at the positional level,
which is constrained more by prosodic factors such as phonetic
form, length, and stress. These features correspond with
heavy-NP shift ordering, which is intuitively driven by length,
and perhaps influenced as well by stress (F. Ferreira, 1993),
but unaffected by grammatical role assignment (which is
assigned at the functional level).
Further evidence that heavy-NP shift ordering occurs at the
positional level of processing derives from the fact that sound
exchanges are more often bound within a phrase, whereas word
exchanges appear more frequently between phrases. Since sound
exchange errors occur at the positional level processes, the
phrasal ordering properties of heavy-NP shift are also accorded
positional level status.
Assignment of heavy-NP shift to processes at the positional
level alone, however, may limit the number of available
explanations for heavy-NP shift to weight-related features.
Such assignment would therefore ignore the evidence for verb
disposition effects as well as other potential factors like
pragmatics, which are typically made accessible at an earlier
stage or production. Experimental evidence in favor of both
weight and verb disposition effects suggests that heavy-NP
shift ordering cannot be adequately accounted for at the
positional level of processing alone. While it is clear how
1 4 6
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
positional processes would take into account the presence of
long NP complements or prosodically prominent complements, for
example, it is not immediately obvious where non-adjacency
history or pragmatic factors would be represented in such a
model. This is especially true for a case like heavy-NP shift,
where phrasal ordering is free and occurs independently of
grammatical role assignment, which occurs at the functional
level of processing. Thus, the challenge to production models
is to account for the possible influence of pragmatics on
structures that are free of grammatical role assignment as well
as for lexical semantic factors like verb disposition that also
influence the spoken production of heavy-NP shift. To address
the latter, existing models do maintain competitive components,
but they do not clearly delineate the extent to which
competition effects ordering. In light of the verb disposition
effects and the performance approach being advocated here, the
next section reviews the role of competition in three
production models.
The role of competition in production models
Garrett's (1980) model recognizes a competing plans
hypothesis to some degree, but it does not delineate the
circumstances under which competition would influence
production. Moreover, his model fails to allow for feedback
from the positional to the functional level (see Bock, 1987a
for review), and hence, assumes more of a transformational
approach to production rather than one that is motivated solely
by performance as advocated in chapter 2.
147
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
The underlying assumptions in De Smedt's (1994a; 1994b)
models are similar in that they also reject feedback to
previous levels within the formulator. However, in the latter
incremental activation-based model, the role of competition is
clearly laid out. This model is an extension of his (1994b)
incremental parallel formulator (IPF) in which the
conceptualizer, formulator, and articulator operate
independently and in parallel. According to De Smedt, this
kind of process does not allow surface features such as weight
to influence syntactic form. Upon recognition of a phrasal
head, IPF immediately integrates it into the structure,
allowing no time to process or integrate length. His current
model, therefore, includes a mechanism by which surface level
features such as length compete for activation (as measured in
short term memory, see Kempen & Vosse, 1989 for review, cited
in De Smedt, 1994). He argues that the heavy constituents
require more time to process. Consequently, shorter items are
more accessible for early position assignment and heavy items
appear in clause-final position. Although this process
undermines the advantages of incremental production, De Smedt
argues that the memory load for the speaker is reduced via the
early production of the more accessible lighter item.
This approach is compatible with the notions of
accessibility discussed in chapter 3 where conceptually and
lexically available items are more accessible for positioning
early on in a structure. For De Smedt, the relationship
between conceptual accessibility and weight is a negative one.
1 4 8
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Shorter constituents are more likely to constitute conceptually
accessible (and salient) items, whereas heavier items tend to
be less readily accessible. Interestingly though, the
competition in this model is not directly tied to lexical-
semantic factors, but instead to length. Whether lexical-
semantic factors or at all plausible in this model is unclear.
In addition to length and conceptual accessibility, De Smedt
specifies only animacy as one of several additional factors
that may compete to influence syntactic form.
Bock (1986a) proposes an alternative model that is more
compatible with a performance approach to ordering. Similar to
Garrett's model, it assumes that both functional and positional
levels are involved. Unlike the previous two models, however,
this model integrates Dell's (1986) lexical interaction
principle so that feedback between levels is allowed. In
addition, a competitive component is clearly evident within the
functional level, where Bock argues conceptual accessibility as
well as other processes contribute to the grammatical functions
that are ultimately assigned to a word. One of the other
processes that Bock mentions may shed light on the verb
disposition issue. Bock argues that the strength of
alternative verb forms in conjunction with the conceptual
accessibility and the thematic role structure of an item will
influence syntactic form. The decision to produce an active
versus a passive structure, for example, demonstrates this
interaction nicely. At the beginning of the production
process, accessibility of the active form is highly activated.
149
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
As additional information is processed, if the conceptual
accessibility of the patient is significantly higher them the
agent, so that early placement of the patient is preferred, the
active structure will begin to lose activation in favor of the
passive structure where the patient assumes subject position.
Applied to heavy-NP shift, verbs would conceivably activate
basic and shifted structures at the functional integration
level. As the weight of the NP and PP are accessed at the
positional level, the accessibility of the NP and the PP are
fed back to the functional level which would result in the
higher activation of one structure over another. A potential
problem with this account is that heavy-NP shift is unaffected
by grammatical role assignment, a fundamental component of
functional level processes. Bock & Warren (1985), showed that
when grammatical roles did not differ (i.e., in conjunctive
phrases), the concreteness of an item did not affect ordering.
Thus, it remains to be seen if alternative verb structures
would compete in the same way at the functional level for
heavy—NP shift ordering, where grammatical role assignment is
unaffected.
F u t u r e D i r e c t i o n s
In sum, relative weight, not properties of the NP alone,
influences the production of heavy-NP shift structures.
Evidence exists, however, that weight is only one factor within
a broader theory of phrasal ordering. In the course of
establishing the nature of weight effects in heavy—NP shift, we
1 5 0
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
have also begun to identify the influence of additional factors
on the production of heavy-NP shift structures.
Future research issues should focus on answering questions
about the precise production mechanisms that affect constituent
ordering and the interactive nature of the multiple constraints
that influence ordering. For instance, are the effects found
in this study limited to structures with free phrasal ordering?
Another important area of research concerns the issue of
production models themselves. Future endeavors in incremental
models of production should focus on clarifying the extent to
which competitive processes exert influence within the model.
Are they limited to the functional level, or are they
distributed throughout the model? Moreover, if an interactive
incremental account can adequately account for heavy-NP shift
data, the next challenge for such models is to account for
shifting in all languages, in particular why the opposite long
before short pattern occurs in Japanese (Hawkins, 1994).
Research must also continue to explore the interactive
nature of multiple factors. Hawkins (1994) proposes that "non—
EIC" factors are more likely to motivate shifting when the
relative weight differences are small, but the precise
interactions between weight and these factors is yet to be
determined. His (submitted) consideration of the interactive
nature between relative weight and the dependency relations
that a particle has with its verb in verb-particle movement
structures is, however, instrumental in achieving that goal.
Recall that as soon as the NP exceeded the dependent particle
1 5 1
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
by just one word, the dependent particle appeared in verb-
adjacent position in 100% of the cases (versus 33% of the time
when the NP and Particle were equal in length) . As the
relative weight increased, therefore, the dependencies also
increased.
This correlation between relative weight and dependency
strength works well within a model of incremental production.
Recall the findings by Bock & Irwin (1980) that were presented
in chapter 3. When an item was referentially and lexically
accessible, the placement of the item early in the sentence was
more likely than when the item was neither referentially nor
lexically accessible or was only referentially accessible, for
instance. A plausible extension of these findings would
suggest that a particle that is dependent on its verb is highly
accessibile at the outset. As the weight of NP increases
relative to the weight of the particle, the amount of
information that needs to be processed to increase the
accessibility of the NP will also increase. As a result,
access of the NP will likely be delayed and the dependency
status of the particle is made even more accessible.
In Stallings, MacDonald, & O'Seaghdha (in press), a
significant interaction between length and verb type obtained,
such that NP/S verbs were most likely to shift in the long NP
condition. In addition, the presence of an NP/S verb
facilitated the production of heavy—NP shift structures.
Production of a shifted utterance occurs much more quickly when
a verb that frequently appears non-adjacent to its complements
152
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
encounters a long NP than if the verb more frequently appears
adjacent to its complements. What is unknown, however, is the
degree to which NP/S verbs shift as the relative weight between
the NP and PP increases or decreases.
One approach to establishing the level of interaction
between these two factors is to compare the distribution of NP-
only and NP/S verbs in shifted structures as relative weight
increases. When the NP and PP are close in length, the impact
of verb type on heavy-NP shift ordering should be stronger than
when the relative weight between the cwo constituents is large.
The coming availability of spoken corpus data will be a
valuable resource through which such claims about the
interactive nature between constraints can be tested.
In conclusion, evidence provided in the preceding chapters
challenges a single-factor grammatical explanation of heavy-NP
shift. Instead, a performance-motivated approach to
constituent ordering in the spoken production of heavy-NP shift
is supported, suggesting that at the very least lexical
semantic factors such as verb disposition and phonological
factors such as weight constrain shifting. The influence of
relative weight on the production of heavy-NP shift structures
may further support the two levels of processing assumption in
Bock (1986a), though the precise mechanism by which such models
operate must be clarified.
153
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
References
Andersen, E. S. (1990) . Speaking with style: The
sociolinguistic skills of children. London: Routledge.
Arnold, J., Losongco, A., Ginstrom, R., Brynolfson, A., 4
Wasow, T. (1997). Save the worst for last: The effects
of syntactic complexity and information structure on
constituent ordering. Chicago, IL. : Paper presented at
the Linguistic Society of America Conference.
Bates, E., & Devescovi, A. (1989). Crosslinguistic studies
of sentence production. In B. MacWhinney & E. Bates
(Eds.), The crosslinguistic study of sentence processing.
New York: Cambridge University Press.
Behaghel, O. (1932). Deutsche syntax: Eine geschichtliche
darstellung. Wortstellung. Periodenbau: Heidelberg:
Carl Winters.
Bever, T. G. (1970). The cognitive basis for linguistic
structures. In J. R. Hayes (Ed.), Cognition and the
development of language. New York: Wiley.
Bock, J. K. (1982). Toward a cognitive psychology of
syntax: Information processing contributions to sentence
formulation. Psychological Review, 89, 1-47.
Bock, J. K. (1986a). Exploring levels of processing in
sentence production. In G. Kempen (Ed.), Natural
language generation. New results in artificial
intelligence (pp. 351-363). Boston: Martinus Nijhoff.
Bock, J. K. (1986b) . Meaning, sound, and syntax: Lexical
priming in sentence production. Journal of Experimental
Psychology: Learning, Memory and Cognition, 12, 575-586.
Bock, J. K. (1986c). Syntactic persistence in language
processing. Cognitive Psychology, 18, 355-387.
Bock, J. K. (1987a). Coordinating words and syntax in
speech plans. In A. Ellis (Ed.), Progress in the
psychology of language. London: Erlbaum.
Bock, J. K. (1987b). An effect of the accessibility of word
forms on sentence structures. Journal of Memory and
Language, 26, 119-137.
Bock, J. K. (1990). Structure in language: Creating form in
talk. American Psychologist, 45, 1221-1236.
Bock, J. K., & Irwin, D. E. (1980). Syntactic effects of
information availability in sentence production. Journal
of Memory and Language, 19, 467-484.
154
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Bock, J. K., & Levelt, W. (1994). Language production:
Grammatical encoding. In M. A. Gemsbacher (Ed.),
Handbook of Psycholinguistics. Hillsdale, NJ: Erlbaum.
Bock, J. K., Loebell, H., & Morey, R. (1992). From
conceptual roles to structural relations: Bridging the
syntactic cleft. Psychological Review, 99, 150—171.
Bock, J. K., & Warren, R. K. (1985). Conceptual
accessibility and syntactic structure in sentence
formulation. Cognition, 21, 47-67.
Bresnan, J. (1976). Nonarguments for raising. Linguistic
Inquiry, 7, 485-501.
Brown, P. M. D., G.S. (1987) . Adapting comprehension to
production: The explicit mention of instruments.
Cognitive Psychology, 19, 441-472.
Caplan, D. (1992). Language: Structure, processing, and
disorders. Cambridge: The MIT Press.
Chafe, W. (1987). Cognitive constraints on information
flow. In R. S. Tomlin (Ed.), Coherence and Grounding in
Discourse. Amsterdam: John Benjamins.
Chafe, W. L. (1977). Creativity in verbalization and its
implications for the nature of stored knowledge.
Norwood: Ablex.
Chomsky, N. (1965) . Aspects of the theory of syntax.
Cambridge, MA: MIT Press.
Clark, H. H. 4 Haviland, S. E. (1977). Comprehension and
the given-new contract. In R. 0. Freedle (Ed.),
Discourse production and comprehension: Volume 1.
Advances in discourse processes (pp. 1-40) . Norwood NJ:
Ablex Publishing Corporation.
Cooper, W. E., & Ross, J. R. (1975). World order. Paper
presented at the Papers from the Parasession on
Functionalism, Chicago, IL.
De Smedt, K. (1994a, July 25-29) . Dragging heavy
constituents around. Paper presented at the Dagstuhl
Workshop on Principles of Natural Language Generation,
Dagstuhl.
De Smedt, K. (1994b) . Parallelism in incremental sentence
generation. In G. Adriaens & U. Hahn (Eds.),
Parallelism in Natural Language Processing : Ablex.
De Smedt, K. J. M. J. (1990) . IPF: An incremental parallel
formulator. In R. Dale, C. Mellish, 4 M. Zock (Eds.),
Current research in natural language generation (pp. 167-
192). London: Academic Press.
Dell, G. S. (1986). A spreading-activation theory of
retrieval in sentence production. Psychological Review,
93, 283-321.
1 5 5
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Dell, G. S., & Brown, P. M. (1991). Mechanisms for
listener-adaptation in language production: Limiting the
role of the "model of the listener". In D. J. Napoli &
J. A. Kegl (Eds.), Bridges between Psychology and
Linguistics: A Swarthmore festschrift for Lila Gleitman
(pp. 105-129). Hillsdale: Lawrence Erlbaum.
Dell, G. S., & O'Seaghdha, P. G. (1992). Stages of lexical
access in language production. Cognition, 42, 287-314.
Dik, S. C. (1978) . Functional grammar. New York: North
Holland Publishing Company.
Dik, S. C. (1989) . The theory of functional grammar. Part
I: The structure of the clause. Dordrecht: Foris.
Emonds, J. (1976). A transformational approach to English
syntax. New York: Academic Press.
Erdmann, P. (1988). On the principle of 'weight* in
English. In C. Duncan-Rose & T. Vennemann (Eds.), On
language, rhetorics phonologies syntactics: A
festschrift for Robert P. Stockwell from his friends and
colleagues. London: Routledge.
Ferreira, F. (1991). Effects of length and syntactic
complexity on initiation times for prepared utterances.
Journal of Memory and Language, 30, 210-233.
Ferreira, F. (1993). Creation of prosody during sentence
production. Psychological Review, 2, 233-253.
Ferreira, F. (1994). Choice of passive voice is affected by
verb type and animacy. Journal of Memory and Language,
33, 715-736.
Ferreira, F., & Clifton, C. (1986). The Independence of
Syntactic Processing. Journal of Memory and Language,
25, 348-368.
Ferreira, F., & McClure, K. K. (1997) . Parsing of garden-
path sentences with reciprocal verbs. Special Issue of
Language and Cognitive Processes, 12(2/3), 273-306.
Ferreira, V. (1996). Is it better to give than to donate?
The consequences of syntactic flexibility in language
production. Journal of Memory and Language, 35, 724-755.
Ferreira, V., & Dell, G. S. (1996). Do speakers choose
their words cooperatively? Investigating the production-
comprehension interface. Chicago, IL: Poster presented
at the 37th Annual Meeting of the Psychonomics Society.
Ferguson, C. A. (1977). Babytalk as simplified register. In
C. E. Snow & C. A. Ferguson (Eds.), Talking to children
(pp. 209-235). Cambridge: Cambridge University Press.
Firfaas, J. (1966). On defining the theme in functional
sentence perspective. Travaux Linguistiques de Prague,
2.
156
permission of the copyright owner. Further reproduction prohibited without permission.
Frazier, L. (1985). Syntactic complexity. In D. Dowty, L.
Karttunen, £ A. Zwicky (Eds.), Natural language parsing:
Psychological, computational, and theoretical
perspectives . Cambridge: Cambridge University Press.
Frazier, L., & Rayner, K. (1987). Resolution of syntactic
category ambiguities: Eye movements in parsing lexically
ambiguous sentences. Journal of Memory and Language, 26,
505-526.
Gamsey, S. M., Pearlmutter, N. J., Myers, E., £ Lotocky, M.
A. (1997). The contributions of verb bias and
plausibility to the comprehension of temporarily
ambiguous sentences. Journal of Memory and Language, 31,
58-93.
Garrett, M. F. (1975). The analysis of sentence production.
In G. H. Bower (Ed.), The psychology of learning and
motivation. New York: Academic Press.
Garrett, M. F. (1976). Syntactic processes in sentence
production. In R. J. Wales £ E. Walker (Eds.), New
approaches to language mechanisms. Amsterdam: North
Holland.
Garrett, M. F. (1980) . Levels of processing in sentence
production. In B. Butterworth (Ed.), language
production. London: Academic Press.
Garrett, M. F. (1982). Production of speech: Observations
from normal and pathological language use. In A. Ellis
(Ed.), Normality and Pathology in Cognitive Functions.
London: Academic Press.
Garrett, M. F. (1988) . Processes in sentence production.
In F. Newmeyer (Ed.), The Cambridge Linguistics Survey:
Biological and social factors. Vol. 3. New York:
Cambridge University Press.
Givon, T. (1988) . The pragmatics of word order:
Predictability, importance, and attention. In M.
Hammond, E. A. Moravcsik, £ J. Wirth (Eds.), Studies in
syntactic typology. Amsterdam: John Benjamins.
Gordon, P. C., Grosz, B. J., £ Gilliom, L. A. (1993) .
Pronouns, names, and the centering of attention in
discourse. Cognitive Science, 17, 311-347.
Grice, H. P. (1975) . Logic and conversation. In P. Cole £
J. L. Morgan (Eds.), Syntax and semantics: Speech acts
(Vol. 3). New York: Academic Press.
Gundel, J. K., Houlihan, K., £ Sanders, G. (1988). On the
function of marked and unmarked terms. In M. Hammond, E.
Moravcsik, £ J. Wirth (Eds.), Studies in Syntactic
Typology . Amsterdam: John Benjamins.
Haegeman, L. (1991) . Introduction to Government £ Binding
Theory. Cambridge, MA: Blackwell.
157
permission of the copyright owner. Further reproduction prohibited without permission.
Haviland, S. E. C., H.H. (1974). What's new? Acquiring
information as a process in comprehension. Journal of
Verbal Learning and Verbal Behavior, 13, 512-521.
Hawkins, J. A. (1994) . A performance theory of order and
constituency. Cambridge, England: Cambridge University
Press.
Hawkins, J. A. (1997). Some issues in a performance theory
of word order. In A. Siewierska (Ed.), Constituent Order
in the Languages of Europe (EALT) . Berlin: Mouton de
Gruyter.
Hawkins, J. A. (submitted). Adjacency, linear precedence,
and processing efficiency in performance and grammars.
Journal of Linguistics.
Horton, W. S., & Keysar, B. (1996). When do speakers take
into account common ground? Cognition, 59, 91-117.
Kawamoto, A. (1993). Nonlinear dynamics in the resolution
of lexical ambiguity: A parallel distributed processing
account. Journal of Memory and Language, 32, 474-516.
Kayne, R. S. (1994). The Antisymmetry of syntax.
Cambridge, MA: The MIT Press.
Kelly, M. H., Bock, K. , & Keil, F. C. (1986).
Prototypicality in a linguistic context: Effects on
sentence structure. Journal of Memory and Language, 25,
59-74.
Kempen, G., & Hoenkamp, E. (1987). An incremental
procedural grammar for sentence formulation. Cognitive
Science, 11, 201-258.
Kimball, J. (1973). Seven principles of surface structure
parsing in natural language. Cognition, 2, 15-47.
Lahiri, L. W. A. (1997). Prosodic units in speech
production. Journal of Memory and Language, 37, 356-381.
Larson, R. K. (1988). On the double object construction.
Linguistic Inquiry, 19, 335-391.
Levelt, W., & Maassen, B. (1981). Lexical search and order
of mention in sentence production. In W. Klein & W.
Levelt (Eds.), Crossing the Boundaries in Linguistics:
Studies Presented to Manfred Bierwisch. Boston: D.
Reidel.
Levelt, W. J. M. (1989) . Speaking: From intention to
articulation. Cambridge, MA: MIT Press.
MacDonald, M. C. (1993). The interaction of lexical and
syntactic ambiguity. Journal of Memory and Language, 32,
692-715.
MacDonald, M. C. (1994). Probabilistic constraints and
syntactic ambiguity resolution. Language and Cognitive
Processes, 9, 157-201.
1 5 8
permission of the copyright owner. Further reproduction prohibited without permission.
MacDonald, M. C., Pearlmutter, N. J-, & Seidenberg, M. S.
(1994) . The lexical nature of syntactic ambiguity
resolution. Psychological Review, 101(A).
MacDonald, M. C., Stallings, L. M., & O'Seaghdha, P. G.
(1998). Distributional information and heavy-NP shift.
Unpublished Manuscript.
MacWhinney, B. (1977). Starting points. Language, 53(1),
152-168.
MacWhinney, B., & Bates, E. (1989). The crosslinguistic
study of sentence processing. Cambridge: Cambridge
University Press.
McCawley, J. D. (1982). Parentheticals and discontinuous
constituent structure. Linguistic Inquiry, 13, 91-106.
McCawley, J. D. (1987). Some additional evidence for
discontinuity. In G. J. Huck & A. E. Ojeda (Eds.),
Discontinuous constituency, syntax and semantics. New
York: Academic Press.
McDonald, J., Bock, K., & Kelly, M. H. (1993). Word and
world order: Semantic, phonological, and metrical
determinants of serial position. Cognitive Psychology,
25, 188-230.
Miller, G. A., & Chomsky, N. (1963) . Finitary models of
language users. In R. D. Luce, R. R. Bush, & E. Galanter
(Eds.), Handbook of mathematical psychology. New York:
Wiley.
Nespor, M., & Vogel, I. (1986). Prosodic phonology.
Dordrecht, The Netherlands: Foris.
O'Grady, W. (1987). Discontinuous constituents in a free
word order language. In G. J. Huck & A. E. Ojeda (Eds.),
Discontinuous constituency, syntax and semantics. New
York: Academic Press.
Osgood, C. E., & Bock, J. K. (1977). Salience and
sentencing: Some production principles. In S. Rosenberg
(Ed.), Sentence production: Developments in research and
theory. Hillsdale: Lawrence Erlbaum.
Pesetsky, D. (1995). Zero syntax: Experiencers and
cascades. Cambridge: The MIT Press.
Pinker, S., & Birdsong, D. (1979). Speakers' sensitivity to
rules of frozen word order. Journal of Verbal Learning
and Verbal Behavior, 18, 497-508.
Postal, P. (1974). On raising: One rule of English grammar
and its theoretical implications. Cambridge, MA.: The
MIT Press.
Primus, B. (1994). The role of grammatical relations in
word order universals. Working Paper 4: Theme Group 2:
Constituent Order Coordination.
159
with permission of the copyright owner. Further reproduction prohibited without permission.
Quirk, R., Greenbaum, S., Leech, G., & Svartvik, J. (1972).
A grammar of contemporary English. London: Longman.
Rickford, J. R., Wasow, T. A., Mendoza-Denton, N., &
Espinoza, J. (1995). Syntactic variation and change in
progress: Loss of the verbal coda in topic-restricting as
far as constructions. Language, 71, 102-131.
Rochemont, M. S. (1978). A theory of stylistic rules in
English. New York: Garland Press.
Rochemont, M. S., & Culicover, P. W. (1990). English focus
constructions and the theory of grammar. New York:
Cambridge University Press.
Ross, J. R. (1967). Constraints on variables in syntax.
Unpublished PhD Dissertation, MIT.
Schwartz, M. F., Saffran, E. M-, Bloch, D. E., & Dell, G. S.
(1994). Disordered speech production in aphasic and
normal speakers. Brain and Language, 47, 52-88.
Selkirk, E. O. (1980). On prosodic structure and its
relation to syntactic structure. Indiana: Indiana
University Linguistics Club.
Siewierska, A. (1991). Functional grammar. London:
Routledge.
Stallings, L. (1994). Performance explanations and
predictions in word order. Unpublished Screening Paper,
University of Southern California, Los Angeles.
Stallings, L., MacDonald, M. C., & O'Seaghdha, P. G. (in
press). Phrasal ordering constraints in sentence
production: Phrase length and verb disposition in heavy-
NP shift. Journal of Memory and Language.
Steedman, M. J., & Johnson-Laird, P. N. (1980) . The
production of sentences, utterances, and speech acts:
Have computers anything to say? In B. Butterworth (Ed.),
Language production: speech and talk. San Francisco:
Academic Press.
Stemberger, J. P. (1985). An interactive activation model
of language production. In A. W. Ellis (Ed.), Progress
in the psychology of language (Vol. 1, pp. 143—187) .
Hillsdale: Lawrence Erlbaum.
Sternberg, S., Monsell, S., Knoll, R. L., & Wright, C. E.
(1978). The latency and duration of rapid movement
sequence: Comparisons of speech and typewriting. In G.
E. Stelmach (Ed.), Information processing in motor
control and learning . New York: Academic Press.
Stockwell, R. P. (1977) . Foundations of syntactic theory.
Englewood Cliffs, NJ: Prentice-Hall Inc.
160
permission of the copyright owner. Further reproduction prohibited without permission.
Tanenhaus, M. K. (1989) . Psycholinguistics: An overview.
In F. Newmeyer (Ed.), Language: Psychological and
biological aspects (Vol. Ill, pp. 1-37). New York:
Cambridge University Press.
Tanenhaus, M. K., & Trueswell, J. C. (1995). Sentence
Comprehension. In J. L. Miller & P. D. Eimas (Eds.),
Speech, language, and communication. San Diego, CA:
Academic Press.
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.
Tyler, L. K., & Marslen-Wilson, W. D. (1977). The on-line
effects of semantic context on syntactic processing.
Journal of Verbal Learning and Verbal Behavior, 16, 683-
692.
Wasow, T. (1997a) . End-weight from the speaker's
perspective. Journal of Psycholinguistic Research,
26(3), 347-361.
Wasow, T. (1997b). Remarks on grammatical weight. Language
Variation and Change, 9(1), 81-105.
Wright, C. E. (1990). Controlling sequential motor
activity. In D. N. Osherson, S. M. Kosslyn, & J. M.
Hollerbach (Eds.), An invitation to Cognitive Science:
Visual cognition and action. Cambridge, MA: MIT Press.
Zee, D., & Inkelas, S. (1990). Prosodically constrained
syntax. In S. Inkelas & D. Zee (Eds.), The phonology-
syntax connection. Chicago: CSLI and the University of
Chicago Press.
161
permission of the copyright owner. Further reproduction prohibited without permission.
A p p e n d i x A
The following items include the subject NP, the NP-only and
NP/S verbs, the long (and short) NPs, and the animate/inanimate
PPs respectively. Each verb pair was used in two sentences.
The numbers at the end of each item correspond to the amount of
shifting that occurred when the item was presented with the NP-
only verb and with the NP/S verb.
1. The manager presented/exhibited the new line of bright
summer beach and resort fashions (the styles) to Jill/at
noon. 4/4
The contractor presented/exhibited the plans for a lagoon
and waterfall with red clay (the display) to Jamie/in July.
3/3
2. Mary returned/mentioned the used car ad for a Chevrolet
with low mileage (the ads) to Frank/at tea. 1/8
Mark returned/mentioned the flyers for some lectures and
slide shows on campus (some flyers) to Julie/very quickly.
2/11
3. The dealer transported/demonstrated a brand new speedy
silver sports car with chrome bumpers (a car) to Kate/at
ten. 2/3
The manager transported/demonstrated the new sports line of
running shoes and bicycling gear (sports gear) to Jim/at
three. 3/3
162
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
4. Robert relinquished/proposed one round trip plane ticket
from New York to Atlanta (plane tickets) to Lizzy/last
semester. 2/3
Donald relinquished/proposed some great front row center
seats for the last performance (the drafts) to Cole/with
joy. 4/8
5. Jake released/explained all of the facts for the import and
export taxes (the figures) to Carol/on Friday. 2/6
Brian released/explained all of the answers on the twenty
point English quiz (the quiz) to Ralph/at lunch. 3/6
6. Alex described/indicated each of the key points of conflict
in an interview (the offer) to Curtis/in detail. 11/7
The woman described/indicated the best routes to the Santa
Monica beach and pier (the route) to Tom/by phone. 3/9
7. The lawyer distributed/acknowledged each piece of the
evidence in the tax fraud cases (the dividends) to
everyone/with confidence. 4/6
The editor distributed/acknowledged the first drafts of a
long article and book review (the details) to Bonnie/at
supper. 1/0
8. The president introduced/announced the new vice chair of
the first corporate finance group (the captain) to Simon/on
Monday. 1/6
163
permission of the copyright owner. Further reproduction prohibited without permission.
The teacher introduced/announced some hard math concepts
for the one hundred point final (the test) to Kim/with
care. 3/7
9. Judy delivered/suggested a gorgeous red and black silk
dress with sparkling sequins (a dress) to Ann/in March. 2/3
Todd delivered/suggested a large package with lots of
presents for small children (the gifts) to Al/at once. 4/7
10. Shelly carried/disclosed top secret plans for a large fleet
of defense missiles (the notes) to Bev/in May. 3/3
The woman carried/disclosed the birth and school records of
the three young children (the records) to Nina/in August.
2/3
11. David furnished/reported the facts on the strange bank
robbery crimes in Utah (the facts) to Will/with hesitation.
1/4
Jessica furnished/reported the awful news of the refugees
in the Far East (the news) to Kris/last summer. 0/3
12. Steven addressed/communicated a seven page letter with many
details of the trip (a letter) to Leslie/during break. 2/5
Raymond addressed/communicated a five page plan of
withdrawal for the allied troops (a memo) to Sarah/very
loudly. 3/3
164
permission of the copyright owner. Further reproduction prohibited without permission.
13. Sam donated/recommended an old treasure chest with some
silver and gold trinkets (a treasure) to Alan/on Sunday.
1/5
The generous man donated/recommended some used toys, winter
clothing, clean linens, and snow boots (some clothes) to
Keith/in person. 0/8
14. Janet transferred/revealed some more specific plans for a
brand new defense plant (the graphs) to Leigh/at dawn. 1/4
Bobby transferred/revealed the short documents to the old
mansion on the hill (the deeds) to Joe/at dusk. 1/5
15. Kathy recited/dictated all three verses from an old and
famous French poem (the poem) to Peter/last night. 1/0
Mike recited/dietated the words to a popular country song
in the South (the words) to Ron/last year. 2/3
16. Amy forfeited/broadcasted the high school girl's state
swimming title in New York (the game) to Jan/last week.
0/4
Jason forfeited/broadcasted all of the winnings for the
past seven horse races (the race) to Jeff/with glee. 0/5
17. Josh dispatched/conveyed a short message on the dangers of
the current storm (a message) to Michelle/at breakfast. 1/3
Brad dispatched/conveyed the instructions for the safest
route to the old town (the thoughts) to Luke/at nine. 3/2
1 6 5
permission of the copyright owner. Further reproduction prohibited without permission.
18. The teacher narrated/confessed each of the series of events
in the secret case (the events) to Rachel/with caution.
5/7
Matthew narrated/confessed the whole story on the defects
in the new Mazda (the story) to Kenny/in secret. 2/9
19. Tim contributed/stated the small reward for helpful clues
on a recent crime (the reward) to Kelly/with reluctance.
4/9
Bob contributed/stated the earnings of the five winners in
the poker games (the earnings) to Ian/with pleasure. 3/8
20. Ed entrusted/muttered a script for a suspense film with a
giant budget (a script) to Fran/at brunch. 1/1
Angie entrusted/muttered the plan for the biggest poster in
the science fair (the plan) to Mel/with hope. 5/9
166
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
A p p e n d i x B
These items appeared in the PP Length Experiment. For each
subject verb phrase (e.g., the radio listeners accepted), a 10-
word NP and 3 PPs of varying lengths (2, 5 and 7 words) were
constructed.
They appear in the following order:
#. Subject verb phrase
2-word PP
5-word PP
8-word PP
10-word NP
1. The radio listeners accepted
without doubt
without any doubt or concern
without doubt or any bit of concern
the whole story on the defects in the new Mazda
2. The executive announced
at lunch
at the weekly dinner meeting
at the annual picnic for company staff
five of the winners in the new ad campaign contest
3. The Senate approved
with joy
with great joy and enthusiasm
with a great deal of vocal enthusiasm
the new and much debated bill on health care reform
167
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
4. The paper assessed
in detail
in a very detailed manner
in a detailed and well-researched manner
the harmful effects of smoke on both kids and adults
5. The suspect confessed
in public
in front of the media
in the middle of the press conference
each of the series of events in the secret case
6. The lawyer considered
with trepidation
with much thought and trepidation
with a lot of regret and trepidation
each of the key points of conflict in the interview
7. The angry wife demanded
in court
in the heated court battle
in the heated and tiring court battle
custody of the kids a new car and more money
8. Lieutenant Schultz disclosed
with caution
with some caution and concern
with a note of caution and concern
the top secret notes about the defense plant in Guam
9. The scientist discovered
by coincidence
by quite a strange coincidence
by quite a strange and lucky coincidence
the gene sequence for a rare form of colon cancer
permission of the copyright owner. Further reproduction prohibited without permission.
10. The student read
during class
during her Spanish Lit class
during her second class of the morning
three short reports about the crime rates in big cities
11. The Supreme Court established
with opposition
with a bit of opposition
with only a small bit of opposition
a new law on the adoption rights of single dads
12. The textbook explained
in brief
in the first edition draft
in a brief summary of the chapter
the names and locations of rare plant and bird species
13. The family heard
at midnight
at a quarter to eleven
at about half an hour before sunrise
rustling noises outside the back door porch of the house
14. Congress included
as expenditures
as some of the expenditures
as some of the few allowed expenditures
a budget for the loss of farm crops and equipment
15. The actress mentioned
on television
on a television talk show
on quite a popular television talk show
the names of two of her most recent movie projects
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
16. The doctor noted
with care
with an acutely trained eye
with a worrisome furrow in her brow
the tiny Tump on the end of her patient's elbow
17. The driver noticed
with horror
with quite a horrified look
with a horrified expression in his eyes
a dent in the rear bumper of his new Ferrari
18. The scouts observed
at camp
at the busy National Park
at the busy but beautiful National Park
a famous array of geysers hot springs and small lakes
19. The investigator omitted
from reports
from the latest police reports
from the district attorney's latest crime report
two pieces of the evidence in the tax fraud case
20. The mystic predicted
with confidence
with a show of confidence
with a show of confidence and bravado
the exact locations and dates of the next two floods
21. The government representative provided
on Monday
on the nineteenth of July
on the final day of the convention
some of the facts about the new market in Europe
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
22. The professor questioned
during class
during the guest lecturer's talk
during the guest speaker's highly publicized talk
the strength of the new theory and data on cloning
23. The teacher discussed
in detail
in a very forceful manner
in a direct and rather offensive manner
a paper on civil rights in the past four decades
24. The dancer realized
with disappointment
with disappointment and some shock
with disappointment and a bit of shock
her grave mistake during the last part of the routine
25. The producer recognized
at work
at a busy coffee bar
at an event to benefit AIDS research
the famous actress from the hit movie First Wives Club
26. The parking attendant recommended
with pleasure
with pleasure and genuine warmth
with pleasure and a very friendly smile
a classy four-star hotel near the John Wayne airport
27. The mayor reported
with regret
with some regret and sorrow
with some regret and a little hesitation
the hard facts about seven bank thefts and five murders
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
28. The restaurant owner revealed
with pride
with a show of enthusiasm
with a show of pride and enthusiasm
the cooking talents of the chef and his kitchen staff
29. The administrators reviewed
with parents
with many of the parents
with many of the parents and faculty
the guidelines for the first through third grade reading
program
30. The report stated
with authority
with authority and some facts
with authority and quite a few facts
the absence of any drugs or weapons in the home
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
A p p e n d i x C
These items appeared in the Relative Weight Experiment. For
each subject verb phrase (e.g., the radio listeners accepted)
2 PPs (2 and 5 words) were constructed and 3 NPs (10, 7, and
words) were constructed.
The stimuli appear in the following order:
#. Subject verb phrase
2-word PP
5-word PP
10-word NP
7-word NP
2-word NP
1. The radio listeners accepted
without doubt
without any doubt or concern
the whole story on the defects in the new Mazda
the whole story on the recent defects
the story
2. The executive announced
at lunch
at the weekly dinner meeting
five of the winners in the new ad campaign contest
five winners in the new ad contest
five winners
3. The Senate approved
with joy
with great joy and enthusiasm
the new and much debated bill on health care reform
the much debated health care reform bill
the bill
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
4. The paper assessed
in detail
in a very detailed manner
the harmful effects of smoke on both kids and adults
the harmful effects of second hand smoke
the risks
5 . The suspect confessed
in public
in front of the media
the series of planned events in the secret crime case
the series of events in the case
the details
6. The lawyer considered
with trepidation
with much thought and trepidation
each of the key points of conflict in the interview
each of the key points of conflict
each point
7. The angry wife demanded
in court
in the heated court battle
more money, custody of the kids, and a new car
more money and custody of the kids
more money
8. Lieutenant Schultz disclosed
with caution
with some caution and concern
the top secret notes about the defense plant in Guam
the secret notes about the defense plant
the notes
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
9. The scientist discovered
by coincidence
by quite a strange coincidence
the gene sequence for a rare form of colon cancer
the rare gene sequence for colon cancer
the gene
10. The student read
during class
during her Spanish Lit class
three short reports about the crime rates in big cities
three short reports about recent crime rates
three reports
11. The Supreme Court established
with opposition
with a bit of opposition
a new law on the adoption rights of single dads
a law on gay couple adoption rights
a law
12. The textbook explained
in brief
in the first edition draft
the names and locations of rare plant and bird species
the names of plant and bird species
the names
13. The family heard
at midnight
at a quarter to eleven
loud rustling noises outside the front porch of the house
loud rustling noises outside the new home
loud noises
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
14. Congress included
as expenditures
as some of the expenditures
a budget for the loss of farm crops and equipment
a budget for damaged sugar cane crops
a budget
15. The actress mentioned
on television
on a television talk show
the names of two of her most recent movie projects
the names of her recent movie projects
the movies
16. The doctor noted
with care
with an acutely trained eye
the tiny lump on the end of her patient's elbow
the tiny lump on the patient ’ s arm
the lump
17. The driver noticed
with horror
with quite a horrified look
a dent in the rear bumper of his new Ferrari
a large dent in his new Ferrari
a dent
18. The scouts observed
at camp
at the busy National Park
a famous array of geysers hot springs and small lakes
a wide array of rivers and geysers
a geyser
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
19. The investigator omitted
from reports
from the latest police reports
two pieces of the evidence in the tax fraud case
two pieces of the tax fraud evidence
two findings
20. The mystic predicted
with confidence
with a show of confidence
the exact locations and dates of the next two floods
the locations and dates of two floods
the floods
21. The government representative provided
on Monday
on the nineteenth of July
some of the facts about the new market in Europe
some facts about the new Common Market
some facts
22. The professor questioned
during class
during the guest lecturer's talk
the strength of the new theory and data on cloning
the strength of the recent cloning data
the data
23. The teacher discussed
in detail
in a very forceful manner
a paper on civil rights in the past four decades
a long paper on civil rights issues
a paper
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
24. The dancer realized
with disappointment
with disappointment and some shock
her grave mistake during the last part of the routine
her mistake during part of the routine
her mistake
25. The producer recognized
at work
at a busy coffee bar
the famous actress from the hit movie Air Force One
the famous actress from Air Force One
the actress
26. The parking attendant recommended
with pleasure
with pleasure and genuine warmth
a classy four-star hotel near the John Wayne airport
a four-star hotel near the airport
a hotel
27. The mayor reported
with regret
with some regret and sorrow
the hard facts about seven bank thefts and five murders
the recent bank thefts and strange murders
the crimes
28. The restaurant owner revealed
with pride
with a show of enthusiasm
the cooking talents of the chef and his kitchen staff
the light and spicy grilled fish recipes
the recipes
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
29. The administrators reviewed
with parents
with many of the parents
the guidelines for the first through third grade reading
program
the guidelines for the first grade program
the guidelines
30. The report stated
with authority
with authority and some facts
the absence of any drugs or weapons in the home
the findings for a recent drug study
the findings
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
IMAGE EVALUATION
150mm
IM/IGE. Inc
1653 East Main Str eet
Rochester, NY 14609 USA
Phone: 716/482-0300
Fax: 716/288-5989
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Linked assets
University of Southern California Dissertations and Theses
Conceptually similar
PDF
Definiteness types in Spanish: A study of natural discourse
PDF
Frequency in sentence comprehension
PDF
In the event of focus
PDF
Charlotte Smith: Life of a novelist, novels of a life
PDF
Morphology and the lexicon: Exploring the semantics-phonology interface
PDF
Justus Georg Schottel and linguistic theory
PDF
On the representation and licensing of Q and Q-dependents
PDF
Effects of working memory and semantic impairments on speech in Alzheimer's disease.
PDF
Intra-lexical noun-verb dissociations: Evidence from Chinese aphasia
PDF
Language and context: toward a model of psycholinguistic competence
PDF
The dramatic technique of Cyril Tourneur
PDF
The syntax of clitic doubling in modern Greek
PDF
'Beowulf' and 'the hobbit': elegy into fantasy in j. R. R. Tolkien's creative technique
PDF
Arthur Machen’s treatment of the occult and a consideration of its reception in England and America
PDF
Satirical social criticism in the novels of John Dos Passos
PDF
Greek drama and the African world: A study of three African dramas in the light of Greek antecedents
PDF
On the status of possessives
PDF
The other world in the New World: Representations of the supernatural in short fiction of the Americas
PDF
The syntax of the Chinese BA-constructions and verb compounds: A morphosyntactic analysis
PDF
The syntax of focus and topic in Mandarin Chinese
Asset Metadata
Creator
Stallings, Lynne Marie
(author)
Core Title
Evaluating heaviness: Relative weight in the spoken production of heavy-NP shift
Degree
Doctor of Philosophy
Degree Program
Linguistics
Publisher
University of Southern California
(original),
University of Southern California. Libraries
(digital)
Tag
language, linguistics,OAI-PMH Harvest,psychology, cognitive,psychology, general
Language
English
Contributor
Digitized by ProQuest
(provenance)
Permanent Link (DOI)
https://doi.org/10.25549/usctheses-c17-396627
Unique identifier
UC11350796
Identifier
9919108.pdf (filename),usctheses-c17-396627 (legacy record id)
Legacy Identifier
9919108.pdf
Dmrecord
396627
Document Type
Dissertation
Rights
Stallings, Lynne Marie
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 au...
Repository Name
University of Southern California Digital Library
Repository Location
USC Digital Library, University of Southern California, University Park Campus, Los Angeles, California 90089, USA
Tags
language, linguistics
psychology, cognitive
psychology, general