University of Southern California Dissertations and Theses

Dynamics of multiple pronoun resolution

(USC Thesis Other)

Dynamics of multiple pronoun resolution

PDF

Download

Open document

Flip pages

Download a page range

Download transcript

Copy asset link

Request this asset

Transcript (if available)

Content DYNAMICS OF MULTIPLE PRONOUN RESOLUTION by Jina Song A Dissertation Presented to the FACULTY OF THE USC GRADUATE SCHOOL UNIVERSITY OF SOUTHERN CALIFORNIA In Partial Fulfillment of the Requirements for the Degree DOCTOR OF PHILOSOPHY (LINGUISTICS) May 2023 Copyright 2023 Jina Song ii Acknowledgments I would like to express my deepest gratitude to my advisor, Elsi Kaiser. I feel incredibly grateful and privileged to have been guided by this exceptionally brilliant, supportive, and warm-hearted person. She always listened to all of my ideas, no matter how small, and helped guide me towards a better path. Her unwavering encouragement, profound expertise, and insight have been invaluable in shaping my research and helping me grow as an independent researcher. Especially, as a mother and a graduate student, I was able to manage both roles thanks to her constant support and understanding. I can honestly say that I would never have made it through this journey without knowing that she would always be there for me. Thank you from the bottom of my heart for everything. I would like to thank my committee members Andrew Simpson and Toby Mintz for their valuable feedback and contributions to this work. Their insights and perspectives from syntax and cognitive processing have helped me to refine my ideas and make my argument more sound. I am also grateful to Roumyana (Roumi) Pancheva, Deniz Rudin, Audrey Li, and Stefan Keine for serving on my qualifying and screening committees and providing feedback that helped me to shape the direction of my research. Their contributions have been instrumental in bringing this dissertation to fruition. I am also deeply grateful to my mentors outside the Linguistics department at USC, including Eun-Jung Yoo, Chang-Yong Sohn, Ki-Sun Hong, Yong-Yae Park, Heejeong Ko, and Hong-Oak Yun, who first led me to the path of linguistics. Especially, I am thankful to my MA advisor, Eun-Jung Yoo, for patiently guiding me through every step of my linguistic journey and for her unwavering support throughout this process, even when I doubted myself. I am also grateful iii to Prof. Hong-oak Yun, who introduced me to the joys of psycholinguistics. Additionally, I want to express my appreciation to Woojoo Kim, a lecture of Korean language program at USC, who showed me how to teach and interact with students in a thoughtful and effective way. I extend my gratitude to my fellow graduate students and staff at USC, including my cohort Silvia Kim, Yijing Lu, and Tommy Tsz Ming Lee. Their company, support, and shared laughter over meals helped me to get through my doctoral program. I am also thankful to the members of the Language Processing Lab, including Sarah Lee, Binh Ngo, Ana Besserman, Jesse Storbeck, Silvia Kim, Cindy Chiang, Ian Rigby, and Jun Lyu, for sharing their thoughts and insights about my research. Especially, I am grateful to Silvia Kim and Sarah Lee for being there to listen and support me through all sorts of issues in my life. Special thanks to Jesse Storbeck for helping me design and implement my webcam-based eye-tracking study, and patiently answering my endless questions. Without your help, I would never have been able to finish Chapter 5. Additionally, I would like to thank our departmental staff Guillermo Ruiz and Lisa Jo Keefer (unfortunately, she is no longer at USC) for always being kind and helpful in handling complex administrative tasks. Finally, I would like to express my sincere thanks to my family. My husband, Joe, deserves a great deal of gratitude for his constant support and encouragement. Without him, I wouldn't have had the courage to start my doctoral program overseas with our child. He has been my lifelong companion, guiding me towards the right path and providing unwavering support throughout this journey. Words cannot express how grateful I am for everything he has done for me. My Son, Noah, has been the driving force behind my perseverance. His presence alone brings me so much happiness and strength, and I couldn't have made it this far without him. Thank you so much for everything, and I love you so much! I am also grateful to my cousin, Jin-Hwan, for selflessly helping us with everything without complaining while living with us. We have gone through iv difficult times together, such as the COVID pandemic, and it feels like we have become one big family. I also would like to thank my in-laws, Il Huh, Do-Hyun Chang, Hwan Huh and Eun-Ji Huh, who supported and prayed for me throughout the doctoral program. Especially, I am grateful to Eun-Ji for always lifting my spirits with enjoyable conversations! My sister-like friend, Ha-Young, has been a constant source of support and care, always being there for me through both happy and sad times, even though we were miles apart. Lastly, I want to express my deep gratitude to my mother, Youn-Sook An, who always prayed for me and traveled all the way to the United States where she didn't know anyone just to help me. I can never thank you enough for all that you've done for me. I love you! This dissertation is dedicated to the memory of my father, Ki-Sup Song, who now rests in peace with God. Parts of the research presented in Chapter 2 has been submitted for publication in Language, Cognition, and Neuroscience by Song and Kaiser (to appear). v Table of Contents Acknowledgements ......................................................................................................................... ii List of Tables ................................................................................................................................. xi List of Figures .............................................................................................................................. xiii Abstract ........................................................................................................................................ xvi Chapter 1: Introduction ................................................................................................................... 1 1.1. Overview and objectives ................................................................................................... 1 1.2. Defining referential structure ............................................................................................ 3 1.3. Factors that guide pronouns resolution at different levels ................................................. 4 1.3.1. Syntactic factor: Parallel Function Strategy (PFS) .................................................. 5 1.3.2. Syntax-memory factor: Introducing the Cue-based parallel function strategy (Cue-based PFS) .................................................................................................... 12 1.3.2.1. Memory-retrieval factor: Cue-based Retrieval Approach ............................ 13 1.3.2.2. Cue-based Parallel Function Strategy (Cue-based PFS) .............................. 15 1.3.3. Discourse coherence factor: Centering Theory (CT) ............................................. 20 1.4. Structure of this dissertation ............................................................................................ 28 Chapter 2: The effects of referential structure in pronoun interpretation ..................................... 30 2.1. Introduction ..................................................................................................................... 30 2.2. Hypotheses on referential structure ................................................................................. 32 2.2.1 Independence View ................................................................................................. 32 2.2.2. Dependence Views ................................................................................................. 36 2.2.2.1. Syntax-level Dependence View .................................................................... 36 2.2.2.2. Discourse-level Dependence View ............................................................... 38 2.2.2.3. Syntax-Discourse Dependence View ........................................................... 41 2.3. Psycholinguistic work on effects of referential structure ................................................ 42 2.4. Effects of Implicit Causality ............................................................................................ 44 2.5. Experiment 1 ................................................................................................................... 45 2.5.1. Participants ............................................................................................................. 46 2.5.2. Materials and design .............................................................................................. 46 2.5.3. Procedure ............................................................................................................... 48 2.5.4. Data processing and analysis ................................................................................. 50 2.5.5. Predictions .............................................................................................................. 51 2.5.6. Results .................................................................................................................... 54 2.5.7. Discussion .............................................................................................................. 56 viii 2.6. Experiment 2 ................................................................................................................... 59 2.6.1. Participants ............................................................................................................. 60 2.6.2. Materials and design .............................................................................................. 60 2.6.3. Procedure ............................................................................................................... 61 2.6.4. Data processing and analysis ................................................................................. 61 2.6.5. Predictions .............................................................................................................. 62 2.6.6. Results .................................................................................................................... 62 2.6.7. Discussion .............................................................................................................. 63 2.7. Experiment 3 ................................................................................................................... 65 2.7.1. Participants ............................................................................................................. 65 2.7.2. Materials and design .............................................................................................. 65 2.7.3. Procedure ............................................................................................................... 66 2.7.4. Data processing and analysis ................................................................................. 67 2.7.5. Predictions .............................................................................................................. 67 2.7.6. Results .................................................................................................................... 68 2.7.7. Discussion .............................................................................................................. 69 2.7.8. Comparing the three experiments .......................................................................... 70 2.8. General discussion ........................................................................................................... 71 2.8.1. Referential structure effects ................................................................................... 72 2.8.2. Implicit Causality effects ....................................................................................... 75 2.8.3. Object shift in IC bias and subjectivity .................................................................. 75 2.8.4. Thematic role effects .............................................................................................. 78 2.9. Conclusion ....................................................................................................................... 80 Chapter 3: Source of the dependency interactions in multiple pronoun resolution ...................... 82 3.1. Introduction ..................................................................................................................... 82 3.2. Experiment 4 ................................................................................................................... 84 3.2.1. Participants ............................................................................................................. 84 3.2.2. Materials and design .............................................................................................. 84 3.2.3. Procedure ............................................................................................................... 86 3.2.4. Data processing and analysis ................................................................................. 87 3.2.5. Predictions .............................................................................................................. 87 3.2.6. Results .................................................................................................................... 91 3.3. Discussion........................................................................................................................ 93 3.3.1. Source of referential structure effects .................................................................... 94 3.3.2. Syntax-level parallelism over discourse-level coherence ...................................... 95 3.3.3. Topicality effects.................................................................................................... 97 3.4. Conclusion ....................................................................................................................... 98 Chapter 4. The effects of referential structure with null pronouns ............................................. 100 4.1. Introduction ................................................................................................................... 100 4.2. Relationship between anaphoric forms and topicality................................................... 101 4.3. Korean null pronouns .................................................................................................... 107 4.4. Experiment 5 ................................................................................................................. 109 ix 4.4.1. Participants ........................................................................................................... 110 4.4.2. Materials and design ............................................................................................ 110 4.4.3. Procedure ............................................................................................................. 113 4.4.4. Data processing and analysis ............................................................................... 113 4.4.5. Predictions ............................................................................................................ 114 4.4.5.1. Syntax-over-Discourse Dependence View ................................................. 115 4.4.5.2. Discourse-over-Syntax Dependence View ................................................. 118 4.4.5.3. Syntax-level / Discourse-level Dependence View ..................................... 119 4.4.6. Results .................................................................................................................. 120 4.5. Discussion...................................................................................................................... 122 4.5.1. Stronger syntactic parallelism effects in null pronoun resolution ........................ 123 4.5.2. Stronger topicality effects: null pronouns vs. overt pronouns ............................. 124 4.6. Conclusion ..................................................................................................................... 128 Chapter 5. Referential structure effects during real-time processing ......................................... 129 5.1. Introduction ................................................................................................................... 129 5.2. Recency effects in pronoun interpretation ..................................................................... 133 5.3. Experiment 6 ................................................................................................................. 137 5.3.1. Participants ........................................................................................................... 138 5.3.2. Materials and design ............................................................................................ 139 5.3.3. Auditory (sentence) stimuli .................................................................................. 141 5.3.4. Visual stimuli ....................................................................................................... 142 5.3.5. Procedure ............................................................................................................. 145 5.3.6. Predictions ............................................................................................................ 148 5.3.6.1. Predictions about last-mentioned selections ............................................... 148 5.3.6.2. Predictions about eye-gaze patterns during pronoun interpretation ........... 152 5.4. Offline mouse-click results............................................................................................ 155 5.4.1. Data processing and analysis ............................................................................... 155 5.4.2. Last-mentioned referent selections ...................................................................... 157 5.5. Online eye-tracking results ............................................................................................ 160 5.5.1. Eye-gaze patterns during the matrix clause ......................................................... 160 5.5.2. Data processing and analysis for the because-clause........................................... 161 5.5.3. Eye-gaze patterns during the subject in the because-clause ................................ 163 5.5.4. Eye-gaze patterns during the indirect object in the because-clause ..................... 165 5.6. Discussion...................................................................................................................... 170 5.6.1. Immediate activation of preceding pronoun’s referential dependencies ............. 171 5.6.2. Recency effects on pronoun resolution ................................................................ 175 Chapter 6: Conclusion................................................................................................................. 178 6.1. Overview and summary ................................................................................................. 178 6.2. Various patterns of effects of referential structure ........................................................ 181 6.3. Implications for models of pronoun resolution ............................................................. 183 6.4. Future directions ............................................................................................................ 184 6.5. Final conclusion............................................................................................................. 185 x References ................................................................................................................................... 187 Appendices .................................................................................................................................. 202 Appendix A: Target sentences .............................................................................................. 202 Appendix B: Statistical models ............................................................................................ 210 xi List of Tables Table 1. Center transition states (Walker et al., 1998, p.5) ......................................................... 23 Table 2. A set of example stimuli in Experiment 1 ...................................................................... 47 Table 3. Transition type predicted by CT in Experiment 1 .......................................................... 53 Table 4. Experiment 1: results of the glmer model ...................................................................... 56 Table 5. Experiment 1: planned comparisons .............................................................................. 56 Table 6. A set of example stimuli in Experiment 2 ...................................................................... 60 Table 7. Experiment 2: results of the glmer model ...................................................................... 63 Table 8. Experiment 2: planned comparisons .............................................................................. 63 Table 9. A set of example stimuli in Experiment 3 ...................................................................... 66 Table 10. Experiment 3: results of the glmer model .................................................................... 69 Table 11. A set of example stimuli in Experiment 4 .................................................................... 85 Table 12. Transition type predicted by CT in Experiment 4 ........................................................ 90 Table 13. Experiment 4: results of the glmer model .................................................................... 93 Table 14. Experiment 4: planned comparisons ............................................................................ 93 Table 15. Transition type predicted by CT in Experiment 5 ...................................................... 117 xii Table 16. Experiment 5: results of the glmer model .................................................................. 121 Table 17. Experiment 5: planned comparisons .......................................................................... 121 Table 18. Transition type predicted by CT in Experiment 6 ...................................................... 151 Table 19. Experiment 6: results of the glmer model comparing One-Object-Pronoun vs. Two-Pronoun condition .............................................................................................................. 159 Table 20. Experiment 6: results of the glmer model comparing One-Subject-Pronoun vs. Object-Pronoun condition ........................................................................................................... 159 Table 21. Experiment 6: results of the glmer model comparing One-Subject-Pronoun vs. Two-Pronoun condition .............................................................................................................. 159 Table 22. Summary of fixed effects for the models predicting the strength of the subject advantage scores at time intervals aligned to the offset of the because-clause subject .............. 165 Table 23. Summary of fixed effects for the models on the subject advantage scores in eight 500 ms time-slices and one 400 ms time-slice aligned to the onset of the indirect object (significant effects are shaded) ................................................................................................... 168 Table 24. Summary of fixed effects for the models on the subject advantage scores in four 500 ms time-slices and one 2000 ms time-slice aligned to the onset of the indirect object (significant effects are shaded) ................................................................................................... 169 xiii List of Figures Figure 1. Cue-based memory retrieval model based on the example (12) (adopted and modified from Figure 1. in Jäger et al., (2017)) ............................................................................ 14 Figure 2a. One-pronoun structure ................................................................................................ 18 Figure 2b. Two-pronoun structure ............................................................................................... 18 Figure 3a. One-pronoun structure ................................................................................................ 38 Figure 3b. Two-pronoun structure ............................................................................................... 38 Figure 4a. Example Two-Pronoun item from Experiment 1 ....................................................... 49 Figure 4b. Example One-Pronoun item from Experiment 1 ........................................................ 49 Figure 5a. One-pronoun structure ................................................................................................ 52 Figure 5b. Two-pronoun structure ............................................................................................... 52 Figure 6. Experiment 1: The proportion of trials where the subject-position pronoun refers to the preceding object. (error bars represent ±1 S.E.) ..................................................................... 55 Figure 7a. Example Two-Pronoun item from Experiment 2 ....................................................... 61 Figure 7b. Example Pronoun+Name item from Experiment 2 .................................................... 61 Figure 8. Experiment 2: The proportion of trials where the subject-position pronoun refers to the preceding object. (error bars represent ±1 S.E.) ..................................................................... 62 Figure 9a. Example Two-Pronoun item from Experiment 3 ....................................................... 67 xiv Figure 9b. Example Pronoun+Name item from Experiment 3 .................................................... 67 Figure 10. Experiment 3: The proportion of trials where the subject-position pronoun refers to the preceding object. (error bars represent ±1 S.E.) ................................................................. 68 Figure 11a. Example Subject-Topic item from Experiment 4 ..................................................... 87 Figure11b. Example Object-Topic item from Experiment 4 ....................................................... 87 Figure 12. Experiment 4: The proportion of trials where the subject-position pronoun refers to the preceding object. (error bars represent ±1 S.E.) ............................................................... 91 Figure 13. Topic-Continuity Hierarchy (Givón, 1983:17) ......................................................... 102 Figure 14a. Example Subject-Topic item from Experiment 5 ................................................... 113 Figure14b. Example Object-Topic item from Experiment 5 ..................................................... 113 Figure 15. Experiment 5: The proportion of trials where the subject-position pro refers to the preceding object. (error bars represent ±1 S.E.) ................................................................... 120 Figure 16. Experiment 4 vs. Experiment 5: The proportion of trials where the subject-position pro refers to the preceding object. (Two-Pro(noun) and Pronoun+Name/ One-Pro conditions are collapsed) (error bars represent ±1 S.E.) .............................................. 126 Figure 17a. One-pronoun structure ............................................................................................ 131 Figure 17b. Two-pronoun structure ........................................................................................... 131 Figure 18. An example screen layout for the target sentence: Tommy encouraged Max after the winter vacation because he/Hanna had wengered the neend to Hanna/him ........................ 143 Figure 19. An example screen layout for the catch trial: The spider was found by Dorothy ..... 144 xv Figure 20. An example screen layout for the homophone type: While getting a perm in the beauty salon, Joanna read the book about the tortoise and the hare ......................................... 144 Figure 21. An example screen layout for the PP-attachment type: At the stroony trindle, Isaac sang to the girl with a microphone.................................................................................... 144 Figure 22. An example screen layout the landmark type: While touring Europe, the students went on a churby trip to London and saw Big Ben ..................................................................... 144 Figure 23a. An example screen of the practice item after success ............................................ 147 Figure 23b. An example screen of the practice item after failure (1st attempt) ........................ 147 Figure 24. A visualization of the immediate activation hypothesis ........................................... 154 Figure 25. A visualization of the delayed activation hypothesis ............................................... 155 Figure 26. How often did participants interpret the last-mentioned referent as referring back to the preceding subject? (error bars represent ±1 S.E.) ........................................................... 157 Figure 27. How often did participants interpret the last-mentioned referent as referring back to the name in the because clause (error bars represent ±1 S.E.) ................................................ 158 Figure 28. Eye movements relative to the onset of the matrix-clause subject (each facet shows looks to the three regions of interest within a condition)................................................. 160 Figure 29. Eye movements relative to the offset of the because-clause subject (each facet shows looks to the four regions of interest within a condition) .................................................. 163 Figure 30. Eye movements relative to the onset of the because-clause indirect object (each facet shows looks to the three regions of interest) ...................................................................... 166 Figure 31. Eye movements relative to the onset of the pronoun in the because-clause indirect 166 xvi Abstract This dissertation explores the mechanisms underlying multiple pronoun resolution in comparison to single pronoun resolution, focusing on differences in referential structure – whether all or only one of the referents in the preceding clause are mentioned with a pronoun. This dissertation aims to answer four questions: (i) whether and how the process of multiple pronoun resolution is different from that of single pronoun resolution (ii) what factors contribute to effects of referential structure (iii) whether and how referential structure effects are influenced by differences in anaphoric forms, and (iv) how multiple pronouns are interpreted during real-time processing. Experiments 1, 2 and 3 investigate the first question and test two hypotheses about the potential effects of referential structure in pronoun resolution. The independence view assumes that each pronoun’s anaphoric dependencies are resolved fully independently while the dependence views assume that there are interactions between multiple referential dependencies. Accordingly, pronoun interpretation is predicted to be sensitive to differences in referential structure type only in the dependence views, but not in the independence view. The results from three offline picture-writing tasks support the dependence view, showing that pronoun interpretation is indeed sensitive to referential structure type,. Experiment 4 addresses the second question and further examines what factors at different levels of representation contribute to effects of referential structure by comparing three types of the dependence view (syntax-level, discourse-level, and syntax-discourse dependence view), based on the influence of syntactic and/or discourse factors. The results from an offline picture- writing task support the syntax-discourse dependence view, suggesting that both syntax (i.e., cue- based parallelism) and discourse factors (i.e., Centering Theory) contribute to the interactions xvii between anaphoric dependencies of the two different pronouns, resulting in referential structure effects. Experiment 5 addresses the third question and broadens the scope of referential structure effects to another type of anaphoric form, null pronouns (pro) in Korean. This study tests whether the discourse-level coherence factor has a greater impact on referential structure effects than the syntax-level factor with Korean pro, which have a reduced anaphoric form compared to overt pronouns in English. The results from an offline picture-writing task show that the discourse factor indeed exerts a greater influence on pro resolution in Korean than overt pronoun resolution in English. Experiment 6 investigates the last question by examining referential structure effects during real-time processing using a novel webcam-based visual-world eye-tracking method. Two competing hypotheses are tested regarding the timing of referential structure effects: the immediate activation hypothesis assumes that when the subsequent pronoun is encountered, the preceding pronoun’s referential dependencies are retrieved immediately while the delayed activation hypothesis assumes that the retrieval is delayed. The results indicate early emergence of referential structure effects during resolution of the subsequent pronoun (i.e., the immediate activation hypothesis), suggesting that the preceding pronoun’s dependency information becomes available as soon as a comprehender encounters the subsequent pronoun. In conclusion, this dissertation provides novel evidence that multiple pronoun resolution is different from single pronoun resolution and that these differences stem from differences in referential structure – resolving referential dependency of one pronoun can influence on resolving dependency of the other pronoun. Thus, the findings of this dissertation highlight the need to consider referential structure in understanding pronoun interpretation. 1 Chapter 1: Introduction 1.1. Overview and objectives In our natural language, we have expressions that are interpreted depending on another expression in context. One type of these expressions is pronouns (e.g., she, he, her, him), and third-person pronouns are informationally underspecified on their own. We can fully understand their meanings only when we identify what they refer to in the previous context. However, it is not always straight- forward to establish a dependency relation with a previously mentioned entity because pronouns typically provide very little semantic information about their referents, as in (1). (1) Janei respected Maryj because shei<j visited Tom yesterday. In example (1), the two preceding nouns Jane and Mary can be potential antecedents for the pronoun she due to being third-person female nouns. Despite the ambiguity, most native speakers of English prefer to interpret the pronoun she as referring to Mary. A number of previous psycholinguistic studies have investigated what factors guide the referential preferences in pronoun resolution from a variety of perspectives. Various factors, including syntactic role, discourse status and thematic roles, have been argued to influence the salience of referents, and, therefore, to influence the likelihood of a particular referent being interpreted as the antecedent of a subsequent pronoun. For example, prior work shows that – other things being equal – a third-person pronoun tends to be interpreted as referring to (i) the preceding entities realized in parallel grammatical roles (e.g., Chambers & Smyth 1998; Sheldon, 1974; Smyth, 1994; Stevenson et al., 1993, 1995), to (ii) the implicit causer entity in causal contexts with 2 implicit causality (IC) verbs (e.g., Au, 1986; Caramazza et al., 1977; Ferstl et al., 2011; Hartshorne & Snedeker, 2013; Hartshorne et al., 2015; McKoon, Greene, & Ratcliff, 1993; Rudolph & Försterling, 1997), and to (iii) a topical entity rather than a non-topic (e.g., Ariel, 1990; Arnold, 1998; Colonna et al., 2010; Cowles et al., 2007; Ellert, 2013; Givón, 1983; Grosz et al., 1995; Kaiser, 2011). However, it is important to highlight that most existing accounts tend to focus on clauses that only contain a single pronoun as in (1) and do not extend straightforwardly to explain how multiple pronouns 1 in the same clause are interpreted – with the exception of work on parallelism effects (e.g., Chambers & Smyth 1998; Sheldon, 1974; Smyth, 1994; Stevenson et al., 1993, 1995). (2) Janei respected Maryj because shei/j visited heri/j yesterday. Clauses that contain more than one pronoun as in the example (2) are frequent in naturalistic language use. Nevertheless, prior psycholinguistic work tended to focus on one- pronoun contexts, and when work has tested two-pronoun contexts, it has typically been in contexts involving semantic parallelism (e.g., Smyth, 1994; Stevenson et al., 1993, 1995). Thus, how we interpret multiple pronouns in a single clause is still poorly understood. For example, whether the presence of another pronoun later in the same clause influences people’s ultimate interpretation of subject-position pronouns– comparing (1) and (2), for example – is not yet known. The subject pronoun she in the two-pronoun example in (2) might or might not show the same preference for the preceding referent Mary as she in the one-pronoun example in (1). 1 Although we acknowledge that multiple pronoun resolution is investigated in prior work like Chambers & Smyth (1998), these prior studies typically investigate non-ambiguous configurations, where each pronoun is resolved in favor of a specific antecedent. (e.g., Debbie punched David in the nose. Then she slugged him in the ribs. (Chambers & Smyth, 1998, p.603). 3 This dissertation aims to shed light on the mechanisms that underlie pronoun resolution, by investigating how and what factors guide the interpretation of multiple pronouns in the same clause. Ultimately, a satisfactory model of pronoun interpretation should capture all relevant factors. Thus, if our aim is to account for the mechanisms that explain how comprehenders ultimately interpret pronouns, it is necessary to gain a better understanding of how the process of reference resolution of multiple pronouns differs from that of interpreting a single pronoun. 1.2. Defining referential structure This dissertation aims to investigate the underlying mechanism that guides multiple pronoun resolution. To do this, I compare pronoun interpretation in configuration where only one pronoun is present in a clause as in the example (3a) and configuration with two pronouns in a clause as in the example (3b). In principle, pronouns are ambiguous in both configurations since the two antecedents are identical in gender. (3) a. Janei respected Maryj because shei<j visited Tom yesterday. [One-Pronoun] b. Janei respected Maryj because shei/j visited herj/i yesterday. [Two-Pronoun] From the perspective of reference resolution, the crucial difference between these two configurations lies in the interpretational consequences of a specific type of information in the pronoun-containing clause, namely its referential structure. The referential structure of these configurations differs in that only one of the antecedents from the preceding clause (e.g., Jane or Mary) is mentioned in the One-Pronoun structure in (3a), while both antecedents (e.g., Jane and Mary) are mentioned in the Two-Pronoun structure in (3b). Since the processing system needs to 4 find an antecedent for each of these pronouns – one pronoun (e.g., she) vs. two pronouns (e.g., she and her) – these two configurations differ in the number of referential expressions that need to establish a dependency relation with an antecedent. In this dissertation, I use the term referential structure for this difference, i.e., whether all or only one of the referents in the preceding clause are mentioned with a pronoun in the subsequent clause. In the rest of this chapter, I explore the idea that referential structure influences pronoun interpretation due to the different number of referential dependencies that the language processing system activates in One- vs. Two-Pronoun configurations. Since each of these pronouns can be interpreted as referring to either of the two antecedents, my research examines whether resolving the referential dependency of one pronoun can influence the resolving the dependency of the other pronoun. 1.3. Factors that guide pronouns resolution at different levels For more than three decades, psycholinguistic work on reference resolution has proposed different models of reference resolution. These models show that pronoun interpretation is guided by various factors at different levels of representation, such as those at the level of syntax (e.g., parallel syntactic functions, and subjecthood), those at the level of discourse (e.g., information structure, and coherence relation between propositions) and those at the level of semantics (e.g., thematic-role parallelism and Implicit Causality (IC) verb bias). This dissertation focuses primarily on the models that were built on syntactic parallelism (i.e., Parallel Function Strategy) and local discourse coherence (i.e., Centering Theory). Since these two models can extend to multiple pronoun resolution as well as single pronoun resolution, I will adopt them to test effects of referential structure – whether or not patterns of pronoun resolution depend on referential structure 5 of the pronoun-containing clause. 1.3.1. Syntactic factor: Parallel Function Strategy (PFS) The Parallel Function Strategy (PFS) (e.g., Chambers & Smyth 1998; Smyth, 1994; Sheldon, 1974; Stevenson et al., 1993, 1995) is one of the approaches that claims that syntactic function plays a key role in pronoun resolution. Crucially, the PFS suggests an independent strategy for each of the pronouns that appears in different syntactic positions (e.g., object-position and subject- position pronouns) because it claims that pronoun interpretation is guided by grammatical roles such that the pronoun is co-referenced with the preceding referent that shares the same grammatical role with, as defined in (4). (4) Parallel Function Strategy (PFS): A pronoun is coreferential with a preceding noun phrase occupying the same grammatical role as the pronoun. (Stevenson et al., 1995, p. 394) In the example (5), the subject pronoun he prefers the subject antecedent Tom, but the object pronoun him prefers the object antecedent Jason. This pattern is accounted for by the PFS based on the parallel grammatical roles shared between the pronouns and their antecedent. (5) a. Tomi hit Jasonj, and then hei kicked Bill. b. Tomi hit Jasonj, and then Bill kicked himj. However, it is important to note that most of prior studies about the PFS tested pronoun 6 interpretation in a context that also involves semantically similar or parallel relations between two clauses. In other words, the two clauses describe events or situations that are in some way semantically similar/parallel (e.g., hitting and kicking in (5)). This means that we cannot tell whether it is the parallel grammatical roles or the parallel semantic relations that are driving the patterns observed in these prior studies (for discussion about the importance of coherence relations, see e.g. Hobbs 1979, 1990; Kehler 2002; Kehler et al., 2008; Wolf et al., 2004). I will revisit this issue at the end of this section. Regarding the mechanism underlying the PFS 2 , Smyth (1994) claims that an underlying mechanism for the PFS is a search process based on a feature-matching system. During the search process, “all NPs in the search space are canvassed for both the morphological and syntactic properties which license coreference with the pronoun” (Smyth 1994:197), and then the anaphoric processor decides the best antecedent based on the number of features, such as a gender, number, person and grammatical role, shared by the pronoun. Since the grammatical role is one of the features that is involved in the feature matching process, the parallel grammatical role guides pronoun resolution if the other features are identical among the potential antecedents. To test the PFS, Stevenson et al (1995) conducted a clause-by-clause self-paced reading task by manipulating grammatical roles between the pronouns and their antecedent: for the non- subject pronoun, the non-subject position antecedent was either realized in a parallel grammatical role as in (6) or in non-parallel grammatical role as in (7). During the task, reading times for the last clause of each target sentence were recorded along with participants’ answers to the final questions (“true” or “false). In the results, the differences between the subject (e.g., she) and non- 2 Stevenson et al. (1995) simply views the PFS as one of the heuristic strategies for pronoun resolution that uses a syntactic cue (grammatical roles). 7 subject pronouns (e.g., her) regarding reading times and forced choices were bigger in what the authors call the “fully parallel sentences” condition compared to the “partially parallel sentences” condition. It is primarily driven by the non-subject pronouns’ preference to the non-subject position antecedent (e.g., Penelope or Priscilla) being reduced when their grammatical role was not fully parallel. Thus, the findings were compatible with the predictions of the PFS. (6) Fully parallel sentences: a. CONTEXT: Jean and Penelope were working late trying to finish a job. Roger was also working late but at nine o’clock had had enough. b. TARGET: Jean asked Penelope to make an effort and {she/Roger} told {Roger/her} to have a break. c. QUESTION: {Jean/Roger} told {Roger/Jean} to take a break? (7) Partially parallel sentences: a. CONTEXT: Mary and Priscilla were about to go into town when they realised the car had a puncture. Graham was their next door neighbor and he knew nothing about cars. b. TARGET: Mary helped Priscilla change the wheel and {she/ Graham} talked to {Graham/her} without interest. c. QUESTION: {Mary/Graham} talked to {Graham/Mary} without interest? (Stevenson et al 1995, p.401) However, the main goal of this dissertation is not limited to single pronoun interpretation and extends to comparing pronoun interpretation in One-Pronoun configurations as in (8a) to Two- 8 Pronoun configurations as in (8b). (8) a. Janei respected Maryj because shei>j daxed Lisa. {3sing, fem, sbj} {3sing, fem, obj} {3sing, fem, sbj} b. Janei respected Maryj because shei>j daxed heri<j. {3sing, fem, sbj} {3sing, fem, obj} {3sing, fem, sbj} {3sing, fem, obj} At least when the PFS is taken at face value, it does seem to not make any explicit claims about subject pronoun interpretation being different by referential structure type (One-Pronoun vs. Two-Pronoun sentences). As illustrated in example (8), the best antecedent for the subject-position pronoun she is the preceding subject Jane regardless of the presence of the object-position pronoun her because the syntactically parallel antecedent shares the greatest number of features with the pronouns. In other words, under the current configuration where two potential antecedents only differ in their grammatical roles, each pronoun in the Two-Pronoun sentences is predicted to show its own complementary preferences to their parallel antecedent, not showing differences from reference resolution of the subject position pronoun in the One-Pronoun sentence. On the contrary, in some cases, subject-position pronoun interpretation might be different depending on the presence or absence of another pronoun given that the PFS posits that it can operate together with other constraints such as a structural priming (Bock, 1986) or a subject assignment strategy (Crawley et al., 1990). Specifically, the PFS hypothesizes that parallel interpretation is facilitated in the clauses that are structurally parallel due to the structural priming effect and that the “subject assignment is the default when the two clauses are nonparallel” (Smyth 1994, p.221). In the structurally non-parallel clauses where the subject assignment strategy comes 9 into play as in (9), a subject pronoun in the One-Pronoun configuration in (9a) could have a stronger subject preference than a subject pronoun in the Two-Pronoun configuration in (9b) because there could be competition for the subject antecedent Tom between the subject and object pronoun in the Two-Pronoun configuration, but there could be no competition for the subject antecedent Tom in the One-Pronoun configuration. (9) a. Tomi passed a ball to Jasonj, and then hei>j saw Bill. [One-Pronoun structure] b. Tomi passed a ball to Jasonj, and then hei>j saw himi>j. [Two-Pronoun structure] However, as long as a pronoun(s) appears in the structurally parallel context as in (8), the PFS does not straight-forwardly predict that interpretation of the subject pronoun would be different depending on the presence vs. absence of another pronoun because parallel grammatical role guides pronoun resolution through the feature-matching process. It is worthy of note that most of the studies about the PFS tested pronoun interpretation under the context where two sentences are linked with a conjunction and (similarly) and presumably interpreted as describing two semantically similar or parallel events. As Smyth (1994) acknowledged that parallel interpretation could be influenced by a pragmatic bias from conjunction type in some cases, more recent research has suggested that semantic relations inferred to hold between propositions (henceforth, coherence relations) play a significant role in pronoun resolution (e.g., Hobbs 1979, 1990; Kehler 2002; Kehler et al., 2008; Wolf et al., 2004). Experimental work by Wolf et al. (2004) and Kehler et al. (2008) provides evidence that the PFS is strongly correlated with parallel coherence relation. For example, Wolf et al. (2004) conducted a self-paced reading time study by manipulating (i) the gender of the pronoun and (ii) the coherence 10 relations between sentences as given in (10) – the parallel relation was signaled by a conjunction and similarly and semantically biased towards a parallel antecedent by using semantically similar verbs; the result relation was signaled by and so and semantically biased towards a non-parallel antecedent by using semantically different verbs. Wolf et al. found faster reading times with the parallel antecedents only in the parallel relation, but the reading times were slower with the parallel antecedents in the result relation. These findings show that coherence relations can reverse PFS effects. (10) a. Fiona complimented Craig and similarly James congratulated {her/him} after the match, but nobody took any notice. [Parallel relation] b. Fiona defeated Craig and so James congratulated {her/him} after the match, but nobody took any notice. [Result relation] (Wolf et al. 2004, p.669) Thus, the studies imply that the semantic coherence relations should be taken into account for a successful analysis of pronoun interpretation. Bearing this in mind, the coherence relation will be controlled in the experiments of this dissertation not to be biased to a specific antecedent in all conditions, by being tested under explanation relation (signaled by because) along with different type of Implicit Causality verbs in the antecedent-containing clause. This manipulation will allow either the subject or the object antecedent to be preferred for pronoun interpretation (see Chapter 2 for more discussion). In addition, it’s important keep in mind that the effects of parallelism can be ‘undone’ by coherence relations, as shown by Wolf et al. (2004) in the conditions where two clauses are connected by a cause-effect (result) relation with verbs that are semantically biased towards a non- 11 parallel antecedent. As a whole, prior work suggests that the semantic coherence relation between sentences influences how strongly subject pronouns prefer subject antecedents and object pronouns prefer object antecedents. Furthermore, pronoun resolution exhibits the strongest parallelism effects when the two clauses are linked by a semantic parallelism relation (e.g., as indicated by ‘and’ or ‘and similarly’) and have clearly similar verbs. Effects of parallelism on pronoun resolution can also emerge with other connectives, especially if the two events/situations can be construed as being in some way similar. For example, Chambers & Smyth (1998) found parallelism effects with contiguity coherence relations 3 signaled by a connective then when the two clauses describe events that can be construed as being somehow similar (e.g., ’Leonard handed Michael a sandwich. Then Carla passed him an apple.’ from Chambers & Smyth 1998, p.597). Parallelism effects on pronoun interpretation can also occur with two clauses are connected by and 4 whose verbs do not seem to be closely parallel, especially if the syntactic structure of the two clauses is parallel. For example, consider ‘Brenda copied Harriet with considerable zeal and Bill watched her with great interest’ from Stevenson et al. (1995, p.414). Here, both clauses contain an adverbial prepositional phrase which easily triggers an interpretation that the events are being compared with (‘paralleled to’) each other. In what follows, I consider – for sake of completeness – the possibility of purely syntactic parallelism effects arising even in contexts where the clauses are structurally parallel but semantically non-parallel. More specifically, I disentangle effects of syntactic parallelism from 3 According to Hendriks (2004), two clauses conjoined by ‘and then’ signals a Contiguity relation, which is defined as follows: “Contiguity relation (Occasion): Infer a change of state for a system of entities from the assertion of S2, establishing the initial state for this system from the final state of the assertion of S1.” (Kehler et al., 2008, p.6) 4 Hendriks (2004, p.6) claimed that “if two clauses are conjoined by and, they can express various coherence relations. If the sentence can be paraphrased with and similarly or and ... too, the relation is Resemblance relation. If the sentence can be paraphrased with and therefore or and as a result, this signals a Cause-Effect relation.” 12 semantic parallelism by using two clauses connected by an explanation relation (indicated by the connective because) in contexts that have nonsense verbs (e.g., daxed, yolled) in the pronoun- containing clause. This is because the explanation relation does not guide pronoun interpretation without verbal semantic information and world knowledge (see Kehler 2002; Kehler et al 2008 for more discussion). 1.3.2. Syntax-memory factor: Introducing the Cue-based Parallel Function Strategy (Cue- based PFS) Pronoun resolution relies on not only linguistics factors (e.g., morpho-syntactic features) but also on the general non-language-specific memory processing system. Since the two elements involved in a referential dependency (e.g., a pronoun and its antecedent) do not usually appear adjacent to each other, the memory processing system is necessary to build such dependencies. (11) Henry went to the grocery store with Lisa and the prices shocked him. For example, to interpret the pronoun him as referring to Henry in the preceding clause in example (11), the antecedent must be stored in our memory until it is retrieved by the following pronoun to establish a dependency with it (e.g., Lewis et al., 2006; Jäger et al., 2017; Wagers & Phillips, 2014). This means that it is not enough for us to just know the relevant morpho-syntactic information for co-reference: we also need sufficient working memory capacity as well as a memory-access/retrieval system to build such dependencies. In this section, I introduce one of the memory-retrieval models, a Cue-based Retrieval Approach (e.g., Bhatia & Dillon, 2022; Clark & Gronlund, 1996; Hammerly et al., 2019; Jäger et 13 al., 2017; Lewis & Vasishth, 2005; McElree, Foraker, & Dyer, 2003; Van Dyke & Lewis, 2003; Yadav et al., 2022), which accounts for the processes underlying dependency formation by linking linguistic factors and memory-based factors. After examining pronoun resolution from the perspective of memory retrieval, I will propose a new take on the PFS (Parallel Function Strategy) inspired by the Cue-based Retrieval Model. The new PFS account will explore the possibility that multiple referential dependencies initiated from two different pronouns interact with each other, and hence that ultimate pronoun interpretation is affected by the interaction. 1.3.2.1. Memory-retrieval factor: Cue-based Retrieval Approach A broad range of linguistic phenomena involving dependency formation have been explored from the perspective of cue-based retrieval (e.g., wh-dependencies, subject-verb agreement, see also Bhatia & Dillon, 2022; Dillon, 2011; Jäger et al., 2015 on reflexive pronouns etc.). Basically, the Cue-based Retrieval Approach (e.g., Bhatia & Dillon, 2022; Dillon, 2011; Hammerly, et al., 2019; Lewis & Vasishth, 2005; McElree, 2000; Van Dyke & Johns, 2012; Van Dyke & McElree, 2011; Yadav, et al., 2022) assumes that individual items are encoded and stored in memory as feature bundles. When the parser encounters a retrieval trigger (e.g., pronouns, reflexives, verbs), a search process begins based on the retrieval cues (features) of the trigger. Each retrieval cue is matched against each feature of items in memory simultaneously and transfers activation to those items that match them in features. Thus, items with higher activation levels – resulting from a higher number of matching features – are more likely to be retrieved. (12) The surgeon who treated Jonathan had pricked himself. (Jäger et al., 2017, p.317) 14 For example, in (12), the reflexive himself triggers an antecedent search process with the retrieval cues [c-command] and [masculine]. There are two feature-matching items in memory, the surgeon and Jonathan, and the features of the surgeon fully match with the retrieval cues but those of Jonathan match only partially. As illustrated in Figure 1, the fully-matched item the surgeon receives comparatively more activation than the partially matched one Jonathan and thus the surgeon is eventually selected and retrieved as the antecedent of the reflexive himself. Figure 1. Cue-based memory retrieval model based on the example (12) (adopted and modified from Figure 1. in Jäger et al., (2017)) Since the Cue-based Approach claims that multiple items are accessed simultaneously and evaluated in parallel, it predicts that there may be competition and interference between the multiple candidates for retrieval. This prediction of competition/inference has been confirmed in numerous studies, such as pronoun/reflexive dependencies (e.g., Badecker & Straub, 2002; Parker and Phillips, 2014; Patil et al., 2016), filler-gap dependencies (e.g., Gordon, Hendrick, Johnson, & Lee, 2006; Gordon, Hendrick, & Levine, 2002; Van Dyke & McElree, 2006), and negative polarity dependencies (e.g., Vasishth et al., 2008; Xiang, Dillon, & Phillips, 2009). For example, Badecker and Straub (2002) tested processing of pronouns and reflexives, using a self-paced reading time task. In sentences like (13), the retrieval trigger him/himself has multiple candidates – John and Bill or Jane and Bill – and one of the candidates’ gender feature is 15 manipulated to either mismatch with the gender of the anaphor, as in (13b), or match the gender feature of the anaphor, as in (13a). (13) a. John thought that Bill owed {himself/him} another opportunity to solve the problem. b. Jane thought that Bill owed {himself/him} another opportunity to solve the problem. Badecker and Straub observed longer reading times at the pronoun/reflexive region when both candidates match the gender of the pronoun/reflexive, compared to when their gender features were different. Thus, this study provides evidence that the presence of multiple candidate antecedents can cause interference during reference dependency formation. In sum, the key intuition relevant to my work from the Cue-based Approach is that multiple candidates can be accessed and activated in parallel, and that the competition and inferences stemming from the presence of multiple antecedents that fully or partially match the features of the anaphor can have an impact on the retrieval process. 1.3.2.2. Cue-based Parallel Function Strategy (Cue-based PFS) The Parallel Function Strategy (PFS) resembles the Cue-based Retrieval Approach in that both view reference resolution as a search process based on feature-matching. However, they differ in that the Cue-based Retrieval Approach provides a more detailed account of the search and retrieval process by presenting how the language processor accesses and activates items in our memory. As previously discussed, the PFS does not give an explicit explanation of (i) whether there are referential dependencies other than the one with the fully-matched item and (ii) how the interaction between the different dependencies, if any, can affect pronoun interpretation. This missing 16 information is important for addressing the nature of multiple pronoun interpretation. In contrast, the Cue-based Retrieval Model provides a possible answer to the missing information: multiple items can establish dependencies with a retrieval trigger because each retrieval cue contributes to each given item’s degree of activation, which gets stronger in proportion to the number of matching features. Moreover, since the memory activation of multiple items occur simultaneously, there follows an interference/competition between them during retrieval process. Here, I propose a Cue-based PFS 5 , a new take on the PFS inspired by the Cue-based Retrieval Model. The Cue-based PFS basically follows the reference resolution logic of the PFS while (i) the process of building dependencies (partially or fully) and (ii) the possibility of interference/competition between multiple items are adopted from the Cue-based Retrieval Model (missing information discussed in the paragraph above). More concretely, a pronoun is resolved through the feature-matching system, where the best antecedent is selected among the multiple candidates activated simultaneously, based on their number of shared features. In particular, under this view, grammatical role is one of the shared features (retrieval cues) that guides pronoun resolution. The examples in (14) show the main configurations I investigate in this dissertation. Under the Cue-based PFS system, referential differences between the One-Pronoun (14a) and the Two- Pronoun configuration (14b) can be captured. If the Cue-based PFS is applied to the One-Pronoun example in (14a), I predict that the preceding subject Henry is more likely to be retrieved than the preceding object Kevin for the pronoun he. This is because Henry is fully matched in all features 5 There are two types of mechanisms in Cue-based Retrieval Models: (i) a serial search process which posits that the parser retrieves the necessary information in a given item by its structural location. (e.g., Gillund & Shiffrin, 1984; Knuth, 1965; McElree & Dosher, 1993) versus (ii) a direct access, content-addressable mechanism which posits that the parser retrieves the necessary information by direct access to a given item (e.g.,Clark & Gronlund, 1996; McElree, Foraker, & Dyer, 2003; Parker et al., 2017). The Cue-based PFS focuses on the direct access, content-addressable mechanism. 17 including the grammatical role [subject], but Kevin is not fully matched (the grammatical role [object]), while both Henry and Kevin can be activated in parallel by the retrieval cues. (14) a. Henryi cheated Kevinj because hei>j daxed. {3sing, masc, sbj} {3sing, masc, obj} {3sing, masc, sbj} b. Henryi cheated Kevinj because hei>>j daxed himi<<j. {3sing, masc, sbj} {3sing, masc, obj} {3sing,masc,sbj} {3sing,masc,obj} When the Cue-based PFS is applied to the Two-Pronoun sentence like (13b), a big difference from the One-Pronoun sentence is revealed. The Two-Pronoun configuration contains two retrieval triggers (he and him), and so a search process is initiated two times, and both antecedents (Henry and Kevin) are activated twice. As illustrated in Figure 2b, each antecedent is linked to the two different pronouns, forming dual dependencies at the end of the sentence, and one of the dual dependencies is activated more than the other due to the grammatical role difference (e.g., the subject Henry has a stronger link with the subject pronoun he than with the object pronoun him). In contrast to the reference resolution process of the One-Pronoun sentence (Figure 2a), now the pronoun that appears later in the clause (object pronoun him) establishes the referential dependencies with the items that have been already linked to the other pronoun (subject pronoun he). 18 Figure 2a. One-pronoun structure Figure 2b. Two-pronoun structure While the Cue-based Retrieval Models predict the interaction/competition between the multiple items activated by a single retrieval trigger, it is not clear, to the best of my knowledge, how the multiple antecedents activated by multiple triggers in the same clause can interact with each other. I assume that patterns of subject pronoun interpretation in the Two-Pronoun sentence will be different from those of the One-Pronoun sentence at the end of the sentence, if the process of building referential dependencies with the object pronoun is affected by the pre-established dependencies with the subject pronoun, which are also affected by the dependency-building process of the object pronoun. That is to say, it may be possible in the Two-Pronoun sentence that interactions occur between potential antecedents that are activated by multiple triggers. Crucially, I assume that the interaction between the dual dependencies initiated from the two pronouns to each antecedent will be realized in the form of competition such that the stronger dependency (parallel in grammatical roles) suppresses the other dependency (non-parallel in grammatical roles), resulting in stronger parallel interpretation at the end. More concretely, for the dual dependencies of each antecedent, a relatively high-activation dependency is formed with a fully-matched parallel pronoun (e.g., Kevin-him) but the other dependency is partially matched with the non-parallel pronoun (e.g., Kevin-he). When the parallel dependency (e.g., Kevin-him) suppresses the non-parallel dependency (e.g., Kevin-he) by winning the competition, the suppressed non-parallel dependency strengthens a parallel dependency of the other antecedent (e.g., Henry-he) in return. In sum, the Cue-based PFS predicts that pronoun resolution parallel to the Figure 1a. One-pronoun structure Figure 1b. Two-pronoun structure 19 grammatical roles will be stronger in the Two-Pronoun sentence than the One-Pronoun sentence. Although the parallel grammatical role consists of one of the retrieval cues in the current configurations like (14), I acknowledge that it is not a hard constraint as much as Binding Theory principle B (Chomsky, 1981) or morphological (phi) features such as person, number, and gender. Nonetheless, I assume that grammatical role should be included in the set of retrieval cues in some cases, including (a) contexts with two structurally parallel clauses (i.e. when syntactic parallelism holds) and (b) contexts where parallel coherence relation is inferred to hold between clauses (i.e. semantic parallelism holds), which is typically signaled by connectives like and similarly or the verbs with similar meanings (e.g., Hobbs, 1979, 1990; Kehler, 2002; Kehler et al., 2008). (15) a. Margaret Thatcheri admires Hillary Clintonj, and George W. Bush absolutely worships heri<j. b. # Margaret Thatcheri admires Ronald Raganj, and George W. Bush absolutely worships herij. (Kehler, 2002, p. 157-159) In the examples in (15), the two clauses are structurally and semantically parallel. In (15a), the object pronoun her is more likely to be interpreted as referring to the preceding object Hillary Clinton, although the parallel interpretation is contradicted by our world knowledge (e.g., Bush is more likely to worship Thatcher). Similarly, the sentence (15b) is confirmed to be infelicitous even if there is a gender-matching referent, Margaret Thatcher. In Kehler (2002), this infelicity is explained by the fact that the object pronoun her cannot be co-referenced with the parallel referent Ronald Ragan in (15b) due to the gender mismatch. These examples suggest that the grammatical role does play an important role in pronoun resolution when the context involves structural and semantic parallelism. Thus, I assume that grammatical role is one of the retrieval features at least 20 in the parallel configurations of this dissertation like (14), but for other configurations, I do not presume that the grammatical role must participate in the resolution process. It should be noted that the current version of the Cue-based PFS presented here ignores (probably incorrectly) coherence relations and only focuses on grammatical role. This simplified version of the Cue-based PFS may be sufficient for this dissertation where only explanation coherence relations (signaled by because) with nonce verbs are tested. This is because the explanation relation does not guide pronoun interpretation without verbal semantic information and world knowledge (see Kehler 2002; Kehler et al 200f8 for more discussion). However, it is possible to extend this account to include coherence relations (e.g., by adding further retrieval cues), but I leave this as a question for future work. 1.3.3. Discourse coherence factor: Centering Theory (CT) Centering Theory (CT) (e.g., Grosz et al., 1995; Walker et al.,1998 see also Beaver, 2004; Kameyama, 1998; Miltsakaki, 2002, 2003, 2005; Poesio, 2004; Suri & McCoy, 1994; Taboada & Hadic Zabala, 2008; Taboada & Wiesemann, 2010) is a discourse-based models that accounts for the local coherence of a discourse by focusing on the interactions between referential continuity and degree of salience among entities in discourse. According to CT, pronoun resolution is explained in terms of a larger model of discourse coherence, in contrast to the previously discussed (Cue-based) Parallel Function Strategy (PFS), where reference resolution largely results from an independent search process based on morpho-syntactic constraints. The general idea of CT is that a discourse is considered more coherent when the most central entity in discourse is held constant or unchanged between utterances, but it is considered as less coherent when the most central entity changes across utterances. The most coherent entity is similar to what is typically called ‘topic’ 21 (see Walker et al., 1998; Beaver, 2004; Poesio, 2004 for relevant discussions). In terms of pronoun resolution, CT provides a valuable insight: there is a preference to interpret anaphoric expressions so that the transitions between different utterances are as coherent as possible. (16) a. Jeffi helped Dickj wash the car. b. Hei washed the windows as Dickj waxed the car. c. Hei soaped a pane. [he = Jeff] c’. Hej buffed the hood. [he = Dick] (Walker et al. 1998, p. 1) In example (16), sentences (a) and (b) are about Jeff, and the most central entity Jeff does not change between them. The same most salient entity is maintained if (16b) is followed by (16c), but it is shifted to Dick in (16c’). CT predicts that (16c) will be easier to process than (16c’), because (16c) yields a more coherent transition – as the most central entity in the previous utterances (Jeff ) is pronominalized – than (16c’) where the most central entity is changed to a different entity (Dick). In CT, the degree of discourse coherence is connected to the relative salience of centers, which are defined as the semantic entities in each utterance. The centers are distinguished into three types – forward-looking centers (Cf), a preferred center (Cp), and a backward-looking center (Cb). The Cfs represent a list of discourse entities evoked by each utterance and are partially ordered based on their relative salience. For English, it is commonly assumed that the level of salience is reflected by grammatical roles, as shown in (17) (e.g., Brennan, et al., 1987; Walker et al., 1998). 22 (17) Cf ranking by grammatical roles: Subject > Object(s) > Other On the Cf ranking for English in (17), the least oblique element, the subject, is ranked the highest. Accordingly, the discourse entities in the Cf-list are ranked accordingly. Although it is widely assumed that grammatical role is central in ranking Cfs in English, this may vary across languages. For example, Strube and Hahn (1999) propose that information-structural factors impact the Cf ranking in German, with discourse-old or hearer-old entities ranked higher than discourse-new or hearer-new entities. Once the discourse entities in the Cf-list are ordered, the highest ranked entity in the list is the preferred center (Cp). Thus, the Cp corresponds to the most salient entity in each utterance. CT also identifies a backward-looking center (Cb), which represents the highest ranked entity from the previous utterance that is realized in the current utterance. Since the Cb links the most salient entity of the previous utterance to the current utterance, it is considered as the closest concept to what is typically known as ‘topic’ (see Walker et al., 1998 and Beaver, 2004, Poesio, 2004 for relevant discussions). Moreover, Grosz et al. (1983) claimed that the Cb is the most central entity in each utterance due to its contribution to local coherence as well as local salience. Based on the relationship between the Cp and the Cb, CT provides four types of transition state between the two utterances – CONTINUE, RETAIN, SMOOTH-SHIFT and ROUGH-SHIFT. These transitions reflect the local coherence of the discourse based on the core idea that the less the most salient or central entity (Cp/Cb) changes within an utterance as well as across utterances, the more coherent the transition is. CT posits that some transition states are preferred over others depending on discourse coherence. 23 Cb (Ui) = Cb (Ui-1), or Cb (Ui-1) = [?] Cb (Ui) ≠ Cb (Ui-1) Cb (Ui) = Cp (Ui) CONTINUE SMOOTH-SHIFT Cb (Ui) ≠ Cp (Ui) RETAIN ROUGH-SHIFT Table 1. Center transition states (Walker et al., 1998, p.5) As Table 1 illustrates, the most coherent transition is CONTINUE, where (a) the Cb of the current utterance (Ui) is unchanged from the previous utterance (Ui-1) or there is no Cb 6 in the previous utterance ([?]), and (b) the Cb is also same as the Cp of the current utterance. This situation corresponds a case where the given entity that has been discussed in the prior utterance continues to be discussed in the current utterance. The RETAIN transition is argued to be less coherent than CONTINUE because the Cb of the current utterance (Ui) is not the same as its Cp, while the current Cb is unchanged from the previous one (Ui-1) or uninstantiated in the previous utterance ([?]). The RETAIN transition signals that the central entity that was talked about in the preceding utterance is ‘downgraded’ in the current utterance into a less salient position in the Cf-list. There are two types of SHIFT transitions, both claimed to be less coherent than RETAIN. In SHIFT transitions, the Cb of the previous utterance (Ui-1) is changed in the current utterance (Ui). This change indicates that the given entity that was discussed in the previous utterance is changed in the current utterance. Crucially, the two types of the SHIFT transitions differ in terms of whether the current central entity (Cb (Ui)) is realized in the most salient position (Cp (Ui)) or not: The SMOOTH-SHIFT is more coherent transition than the ROUGH-SHIFT because the 6 Walker et al (1994, 1998) called the notation Cb = [?] an uninstantiated Cb (c.f., a similar notion as Kameyama’s (1986) ‘Center Establishment’). It was proposed to deal with the cases where there is no Cb in an utterance. The uninstantiated Cb is typically found in the initial utterance, where the uninstantiated Cb becomes specified in the following utterance. 24 former transition signals that the central entity of the current utterance is still mentioned in the next utterance, but the latter signals a further shift of the central entity in the following utterances. In sum, the transitions are ordered with regard to preference as summarized in (18). (18) “Transition states are ordered. The CONTINUE transition is preferred to the RETAIN transition, which is preferred to the SMOOTH-SHIFT transition, which is preferred to the ROUGH-SHIFT transition.” (Walker et al., 1998, p.4) In essence, keeping one’s attention ‘centered’ on the same entity across utterances yields a maximally coherent discourse, and thus shapes comprehenders’ pronoun interpretation biases. In other words, when resolving pronouns, multiple interpretations may correspond to different types of transition. Among them, the interpretation that maximizes coherence of transitions is preferred over the other transitions. Let us now consider the possible interpretations of the One-Pronoun configuration tested in this dissertation, shown in (19). 7 (19) a. Janei called Maryj [Cf : Jane, Mary Cp : Jane Cb : ∅] b. because shei liked Tom. [Cf : Jane, Tom Cp : Jane Cb : Jane ] - CONTINUE b’. because shej liked Tom. [Cf : Mary, Tom Cp : Mary Cb : Mary ] – CONTINUE 7 Here I have shifted to discussing a sequence of only two, not three clauses, unlike what we saw in the examples in (16) above. While this kind of short sequence is not what CT was initially developed for, there are several recent studies that have also used CT for these kinds of two-clauses sequences, as I discuss in more depth at the end of this section. Furthermore, later theoretical modifications to CT, discussed by Walker et al (1994, 1998) – building on Kameyama’s (1986) idea of ‘Center Establishment’ – incorporate sentences without Cbs into the typology of transitions. They introduce the notation Cb = [?] – what we can call an uninstantiated Cb. Such uninstantiated Cbs are typically found in discourse-initial utterance and become specified in the subsequent utterance. 25 The two-clause sentence (19) contains one pronoun in the following clause, which can refer to either antecedent (Jane vs. Mary). According to CT, no matter which antecedent is referred to by the pronoun, the interpretations are equally preferred: both yield the most coherent transition (i.e., CONTINUE). Regardless of whether she refers to Jane or Mary, the Cb of the second clause is identical to its Cp (the Cb of the first clause is uninstantiated). Thus, both the preceding subject and object are equally ‘good’ antecedents for she, as they yield equally coherent transitions. Recall that the main aim of this dissertation is to elucidate the differences between multiple pronoun resolution and single pronoun resolution (i.e., One-Pronoun vs. Two-Pronoun configurations). Basically, CT predicts that pronoun resolution can pattern differently depending on whether all or only one of the referents in the preceding clause are mentioned with a pronoun in the subsequent clause, because this can affect discourse coherence. In the Two-Pronoun configuration, there is another position to mention discourse-old entities (e.g., subject/object- position pronoun), which enables both the most salient entity and less salient entity of the preceding utterance to be mentioned in the current utterance. In some cases, when the non-Cp entity of the previous utterance (e.g., object antecedent) is upgraded to the higher-salience position (e.g., subject-position pronoun), the Two-Pronoun configuration involves the previous Cp (e.g., subject antecedent) that is downgraded to the lower-salience position (e.g., object-position pronoun), unlike the One-Pronoun configuration. Thus, differences in pronoun interpretation can yield different transition states. (20) a. Janei called Maryj [Cf : Jane, Mary Cp : Jane Cb : ∅] b. because Shei liked herj. [Cf : Jane, Mary Cp : Jane Cb : Jane ] - CONTINUE b’. because Shej liked heri. [Cf : Mary, Jane Cp : Mary Cb : Jane ] - RETAIN 26 Example (20) is the same as (19) except that now the second clause contains two pronouns, which presumably refer to both of the antecedents in the first clause (Jane and Mary). Unlike the One-Pronoun sentence (19), CT predicts an asymmetry between the two ways of resolving the pronouns – both the subject-position pronoun she and the object-position pronoun her refer to parallel antecedents in (20b) but to non-parallel antecedents in (20b’). Between the two interpretations, according to CT (20b) is preferred over (20b’) because (20b) yields a more coherent transition (CONTINUE) than (20b’) (RETAIN). This difference is due to the fact that the Cb of the second clause is identical to its Cp only under the parallel interpretation (20b), but different under the non-parallel interpretation (20b’). Taken together, the key implication is that CT predicts that the patterns of pronoun interpretation will differ as a function of the referential structure of the pronoun-containing clause. Before continuing, it’s worth noting that in prior work using CT, the question about the basic unit of utterance has been a particular interest because CT computes discourse coherence by utterance unit. However, there is no consensus regarding what counts as an utterance – in particular, whether an utterance should be defined as a sentence (e.g., Grosz, et al., 1995; Miltsakaki, 2002, 2003, 2005) or as a tensed clause (e.g., Kameyama, 1998; Suri & McCoy, 1994; Taboada & Hadic Zabala, 2008; Taboada & Wiesemann, 2010). The Sentence-based View is supported by a series of studies by Miltsakaki. She argued that subordinate-clause entities are less topical than matrix- clauses entities based on the results of sentence-completion tasks in English and Greek, where subject pronouns in the subordinate clauses showed weaker subject-antecedent preference than those in the main clauses. However, the Sentence-based View was challenged by Taboada & Hadic Zabala’s (2008) corpus study (using telephone conversations in English and Spanish from the CallHome corpus). They evaluated the Sentence-based View and the Tensed Clause-based View 27 with regard to the four characteristics of CT: (i) the number of empty Cbs, (ii) coincidence between Cbs and topics, (iii) coincidence between Cp (Ui-1) and Cb (Ui), and (iv) the number of antecedents for pronouns in the same utterance. While there were almost no differences between the two views in their results with regard to the characteristics (i) - (iii), the Tensed Clause-based View was found to perform better than the Sentence-based View regarding the characteristic (iv). Based on the finding that pronouns cannot be properly resolved unless the clause is an independent unit of utterance, they argue in favor of the Tensed Clause-based View. Moreover, de la Fuente (2015) argued for the Tensed Clause-based View when two clauses are in a causal relation (e.g., connected by because) but argued for the Sentence-based View when two clauses are in a temporal relation (e.g., connected by when). In my dissertation, given that the primary domain of inquiry is the referential dependencies between potential antecedents and multiple pronouns, and the resolution process is examined under explanation relation context; I follow the Tensed Clause-based View. In addition, I’d like to point out that many experiments in this dissertation involve predictions regarding centering transition states between two utterances (instead of three or more). This is partly due to reasons related to experimental design: many psycholinguistic studies have used two-sentence sequences and thus by using a similar set up, I can maximize the comparability and relevance of my results for prior psycholinguistic experiments. However, I acknowledge that, within the centering theoretic tradition, predictions regarding transitions are typically made for sequences of at least three utterances, in the original spirit of CT (e.g., Grosz et al., 1995). Nevertheless, there are also some studies that apply CT to two-clause sequences (e.g., Kameyama 1998, Walker et al., 1994), and thus my approach is not without precedent. In addition, Walker et al.’s (1998) version of CT proposes a new type of Cb, uninstantiated Cb, which allows transition state to be predicted between sequences of utterances where the Cb of the initial utterance is 28 underspecified. Thus, the fact that in this dissertation I apply insights from centering to two-clause sequences as well as three-clause sequences builds on these prior studies. 1.4. Structure of this dissertation This dissertation aims to shed light on the mechanisms that underlie pronoun resolution, by investigating which factors guide multiple pronoun interpretation and whether it differs from single pronoun interpretation. These issues are examined by comparing pronoun interpretation in configurations where only one pronoun is present in a clause (One-Pronoun structures) to configurations with two pronouns in a clause (Two-Pronoun structures). I propose that the crucial difference between these two configurations lies in referential structure – whether all or only one of the referents in the preceding clause are mentioned with a pronoun in a subsequent clause. The experiments reported in this dissertation explore whether this difference in referential structure type can influence pronoun interpretation in offline and online tasks. The dissertation is organized as follows: Chapter 2 reports an initial series of picture- description experiments on English testing whether and how the process of reference resolution of multiple pronouns differs from single pronouns. By testing the referential structure effects with different types of verbs (Experiment 1 & 2: psych verbs and Experiment 3: action verbs), Chapter 2 also investigates whether referential structure effects generalize across verb classes with different thematic roles. Chapter 3 explores the mechanisms underlying referential structure effects, focusing on how factors at different levels of representation – in particular, syntactic information and discourse-level information – contribute to referential structure effects. I report a picture- description experiment on English (Experiment 4) that disentangles discourse-coherence related 29 factors from syntactic parallelism related factors. Chapter 4 looks at referential structure effects with a different type of anaphoric form, namely null arguments in Korean (henceforth, null pronouns for ease of exposition, see section 4.1. for relevant discussions). Using the same kind of picture-writing task as in Experiments 1 to 4, Experiment 5 investigates whether and how the patterns of null pronoun (pro) resolution differ in One-pro vs. Two-pro contexts, thereby also shedding light on whether null pronouns (in Korean) differ from overt pronouns (in English) in this regard. Chapter 5 turns top the question of real-time processing. In a webcam-based visual-world eye-tracking study (Experiment 6), I examine referential structure effects from the perspective of incremental processing, focusing on how rapidly referential structure effects will emerge. As a whole, this dissertation provides novel evidence for the differences between multiple pronoun resolution and single pronoun resolution through both offline and online studies, and shows that these differences stem from differences in referential structure. 30 Chapter 2: The effects of referential structure in pronoun interpretation 2.1. Introduction Current models of reference resolution show that pronoun interpretation is guided by various factors at different levels of representation such as syntax, semantics and discourse. However, most coreference models have explored how these factors influence the likelihood of a particular referent being interpreted as the antecedent of a subsequent single pronoun, as in (21) – with the exception of work on parallelism effects (e.g., Chambers & Smyth 1998; Sheldon, 1974; Smyth, 1994; Stevenson et al., 1993, 1995). Although Two-Pronoun sentences as in example (22) are frequent in naturalistic language use, previous psycholinguistic work has not systematically investigated the nature of the mechanisms that govern multiple pronoun resolution. Thus, it is not clear whether the current models of pronoun resolution (focusing on single pronoun resolution) are sufficient to account for multiple pronoun resolution. As an initial step towards these larger questions, this chapter investigates whether the process of reference resolution of two pronouns in the same clause could be different from that of a single pronoun. (21) Jennyi feared Maryj because shei/j verbed (Paula). [One-Pronoun structure] (22) Jennyi feared Maryj because shei/j verbed herj/i. [Two-Pronoun structure] In this dissertation, the difference between single-pronoun resolution and two-pronoun resolution is examined in configurations with two potential antecedents that are identical in gender. In the experiments reported in this chapter, the pronoun-containing clause contains either one pronoun in the subject position (One-Pronoun configuration: she verbed or she verbed Paula (21)) 31 or two pronouns in both the subject and the object positions (Two-Pronoun configuration: she verbed her (22)). I explore the idea that when pronoun interpretation is ambiguous (a given pronoun could in principle be interpreted as referring to either antecedent, as in (21) and (22)), the crucial difference between these two configurations lies in the interpretational consequences of a specific type of information in the pronoun-containing clause, namely its referential structure. As discussed in Chapter 1, the referential structure of these configurations differs in the number of referential expressions that need to form a pronominal dependency relation with an antecedent. In the One- Pronoun configuration as in (21), there is one pronoun (e.g., she) that the processing system ultimately needs to connect to one of the two preceding antecedents (e.g., Jenny or Mary). However, in the Two-Pronoun structure as in (22), there are two pronouns (e.g., she and her) and the processing system ultimately needs to find a distinct antecedent for each pronoun (e.g., Jenny and Mary). Thus, the term referential structure describes this difference – whether all or only one of the referents in the preceding clause are mentioned with a pronoun(s) in the subsequent clause. (In this dissertation I focus on clauses with two referents, and leave clauses with three or more referents for future work.) In this chapter, I report initial evidence that interpretation of subject-position pronouns is different between the One-Pronoun and the Two-Pronoun configurations, and I attribute this to the differences in referential structure (the different number of referential dependencies that the language processing system activates). In the following sections, I present two hypotheses about whether and how referential structure affects pronoun interpretation. The hypotheses differ in the claims they make about whether and how the referential dependencies involving the two different pronouns interact with each other and affect pronoun interpretation. In the rest of this chapter, I 32 report three experiments that test these hypotheses by means of an end-state focused picture- writing task. 2.2. Hypotheses on referential structure I present two hypotheses regarding the question of whether pronoun interpretation is guided by referential structure – the Independence View vs. the Dependence View. I further sketch out three possibilities about how the interaction of coreferential operations at different levels of representation can yield referential structure effects by proposing three subtypes of the Dependence View – (i) the Syntax-level Dependence View, (ii) the Discourse-level Dependence View, and (iii) the Syntax-Discourse Dependence View. Please note that three experiments reported in Chapter 2 (Experiment 1,2, and 3) focus on testing the Independence vs. the Dependence Views as an initial step towards addressing the question about whether and how multiple pronoun resolution is different from single pronoun resolution. Although three different subtypes of the Dependence View are introduced in this chapter, the current studies do not yet aim to distinguish between these three subtypes of the Dependence View. In the next chapter (Chapter 3), the three subtypes will be disentangled, as I take a closer look at how factors at different levels of representation contribute to referential structure effects. 2.2.1 Independence View The Independence View hypothesizes that when resolving multiple pronouns in Two-Pronoun configuration, the anaphoric dependencies constructed for the two different pronouns are resolved fully independently. The Parallel Function Strategy (PFS), defined in (23), is an example of this view (e.g., Smyth, 1994; Chambers & Smyth 1998; see also Stevenson et al., 1993, 1995; Sheldon, 33 1974). (23) Parallel Function Strategy (PFS): A pronoun is coreferential with a preceding noun phrase occupying the same grammatical role as the pronoun. (Stevenson et al., 1995, p. 394) At least when taken at face value, the PFS does seem to not make any explicit claims about subject pronoun interpretation being different between One-Pronoun sentence as in the example (24a) and the Two-Pronoun sentence as in (24b) because the PFS claims that pronoun resolution is a search process based on a feature-matching system where the anaphoric processor decides the best antecedent based on the number of features, such as a gender, number, person and grammatical role, shared by the pronoun (see Smyth (1994) for more details). Since the grammatical role is one of the features that contributes to this matching process, the parallel grammatical role guides pronoun resolution if the other features are identical between the potential antecedents. (24) a. Henryi cheated Kevinj because hei>j daxed. Tom. {3sing, masc, sbj} {3sing, masc, obj} {3sing, masc, sbj} b. Henryi cheated Kevinj because hei>j daxed himi<j. {3sing, masc, sbj} {3sing, masc, obj} {3sing,masc,sbj} {3sing,masc,obj} Note that, as I already mentioned in Chapter 1, I acknowledge that the effects of parallelism can be ‘undone’ due to coherence relations (e.g., Wolf et al., 2004) although it is not always the 34 case that parallelism effects are found only in parallel (resemblance) coherence relation. Nonetheless, for sake of completeness, this chapter explores the possibility of purely syntactic parallelism effects arising even in contexts where the clauses are structurally parallel but semantically non-parallel, as in (24). Let us consider the basic configurations of the studies in this chapter from the perspective of the Independence View. In the One-Pronoun configuration (24a), the best antecedent for the subject-position pronoun he is the preceding subject Henry because the antecedent parallel in grammatical role shares the greatest number of features with the pronoun. Likewise, in the Two- Pronoun configuration (24b), the best antecedent for the subject-position pronoun he is the preceding subject Henry, and the best antecedent for the object-position pronoun him is the preceding object Kevin because the parallel grammatical role guides pronoun resolution. Since the resolution processes initiated from the two different pronouns (the subject and object pronoun) are mutually exclusive based on this claim, the Independence View does not predict that the patterns of subject-position pronoun interpretation would be sensitive to referential structure of the pronoun-containing clause (i.e., whether it contains one or two pronouns). It may seem puzzling to assume that a comprehender would build a dependency between a pronoun and an antecedent while ignoring potential existing dependencies. However, this idea is not without precedent. For example, one possible explanation is found in the serial search account, one of the earlier models of memory retrieval (e.g., Ehrlich, 1980; Gordon & Hendrick, 1998; Gillund & Shiffrin, 1984; Knuth, 1965; McElree & Dosher, 1993). According to this account, the parser selectively accesses a target item in memory based on its linear or structural location. The search process begins with the most local item and continues until a fully feature-matching item is found, without retrieving any partially-matching items. For example, under this full-matching 35 requirement, the subject pronoun he in the One-Pronoun sentence (24a) and in the Two-Pronoun sentence (24b) only forms a referential dependency with a fully-matching item like the subject Henry, because a partially-matching item like the object Kevin can never be retrieved even though its features are evaluated first due to its closer location. Similarly, the search triggered by the object pronoun him in the Two-Pronoun sentence (24b) first encounters the object Kevin and stops searching for farther items (e.g., the subject referent) and thus retrieves the object referent – because a fully-matched item has already been found. If the feature matching system of the PFS follows the serial search mechanism and exhibits this kind of full-matching requirement, the referential dependencies constructed for the two pronouns could (in principle) be resolved fully independently. However, it should be noted that the PFS is sometimes compatible with a situation where subject pronoun resolution in the Two-Pronoun configuration patterns differently from the One- Pronoun configuration, given that the PFS posits that parallelism effects can operate together with other constraints such as a structural priming (e.g., Bock, 1986) or a subject assignment strategy (e.g., Crawley et al., 1990). The PFS hypothesizes that parallel interpretation can be facilitated in structurally parallel clause due to structural priming and that the “subject assignment is the default when the two clauses are nonparallel” (Smyth 1994, p.221). However, given that “when two clauses are fully parallel, the probability of coindexing a pronoun with a parallel NP by the feature-match process is at its maximum” (Smyth, 1994, p.221), and the subject assignment strategy is very unlikely to come into play with the parallel clauses according to the PFS. Thus, I assume that the PFS predicts that referential dependencies from the two pronouns in the Two-Pronoun configuration will be constructed fully independently as long as the two pronouns appear in the parallel structure. Under the configurations used in this chapter, 36 subject pronoun interpretation would not be different between One-Pronoun sentence and the Two- Pronoun sentence. 2.2.2. Dependence Views Contrary to the Independence View, Dependence Views assume that resolving one of the pronominal dependencies influences the formation of the other dependency in the Two-Pronoun structure. Here I introduce three subtypes of the Dependence View depending on factors at different levels of representation that influence the interaction between multiple referential dependencies. 2.2.2.1. Syntax-level Dependence View First, the Syntax-level Dependence View hypothesizes that resolving anaphoric dependencies initiated from one of the pronouns in the Two-Pronoun configuration (e.g., He vs. Him in (24)) influences how the dependency for the other pronoun is resolved, and that syntax-level factors guide this dependency-building operation. The Syntax-level Dependence View predicts the different patterns of pronoun resolution depending on referential structure, contrary to the Independence View. Regarding the syntax-level factors’ influence on referential dependencies, the Independence View assumes that two different pronouns are resolved fully independently at least under the structurally parallel context because according to the PFS, construction of the referential dependencies involves only linguistic information itself, without considering how the information is accessed in memory. However, the Syntax-level Dependence View, which is built on a new approach on the PFS – namely, a Cue-based PFS – focuses on not only the morphosyntactic information but also the 37 nonlinguistic aspect that how comprehenders gain access to the information in memory. Recall that Chapter 1 introduced the Cue-based PFS, where the PFS is inspired by one of the memory retrieval models, a Cue-based Retrieval Approach (e.g., Bhatia & Dillon, 2022; Dillon, 2011; Hammerly, et al., 2019; Jäger et al., 2017; Lewis & Vasishth, 2005; McElree, 2000; Van Dyke & Johns, 2012; Van Dyke & McElree, 2011; Yadav, et al., 2022). The Cue-based Retrieval Approach claims that individual items are encoded and stored in memory as feature bundles. When the language processing system encounters a retrieval trigger (e.g., pronouns, reflexives, verbs), each retrieval cue in the trigger is matched against each feature of items in memory simultaneously and transfers activation to these items. During the search process, multiple items can be accessed and activated in parallel, and, accordingly, competition and interference can arise between items being retrieved because the item with higher activation levels, which has the higher number of matching features, is more likely to be retrieved. The Cue-based PFS basically follows the basic reference resolution logic of the PFS, and the process of building dependencies (with items that partially- or fully-match the retrieval features) and the possibility of interference/competition between multiple items are adopted from the Cue- based Retrieval Model. More concretely, a pronoun is resolved through the feature-matching system, where the best antecedent is selected among multiple candidates activated simultaneously, based on their number of shared features. In particular, I assume that grammatical role is one of the shared features (retrieval cues) that guides pronoun resolution. If the Cue-based PFS is applied to the One-Pronoun example as illustrated in Figure 3a, it is predicted that the fully-matched item (the preceding subject Henry) is more likely to be retrieved than the partially-matched item (the preceding object Kevin) for the pronoun he, while both Henry and Kevin are activated in parallel by the retrieval cues. 38 Figure 3a. One-pronoun structure Figure 3b. Two-pronoun structure When the Cue-based PFS is applied to the Two-Pronoun sentence as illustrated in Figure (3b), the search process is initiated two times by the two retrieval triggers (he and him), and both antecedents (Henry and Kevin) are activated twice. Crucially, I assume that the interaction between the dual dependencies initiated from the two pronouns to each antecedent will be realized in the form of competition such that a stronger dependency with a relatively high activation (parallel in grammatical roles) suppresses the other dependency (non-parallel in grammatical role), resulting in stronger parallel interpretation at the end. Specifically, when the parallel dependency (e.g., Kevin-him) suppresses the non-parallel dependency (e.g., Kevin-he) by winning the competition, the suppressed non-parallel dependency strengthens a parallel dependency of the other antecedent (e.g., Henry-he) in return. In sum, the Cue-based PFS predicts that pronoun interpretation parallel to the grammatical roles will be stronger in the Two-Pronoun sentence than the One-Pronoun sentence. Since the competition between the referential dependencies initiated from the two different pronouns do not exist in the process of single-pronoun resolution, the Syntax-level Dependence View predicts the different patterns of pronoun resolution depending on referential structure. 2.2.2.2. Discourse-level Dependence View The second type of the Dependence View is the Discourse-level Dependence View. It resembles the Syntax-level Dependence View in hypothesizing that resolving anaphoric dependencies Figure 1a. One-pronoun structure Figure 1b. Two-pronoun structure 39 initiated by one of the two pronouns influences how we interpret those triggered by the other pronoun, but crucially posits that discourse-level factors guide the process of dependency construction. This hypothesis is inspired by one of the discourse-based models, Centering Theory (CT), which accounts for local discourse coherence by focusing on the interactions between referential continuity and degree of salience among entities in discourse (e.g., Grosz et al., 1995; Walker et al.,1998 see also Beaver, 2004; Kameyama, 1998; Miltsakaki, 2002, 2003, 2005; Poesio, 2004; Suri & McCoy, 1994; Taboada & Hadic Zabala, 2008; Taboada & Wiesemann, 2010). As discussed in Chapter 1, the general idea of CT is that a discourse is considered as more coherent when the most central entity in discourse (similar to what is typically called ‘topic’, see Walker et al., 1998 and Beaver, 2004, Poesio, 2004 for relevant discussions) is held constant between utterances, but less coherent when the most central entity changes across utterances. CT proposes that how pronouns are interpreted interactively affects the coherence of discourse transitions – CONTINUE, RETAIN, SOFT-SHIFT and ROUGH-SHIFT – and, thus, there is a preference to interpret pronouns so that the transitions between different utterances are as coherent as possible, as summarized in (25). (25) “Transition states are ordered. The CONTINUE transition is preferred to the RETAIN transition, which is preferred to the SMOOTH-SHIFT transition, which is preferred to the ROUGH-SHIFT transition.” (Walker et al., 1998, p.4) Crucially, CT predicts that the coherence of transitions is influenced by referential structure (One-Pronoun vs. Two-Pronoun structure) because discourse coherence can be affected by whether 40 all or only one of the referents in the preceding clause are mentioned in the next utterance. To be specific, this is because there is another position to mention discourse-old entities in the Two- Pronoun configuration (e.g., subject/object-position pronoun) as in (27), which enables both the most salient entity and less salient entity of the preceding utterance to be mentioned in the current utterance; on the contrary, there is only one position for the discourse-old entity in the One- Pronoun configuration (namely, the subject-position pronoun) 8 as in (26), which allows either the most salient preceding entity or less salient preceding entity to be mentioned in the current utterance (26) a. What about Henryi? Hei cheated Kevinj because hei daxed Tom. [CONTINUE] b. What about Henryi? Hei cheated Kevinj because hej daxed Tom. [SOFT-SHIFT] (27) a. What about Henryi? Hei cheated Kevinj because hei daxed himj. [CONTINUE] b. What about Henryi? Hei cheated Kevinj because hej daxed himi. [RETAIN] For example, in the because-clause of the examples above, if the previous ‘topic’ (interchangeably used with the term Cb ‘backward-looking center’) Henry is mentioned in the salient subject position by means of the pronoun he (26a) and (27a), the same topic continues in the final clause as Henry. In this case, regardless of the presence vs. absence of another pronoun in the object position, both the One-Pronoun structure (26a) and the Two-Pronoun structure (27a) 8 For now, I put aside configurations with anaphoric nouns in object position (e.g., He verbed the man) where the man refers to an antecedent in the prior sentence. Thus, I use the term One-Pronoun configuration to mean ‘configurations with only one anaphoric element.’ 41 yield the most coherent transition, CONTINUE. On the contrary, if the previous ‘non-topic’ Kevin is mentioned by the subject pronoun he in the final clause, the topic (Cb) changes to Kevin in the One-Pronoun structure (26b), but the same topic (Cb) Henry continues in the Two-Pronoun structure (27b) with the object pronoun him referring to Henry. In this case, the Two-Pronoun structure (27b) yields a more coherent transition (RETAIN) than the One-Pronoun structure (26b) (SOFT-SHIFT). Thus, due to the difference in transition coherence resulting from referential structure, CT predicts that the patterns of pronoun interpretation will differ as a function of the referential structure of the pronoun-containing clause. In this case, it is predicted that the object antecedent bias for subject-position he will be stronger in the Two-Pronoun sentence (27) than in the One- Pronoun sentence (26), because the least coherent transition is established from the subject- position pronoun referring to the object antecedent in the One-Pronoun condition. In sum, the Discourse-level Dependence View predicts different patterns of pronoun resolution depending on referential structure. 2.2.2.3. Syntax-Discourse Dependence View The last type of the Dependence View is the Syntax-Discourse Dependence View. It hypothesizes that how we resolve anaphoric dependencies from one of the two pronouns affects dependency formation of the other pronoun, and both factors at the level of syntax and discourse guide the process of dependency construction. Under the syntax-discourse dependency view, I expect that dependency formation will be guided by both the syntax-level Cue-based PFS and the discourse- level coherence mechanism, CT. Basically, the Syntax-Discourse Dependence View predicts different patterns of pronoun resolution depending on referential structure because both the Cue- 42 based PFS and CT assume that there will be interaction between referential dependencies initiated from the two different pronouns in the same clause. In the following subsections, I will provide detailed predictions for each case. 2.3. Psycholinguistic work on effects of referential structure To best of our knowledge, the existence of referential structure effects on pronoun resolution has received little attention in psycholinguistic experiments. Although Gordon, Grosz & Gilliom (1993) did not focus on the kinds of referential structure effects investigated in the present paper, looked mostly at unambiguous pronouns, and did not probe pronoun resolution; their reading-time data provides evidence for many of the core tenets of Centering Theory. In other related work, Kaiser (2009) conducted a production-based fragment-completion task by manipulating syntactic form (it-cleft vs. SVO) and the grammatical role of the focused element (subject-focus vs. object-focus), as illustrated in (28). (28) a. [Cleft.Object/Subject = focus] A: The maid scolded the bride. B: No, that’s wrong! It was the secretaryfocus {that she scolded / who scolded her}. She . . . b. [SVO.Object/Subject = focus] A: The maid scolded the bride. B: No, that’s wrong! {She/ the secretaryfocus} scolded {the secretaryfocus /her}. She . . . . (Kaiser, 2009, p.342) 43 Kaiser found that subject-position pronoun interpretation is modulated by whether or not the other referent from the preceding sentence is mentioned later in the pronoun-containing sentence: On trials where participants interpreted the subject-position pronoun as referring to the preceding subject, they were more likely to mention the other argument from preceding clause more than on trials where they interpreted the subject-position pronoun as referring to the preceding object. Kaiser takes this as evidence that if a less salient referent from the preceding sentence is promoted by being mentioned in the subsequent subject-position with a pronoun, then participants will avoid mentioning a highly-salient referent from the preceding sentence later in the pronoun-containing sentence. However, this work only tested sentences with contrastive focus (i.e., a particular information-structural configuration), and did not systematically assess the impact of different kinds of referential structures or verb classes. Thus, although it provides valuable foundational data, more systematic investigation is needed to assess the validity of the effects of referential structure. Experiments in this chapter build on Kaiser (2009) but go beyond it by directly testing how and whether pronoun interpretation varies depending on differences in referential structure (One- Pronoun and Two-Pronoun configurations). Another way in which this work goes beyond Kaiser (2009) is by investigating referential structure effects in contexts with Implicit Causality (IC) verbs, which allow us to assess whether well-known IC effects replicate with our task while also building carefully on prior norming data that allows us to control verbs’ referential biases. I also test types of IC verbs (psych verbs and action verbs) in a systematic way to see whether referential structure effects generalize across verb classes with different thematic roles. In the next section, I discuss IC verbs in more depth. 44 2.4. Effects of Implicit Causality Experiments 1 to 3 use Implicit Causality (IC) contexts where two clauses are linked by the explanation connective because and the first clause is a transitive clause containing an IC verb. It is well known that subject-position pronouns following certain verbs, in the presence of an explanation relation, exhibit systematic interpretational biases depending on the verb (e.g., Bott & Solstad, 2014; Caramazza et al., 1977; McKoon et al., 1993; Hartshorne & Snedeker, 2013; Hartshorne et al., 2015; Ferstl et al., 2011). Researchers have identified subject-biased verbs (IC1 verbs) that elicit subject interpretations (e.g., X bothered Y because she…, X surprised Y because she…) as well as object-biased verbs (IC2 verbs) that elicit object interpretations (e.g., X worshipped Y because she…, X criticized Y because she….). I included an implicit causality manipulation for two main reasons. First, it makes it possible to assess whether classic implicit causality effects replicate with the new task developed for the purpose of the experiments reported in this chapter. If the same IC effects are found as previous studies, this would provide evidence that the method provides meaningful data about pronoun interpretation and can thus be used to test for effects of referential structure. Second, implicit causality contexts allow us to test referential structure effects with previously-established verbs’ referential biases being controlled. IC1 verbs trigger biases towards preceding subjects (e.g., surprised, terrified, frustrated, cheated) while IC2 verbs trigger biases towards preceding objects (e.g., respected, feared, hated, criticized), in contexts where the two clauses are linked by an explanation relation. Using the two types of IC verbs means that we can make use of previously-established biases towards either preceding referent (subjects and objects) to see whether and how these biases are modulated by referential structure. For example, if the same type of the IC verb is used in the preceding clause in both the One-Pronoun sentence (e.g., 45 Lisa surprised Kate because she verbed) and the Two-Pronoun sentence (e.g., Lisa surprised Kate because she verbed her), patterns of subject pronoun interpretation can be compared between the two configurations with the verb bias kept constant across conditions. In what follows, I report the outcomes of three experiments. Experiment 1 tests for effects of referential structure and implicit causality by comparing pronoun interpretation in intransitive and transitive continuations that contain one and two pronouns respectively, following clauses containing Stimulus-Experiencer (IC1) verbs (e.g., surprised, terrified, frustrated) and Experiencer-Stimulus (IC2) verbs (e.g., respected, feared, hated). Experiment 2 addresses a potential confound in Experiment 1 by eliminating differences in transitivity and comparing pronoun interpretation in transitive clauses with one vs. two pronouns, again after clauses with Stimulus-Experiencer (IC1) and Experiencer-Stimulus (IC2) verbs. Experiment 3 broadens the domain of investigation beyond Stimulus-Experiencer/Experiencer-Stimulus verbs and tests whether referential structure effects also obtain in configurations with verbs that have agentive subjects (e.g., cheated, criticized, blamed). As we will see, these three studies provide clear evidence in favor of our claim that referential structure of the entire pronoun-containing clause influences participants’ final (offline) interpretation of pronouns in subject position, supporting the Dependence Views. 2.5. Experiment 1 Experiment 1 aims to investigate whether pronoun resolution is affected by referential properties of the subsequent pronoun-containing clause by testing pronoun interpretation in intransitive and transitive continuations, containing one and two pronouns respectively. Since this experiment focuses on the presence vs. absence of referential structure effects, it basically tests the two 46 hypotheses – the Independence vs. the Dependence Views. The three subtypes of the Dependence View (the Syntax-level Dependence View, the Discourse-level Dependence View, and the Syntax- Discourse Dependence View) will not be distinguished in this chapter but will be individually compared in the next chapter (Chapter 3) to further investigate what factors at different levels of representation guide referential structure effects. 2.5.1. Participants Forty-five adult native speakers of English were recruited from Amazon’s Mechanical Turk (MTurk) and participated in this study. Five participants were excluded from data analysis because they were not self-reported US-born native speakers of English. In addition, no participants were excluded for their poor performance on catch trials, as everyone met the pre-specified threshold (four or more correct out of six catch trials). Thus, a total of 40 participants were included in the final analyses. Note that all experiments reported in the dissertation were reviewed and approved by the Institutional Review Board of USC, and all their exclusion criteria were determined prior to data analysis. 2.5.2. Materials and design In Experiment 1, I employed a 2 × 2 design by manipulating (i) the type of IC verb in the first clause (IC1 vs. IC2) and (ii) the nature of the referential structure type in the second clause (One- Pronoun vs. Two-Pronoun). An example 9 is given in Table 2. 9 As mentioned in Chapter 1, I acknowledge that, within the centering theoretic tradition, predictions regarding transitions are typically made for sequences of at least three utterances, in the original spirit of CT (e.g., Grosz et al., 1995). Nevertheless, there are also some studies that apply CT to two-clause sequences (e.g., Kameyama 1998, Walker et al., 1994), and thus my approach is not without precedent. In addition, Walker et al.’s (1998) version of CT proposes a new type of Cb, uninstantiated Cb, which allows transition state to be predicted between sequences of utterances where the Cb of the initial utterance is underspecified. Thus, the 47 Verb Type Referential structure Examples IC1 (SE) Two-Pronoun Henry surprised Kevin because he daxed him One Pronoun Henry surprised Kevin because he daxed. IC2 (ES) Two-Pronoun Henry respected Kevin because he daxed him. One Pronoun Henry respected Kevin because he daxed. Table 2. A set of example stimuli in Experiment 1 In the first clause, two referents were introduced in the subject position and the object position (e.g., Henry and Kevin) along with the IC verb. The verb semantics was manipulated by using Stimulus-Experiencer (SE) verbs (subject-biased IC1 verbs) vs. Experiencer-Stimulus (ES) verbs (object-biased IC2 verbs). Thus, I tested IC1 verbs (SE) and IC2 verbs (ES). To ensure that the SE verbs were indeed subject-biased and that the ES verbs were object- biased in the expected way, I used the norms from Hartshorne & Snedeker (2013) and Ferstl et al. (2011) to select 24 SE verbs (mean subject bias=67.4%, SD=13.6) and 24 ES verbs (mean object bias=76.3%, SD=11.7). Moreover, the two referents matched in gender so that the following pronouns can be ambiguous in reference resolution. The gender of the names was balanced across target items (12 female name and 12 male names). In the second clause, referential structure was manipulated by containing one or two pronouns. In the Two-Pronoun structure, two pronouns are present in the subject and object position (e.g., he daxed him), and both of the referents in the first clause are mentioned by them. In the One-Pronoun structure, only one pronoun is present in the subject position (e.g., he daxed), and one of the preceding referents is mentioned by it. All the verbs used in the second clause were nonce verbs (e.g., daxed, zoobed, frobbed). This allowed us to minimize potential effects of these fact that in this dissertation I apply CT insights to two-clause as well as three-clause sequences builds on the prior studies. 48 verbs’ semantics on pronoun interpretation, a question I do not aim to test in the present work. Each nonce verb was used once. In addition, the first and the second clause were connected with a connective because for each target sentence, based on the previous findings (e.g., Rohde, 2008; Rohde & Kehler, 2008) – the IC verb effects on pronoun resolution have been clearly found in explanation relation. In addition to the 24 targets, the study included 36 fillers. The fillers were of various types, including ambiguous relative-clause structures and negative ‘stripping’ structures (e.g., X verbed Y but not Z). In addition, six of these fillers were unambiguous catch trials that allowed us to check whether participants were paying attention to the task. The target and filler items were pseudo- randomly intermixed and presented using a Latin-Square design. The full list of target stimuli can be found in Appendix A. 2.5.3. Procedure I used a picture-writing task, where participants saw a sentence coupled with a schematized picture depicting the underlined part of the sentence. An example is in Figure 4. In target items, the underlined part was the critical pronoun-containing clause. The participants’ task was to type the names of the characters into the textboxes. This reveals how they interpret the pronoun(s). The study was conducted using Qualtrics (Provo, UT). 49 Figure 4a. Example Two-Pronoun item from Experiment 1 Figure 4b. Example One-Pronoun item from Experiment 1 The target pictures included one or two stick figures (depicting the characters involved in the event), and a nonce verb for the action in the pronoun-containing clause (e.g., daxed). For one- character pictures (Figure 4b; intransitive One-Pronoun conditions), the character in the picture is the one who does the action (Agent). For pictures with two characters (Figure 4a; transitive Two- Pronoun conditions), the character at the beginning of the arrow does the action (Agent), and the character at the end of the arrow is the undergoer (Patient/Theme). This was explained to participants as part of the instructions. Thus, with two-character images, the nonce verbs are transitive and with one-character images they are intransitive. The direction of the arrows (leftward vs. rightward) and the location of the agents and Patients/Themes (left vs. right) in the two-character pictures were counterbalanced. The task was to type a name into each text box such that the picture fits with the underlined part of the sentence. For example, in Figure 4a, if a participant interprets the second clause as ‘Henry daxed Kevin’, they should type Henry in the left textbox and Kevin in the right textbox. Thus, participants’ responses reveal how they interpret the pronouns. 50 2.5.4. Data processing and analysis Responses were coded according to which of the referents in the first clause (subject or object) was chosen as the referent of the subject-position pronoun: I analyze how often this subject- position pronoun in the second clause is interpreted as referring to the object of the first clause by coding object choices as 1 and subject choices as 0. Because there are only two possible antecedents, looking at how the subject-position pronoun is interpreted also provides information about how the object pronoun (if present) is interpreted: If the subject-position pronoun refers to the preceding subject, the object-position pronoun refers to the preceding object; if the subject- position pronoun refers to the preceding object, the object-position pronoun refers to the preceding subject. However, the trials that were responded by the names that did not refer to the preceding subject or the preceding object were excluded from analysis by marking them as N/A (0.7% of all responses). Thus, 99.3 % of the data were submitted for the statistical analysis. The final data were analyzed with a generalized linear mixed logit model (glmer), using R statistical software (R Development Core Team, 2019) and the lme4 package (Bates et al., 2015) and lmertest (version 3.1.1) (Kuznetsova et al., 2017). The dependent variable was the proportion of object interpretations. As fixed effects, referential structure (contrast-coded, One-Pronoun = 0.5, Two-Pronoun = −0.5), IC verb type (contrast-coded, IC1 = -0.5, IC2 = 0.5), and the referential structure x IC verb type interaction were included in models. As random effects, intercepts for subjects and items were entered as well as by-subject and by-item random slopes for the effects of IC verb type, referential structure type, and their interaction. A maximum random effect structure was used when justified by model comparison. The initial random effect structure that was fully crossed and fully specified with by- subject and by-item effects of IC verb type, referential structure type, and their interaction was 51 simplified (starting with by-item effects) via model comparison. Only random effects that show a significant contribution to the model (p < 0.05) were included (Baayen et al., 2008). 2.5.5. Predictions Experiment 1 asks whether pronoun resolution is affected by referential properties of the subsequent pronoun-containing clause. Since the goal of this first experiment is simply take initial steps in terms of looking for referential structure effects, it tests the Independence vs. the Dependence Views, without yet distinguishing the different subtypes of the Dependence View (see Chapter 3). Moreover, referential structure effects are tested in the presentence of well-known implicit causality (IC) effects, to confirm whether the novel task of Experiment 1 reliably detects the expected IC bias effects. Implicit causality effects: I expect to replicate the IC verb effects seen in prior work. If pronoun interpretation is guided by IC information of the preceding verb, it is predicted that a subject-position pronoun will be more likely to refer to the preceding object antecedent in the IC2 (ES) verb condition than in the IC1 (SE) verb condition. Referential structure effects: Two different predictions are made regarding the effects of referential structure, depending on whether or not resolving one of the pronominal dependencies influences the formation of the other dependency. The three types of Dependence View (Syntax- level, Discourse-level, and Syntax-Discourse Dependence View) predict that the Two-Pronoun structure will pattern differently from the One-Pronoun structure in terms of how people interpret the subject-position pronoun such that object antecedent preference of the subject-position pronoun will be stronger in the One-Pronoun structure than the Two-Pronoun structure. On the other hand, the Independence View predicts that interpretation of the subject-position pronoun 52 will pattern alike in One- and Two-Pronoun structures. Let us consider the Dependence Views first. The first type of the Dependence View, the Syntax-level Dependence View, predicts that the object antecedent bias of the subject pronoun (e.g., he in Figure 5) is weaker in the Two-Pronoun structure, relative to the One-Pronoun structure. The Cue-based PFS predicts that each pronoun forms a stronger dependency with the antecedent parallel in grammatical roles because the interactions between the dual dependencies are realized in the form of competition that a stronger dependency suppresses the other dependency that is linked to the same antecedent. Thus, the object antecedent will be more likely to be retrieved for the subject-position pronoun in the One-Pronoun structure, compared to the Two-Pronoun structure. As illustrated in Figure 5, when the object-position pronoun is present, the object antecedent (Kevin) becomes more strongly connected to the object-position pronoun (him), which weakens the dependency connecting it to the subject-position pronoun (he). Thus, compared to the case where the object-position pronoun is absent (One-Pronoun structure), the object antecedent bias of the subject pronoun is expected to be weaker in the Two-Pronoun structure. Figure 5a. One-pronoun structure Figure 5b. Two-pronoun structure The second type of Dependence View, the Discourse-level Dependence View, predicts the identical patterns of reference resolution – the object antecedent bias will be weaker in the Two- Pronoun condition than in the One-Pronoun condition in resolving the subject-position pronoun. However, the underlying mechanism is different from the Syntax-level Dependence View. Figure 1a. One-pronoun structure Figure 1b. Two-pronoun structure 53 According to Centering Theory (CT), the most coherent transition in the Two-Pronoun condition can be established from the subject-position pronoun referring to the subject antecedent (CONTINUE in (c)), not the object antecedent (RETAIN in (d)), as given in Table 3. In contrast, in the One-Pronoun condition, the most coherent transition (CONTINUE in (a,b)) is elicited, no matter which antecedent is referred to by the subject-position pronoun. In other words, the topic (Cb, backward-looking center) of the following utterance can be either the preceding subject (Henry) or the preceding object (Kevin). If there are no other preceding referents to be mentioned (One-Pronoun structure), the two options make the discourse transition equally coherent. However, if both of the preceding referents are to be mentioned (Two-Pronoun structure), the topic of the following because clause becomes the preceding subject (Henry) regardless of the position it is re- mentioned in. Thus, re-mentioning the higher-salient referent (preceding subject) in the less salient position (object-position pronoun) yields a less coherent transition than re-mentioning it in the salient position (subject-position pronoun), because the former signals that the topic might be shifted in the next utterance. CT Transition Ref. structure Examples CONTINUE One-Pronoun a. Henryi verbed Kevinj because hei daxed. b. Henryi verbed Kevinj because hej daxed. CONTINUE Two-Pronoun c. Henryi verbed Kevinj because hei daxed himj. RETAIN d. Henryi verbed Kevinj because hej daxed himi. Table 3. Transition type predicted by CT in Experiment 1 The last type of the Dependence View, the Syntax-Discourse Dependence View, predicts that the patterns of referential structure effects will be the same as the other two views because the 54 syntax-level and the discourse-level co-referential mechanisms are expected to exert an influence on pronoun resolution in the same direction. Thus, it is predicted that the object antecedent bias will be weaker in the Two-Pronoun condition compared to the One-Pronoun condition in resolving the subject-position pronoun. On the contrary, the Independence View predicts that there will be no effects of referential structure on subject-position pronoun interpretation. According to the PFS, the coreferential processes from the subject-position pronoun and the object-position pronoun in the Two-Pronoun configuration are mutually exclusive at least in the structurally parallel context because only their grammatical roles guide the interpretation (other features being equal). As the grammatical roles of the two pronouns are different, their anaphoric dependencies are expected to be resolved independently in the Two-Pronoun structure – the subject pronoun prefers to be co-referential with the subject antecedent while the object pronoun prefers to be co-referential with the object antecedent. Likewise, the subject-position pronoun in the One-Pronoun structure 10 is predicted to show preference for the subject antecedents. Thus, it is predicted that subject-position pronoun interpretation in the One-Pronoun and Two-Pronoun conditions will pattern alike. 2.5.6. Results Figure 6 shows the proportion of trials on which the subject-position pronoun was interpreted as referring to the preceding object, as function of verb type (ES vs. SE) and referential structure (One-Pronoun vs. Two-Pronoun). 10 Since two clauses are structurally non-parallel (a transitive matrix clause is connected to an intransitive embedded clause), the subject assignment strategy might guide pronoun interpretation as claimed by Smyth (1994). Even if it is the case, the subject assignment strategy predicts the same as the PFS: the subject antecedent preference. Note that regardless of parallelism in structure, the subject-position pronoun is predicted to be interpreted as referring to the subject antecedent by the PFS and the subject assignment strategy. 55 Figure 6. Experiment 1: The proportion of trials where the subject-position pronoun refers to the preceding object. (error bars represent ±1 S.E.) Recall that the focus of Experiment 1 is simply on testing whether the Independence View or the Dependence Views better matches the data, in order to address the question of whether two- pronoun resolution is different from single pronoun resolution, not distinguishing between the three subtypes of the Dependence View. As can be seen in Figure 6, referential structure influences pronoun interpretation in the ES (IC2) condition in the direction predicted by the Dependence Views of referential structure hypotheses – the Syntax-level, the Discourse-level, and the Syntax- Discourse Dependence Views. To be specific, with ES verbs, the presence of an object-position pronoun in the second clause (Two-Pronoun condition) markedly reduces the percentage of object interpretations (69.6%) relative to configurations without an object-position pronoun (One- Pronoun condition) (92.1%). However, with SE verbs, the Two-Pronoun condition (43.3%) does not differ from the One-Pronoun condition (42.1%) in this regard: Figure 6 suggests that in the SE verb conditions, pronoun interpretation is not sensitive to the One-Pronoun vs. Two-Pronoun manipulation. In addition, Figure 6 shows that IC verb type influences pronoun interpretation in the 56 expected direction – more object interpretations with ES/IC2 verbs compared to the SE/IC1 verbs. As can be seen in Table 4, I find main effects of referential structure, IC verb type and a significant interaction. Further planned comparisons (Table 5) show that the effect of referential structure is significant in the ES conditions, but not in the SE conditions, as expected based on Figure 6. The patterns in the ES condition support the prediction of the Dependence Views such that object preference is stronger in the One-Pronoun than in the Two-Pronoun conditions. However, no effect of the referential structure was found in the SE condition. Fixed effect Estimate Std. Error z value Pr(>|z|) (Intercept) 0.8162 0.2221 3.675 0.000238 *** Ref.st type 0.8589 0.2010 4.273 1.93e-05 *** IC type 2.3195 0.1963 11.815 < 2e-16 *** Ref.st type:IC type 1.9618 0.3800 5.163 2.43e-07 *** Table 4. Experiment 1: results of the glmer model IC type Fixed effect Estimate Std. Error z value Pr(>|z|) SE-verbs (Intercept) -0.3737 0.3018 -1.238 0.216 Ref.st type -0.1730 0.2223 -0.778 0.436 ES-verbs (Intercept) 2.0400 0.2895 7.046 1.84e-12 *** Ref.st type 1.9911 0.3277 6.075 1.24e-09 *** Table 5. Experiment 1: planned comparisons 2.5.7. Discussion As a first step towards elucidating the nature of multiple pronoun resolution, Experiment 1 tested whether pronoun resolution is guided by the referential structure of the pronoun-containing clause, 57 by comparing the Independence View vs. the Dependence Views. Although I sketched out three subtypes of the Dependence View (the Syntax-level Dependence View, the Discourse-level Dependence View, and the Syntax-Discourse Dependence View), the current study does not aim to distinguish between them to further investigate what factors at different levels of representation would guide referential structure effects (see Chapter 3 for this question). The effects of referential structure were tested under implicit causality (IC) context, where implicit causality of the verb in the preceding clause was manipulated. As regards implicit causality, our results replicate earlier results: subject-position pronouns were more likely to be interpreted as referring to objects after Experiencer-Stimulus (ES/IC2) verbs than after Stimulus-Experiencer (SE/IC1) verbs. This is in line with prior work (e.g., Bott & Solstad, 2014; Caramazza et al., 1977; McKoon et al., 1993; Ferstl et al., 2011, Hartshorne & Snedeker, 2013, Hartshorne et al., 2013) and also serves as a ‘sanity check’ to show that the expected patterns replicate with our novel picture-writing task and with nonce verbs in the second clause. As regards referential structure, I find significant effects with Experiencer-Stimulus (ES) verbs: the One-Pronoun configuration shows a stronger object bias than the Two-Pronoun configuration. This finding supports the three types of the Dependence View – the syntax-level, the discourse-level 11 , and the Syntax-Discourse Dependence View – that predicted the same 11 One might wonder whether the Cue-based PFS could be expanded to include a feature such as topicality. While this is an interesting question for future work, for now I’d like to point out that simply adding a [topic] feature appears unlikely to be able to capture the fine-grained predictions that I currently generate using insights from Centering Theory. For example, under the Cue- based PFS, it is unclear (a) which referents in the preceding context can have a [topic] feature (e.g., a sentence topic vs. a discourse topic) and (b) whether a pronoun can have the [topic] feature regardless of its syntactic position (e.g., subject-position pronoun vs. object-position pronoun). For instance, let us assume that the sentence topic referent (e.g., subject) can have the [topic] feature and the pronouns in all syntactic positions can have the [topic] feature. In a Two-Pronoun sentence (e.g., Henry respected Kevin because he daxed him), there could be competition between the subject pronoun and object pronoun for the subject antecedent (sentence topic) to build referential dependencies, while there would be no competition for the subject antecedent in a One-Pronoun sentence (e.g., Henry respected Kevin because he daxed Tom). In this case, the object antecedent preference of the subject-position pronoun would be stronger in the Two-Pronoun structure than the One-Pronoun structure – but this is the exact opposite of what I find in Experiment 1. Overall, for now, due to the unclear aspects on topicality in the cue-based model, a full discussion of this issue is left for future work. 58 patterns of the referential effects: the interpretation of subject-position pronouns in One-Pronoun conditions exhibits a stronger object preference (and a weaker subject preference) than in Two- Pronoun conditions. It shows that the resolution of one of the anaphoric dependencies influences the resolution of the other. However, with Stimulus-Experiencer (SE) verbs, I fail to see a stronger object preference in the One-Pronoun condition relative to the Two-Pronoun condition. Rather, the patterns of pronoun interpretation were found to be insensitive to the referential structure of the pronoun- containing clause. The results seem to in line with the Independence View. However, the null result with SE verbs might stem from a particular semantic property of SE verbs coinciding with a potential confound between the One-Pronoun and Two-Pronoun conditions. To see why, let us first note that in the Two-Pronoun condition, the pronoun-containing clauses contained transitive (nonce) verbs (e.g., she frobbed her), whereas in the One-Pronoun condition, they used intransitive (nonce) verbs (e.g., she frobbed). This transitivity manipulation allowed us to construct One-Pronoun vs. Two-Pronoun configurations, but it is also associated with semantic differences: prior work shows that subjects of intransitive verbs are less semantically agentive than subjects of transitive verbs (e.g., Dixon, 1979; Hopper & Thompson, 1980). This means that the referents of the subject pronouns in the (intransitive) One-Pronoun conditions are likely to be regarded as less agentive and less volitional than the referents of the subject pronouns in the Two-Pronoun conditions. Second, it has been claimed that subjects of ES and SE verbs also differ in agentivity, with subjects of SE verbs (which have the thematic role of ‘Stimulus’) being less agentive than subjects of ES verbs (which have the thematic role of ‘Experiencer’) (e.g., Van Valin & LaPolla, 1998). Put together, these two considerations (intransitive subjects being less agentive and SE 59 subjects being less agentive) mean that sentences in the SE (IC1) + One-Pronoun condition are likely to be interpreted as having relatively non-agentive subjects in both clauses. If pronoun interpretation is guided by a preference for semantic role parallelism, such that pronouns with a (non)agentive role prefer antecedents with a (non)agentive role, this could result in the subject preference in the SE (IC1) + One-Pronoun condition being boosted (and the object preference being lowered) – which would explain why there is no referential structure effect with SE verbs (IC1). To address this concern and to test whether we find support for Dependence Views of referential structure hypotheses when the thematic roles in the One-Pronoun and Two-Pronoun configurations are held constant, Experiment 2 used transitive verbs in both clauses. 2.6. Experiment 2 The results of Experiment 1 provide evidence that referential structure of the pronoun-containing clause guides pronoun interpretation in contexts with Experiencer-Stimulus (ES) verbs, but show no effects of referential structure with Stimulus-Experiencer (SE) verbs. As discussed above, this unexpected lack of across-the-board referential structure effects may be due to concerns related to verb transitivity and associated semantic properties. To test the effects of referential structure more directly by comparing the Independence View vs. the Dependence Views, in Experiment 2, the nonce verbs in the pronoun-containing clause were consistently transitive. The Two-Pronoun conditions were as in Experiment 1 (e.g., she frobbed her), and the One-Pronoun conditions now consist of transitive sentences with a new name in object position (e.g., she frobbed Kate). I will refer to these as Pronoun+Name conditions. 60 2.6.1. Participants Forty-eight adult native speakers of English, who had not participated in Experiment 1, took part. Participant recruitment was identical to Experiment 1. Four participants were excluded from data analysis because they were not self-reported US-born native speakers of English, gave random answers throughout the experiment or showed poor performance on catch trials (eight or more correct out of eleven catch trials). 12 Additional four participants were excluded for balancing the number of participants in each list. Thus, 40 participants were included in the final analyses. 2.6.2. Materials and design The design and materials of Experiment 2 were the same as Experiment 1, except that I now manipulated referential structure without introducing an (in)transitivity confound, as shown in Table 6. The critical pronoun-containing clause is now transitive in both the Two-Pronoun condition (e.g., he daxed him) and the Pronoun+Name condition (e.g., he daxed Tom). In addition to the 24 targets, Experiment 2 included 36 fillers (including 11 catch trials). Verb Type Referential structure Examples IC1 (SE) Two-Pronoun Henry surprised Kevin because he daxed him Pronoun+Name Henry surprised Kevin because he daxed Tom. IC2 (ES) Two-Pronoun Henry respected Kevin because he daxed him. Pronoun+Name Henry respected Kevin because he daxed Tom. Table 6. A set of example stimuli in Experiment 2 12 The number of catch trials in Experiment 2 is higher than in Experiment 1, because Experiment 2 included transitive as well as intransitive sentences as unambiguous catch trials. This was done to ensure participants saw a mix of transitive and intransitive sentences over the course of Experiment 2 (all transitive targets), to keep it similar to Experiment1 (in/transitive targets). 61 2.6.3. Procedure The task and the procedure were the same as in Experiment 1. All target items now involve two characters (Figure 7), because all nonce verbs in the pronoun-containing clause denoted transitive events. The displays that participants saw in the Two-Pronoun conditions were the same as in Experiment 1 (Figure 4a). An example display for the Pronoun+Name condition is provided in Figure 7b. Figure 7a. Example Two-Pronoun item from Experiment 2 Figure 7b. Example Pronoun+Name item from Experiment 2 2.6.4. Data processing and analysis Experiment 2 was analyzed as in the same way as in Experiment 1. 1.1% of all responses were excluded from all the data (marked as N/A) because the subject-position pronoun or object- position pronoun was interpreted as referring to neither the subject nor the object of the first clause. Thus, in total of 98.9 % of the data were included in the statistical analyses. Statistical analyses were carried out the same was as Experiment 1. Experiment 2 also used a generalized linear mixed logit model (glmer) with the proportion of object interpretations as the dependent variable. As fixed effects, referential structure (contrast-coded, Pronoun+Name = 0.5, Two-Pronoun = −0.5), verb type (contrast-coded, IC1 = -0.5, IC2 = 0.5), and the referential structure x verb type interaction were included in models. Random effects were determined in the same way as in 62 Experiment 1. 2.6.5. Predictions The predictions are the same as for Experiment 1. 2.6.6. Results Figure 8 presents the proportion of trials on which the subject-position pronoun is interpreted as referring to the preceding object. Figure 8. Experiment 2: The proportion of trials where the subject-position pronoun refers to the preceding object. (error bars represent ±1 S.E.) In Experiment 2, I find a main effect of the IC verb type in the expected direction (more object interpretations with ES/IC2 verbs), a main effect of referential structure (more object interpretations with Pronoun+Name conditions than Two-Pronoun conditions), and an interaction between verb type and referential structure (stronger effects of referential structure with ES than SE verbs). See Table 7 for statistical details. Importantly, planned comparisons (Table 8) confirm that there are more object interpretations in the Pronoun+Name configurations than Two-Pronoun 63 configurations with both SE verbs (Pronoun+Name: 56.7% vs. Two-Pronoun: 45.4%) and ES verbs (Pronoun+Name: 87.5% vs. Two-Pronoun: 66.7%). Thus, although I obtain an interaction, I now see that – in contrast to Experiment 1 – Experiment 2 reveals significant effects of referential structure with both verb classes. Fixed effect Estimate Std. Error z value Pr(>|z|) (Intercept) 0.8970 0.2148 4.175 2.98e-05 *** Ref.st type 1.1625 0.1931 6.021 1.73e-09 *** IC type 1.6657 0.1818 9.163 < 2e-16 *** Ref.st type:IC type 1.0760 0.3554 3.027 0.00247 ** Table 7. Experiment 2: results of the glmer model IC type Fixed effect Estimate Std. Error z value Pr(>|z|) SE-verbs (Intercept) 0.09751 0.26557 0.367 0.71349 Ref.st type 0.66565 0.24535 2.713 0.00667 ** ES-verbs (Intercept) 3.590 1.196 3.003 0.00267 ** Ref.st type 5.375 2.294 2.343 0.01914 * Table 8. Experiment 2: planned comparisons 2.6.7. Discussion The results of Experiment 2 echo Experiment 1 in showing that both referential structure and IC verb bias guide pronoun interpretation. In a set-up where I eliminate potential confounds from transitivity, I now find significant referential structure effects with both SE and ES verbs: the preference for object antecedents is stronger in the Pronoun+Name configuration (where only one of the referents from the preceding clause is mentioned) than in the Two-Pronoun configuration 64 (where both of the referents from the preceding clause are mentioned), with both verb types. This finding is in line with the Dependence Views, which posit that the ultimate interpretation of pronouns is influenced by referential structure of the entire pronoun-containing clause, i.e., whether the other referent from the preceding clause is also mentioned in that clause. The results provide supporting evidence to the Dependence Views, which assume that the multiple-pronoun resolution process is fundamentally different from the single-pronoun resolution due to the interactions between the anaphoric dependencies of the two different pronouns.(Like Experiment 1, Experiment 2 does not yet seek to disentangle the three subtypes of the Dependence View – the Syntax-level Dependence View, the Discourse-level Dependence View, and the Syntax-Discourse Dependence View – which I turn to in Chapter 3.) Experiment 2 also shows an interaction between IC verb type and referential structure: referential structure effect is present with both ES and SE verbs, but stronger with ES verbs than with SE verbs. This verb-type difference, while entirely compatible with the Dependence Views, is not directly predicted by them. I speculate that it may stem from differences in the thematic roles associated with the subjects and objects of ES vs. SE verbs and the assumptions that participants make about the thematic roles associated with the subject and object of the nonce verbs. I return to this in the General Discussion section. As a whole, Experiment 2 provides evidence in favor of the Dependence View of referential structure hypotheses. However, both Experiments 1 and 2 used psych verbs (ES/SE verbs), with Experiencer and Stimulus thematic roles. This raises the question of whether referential structure effects also generalize to other verb classes. To assess the robustness of referential structure effects, Experiment 3 tests Agent-Patient and Agent-Evocator verbs. These verb classes have clearly agentive subjects and describe the actions of the event participants, in contrast to ES/SE psych 65 verbs that describe experiencers’ mental states. 2.7. Experiment 3 Experiment 3 tests whether referential structure effects extend to action verbs with different thematic roles than the psych verbs used in Experiments 1 and 2. Given that Dependence Views of referential structure hypotheses make no reference to verb semantics, these views predict that referential structure effects should occur with all verb types (other things being equal). 2.7.1. Participants Seventy-one adult native speakers of English participated in the experiment in exchange for $2.75. Participant recruitment and exclusion criteria were identical to Experiment 2. Three participants were excluded because they were not self-reported US-born native speakers of English. Five participants were excluded for giving random answers throughout the task or failing to pass the threshold for catch trials (at eight or more correct out of 11 catch trials). Three additional participants were excluded for the purpose of balancing the number of participants per list. Thus, 60 participants were submitted for final analyses. 2.7.2. Materials and design The materials and design are the same as Experiment 2, except that the first clause now used Agent- Patient (AP) verbs (which have an IC1 bias) and Agent-Evocator (AE) verbs (which have an IC2 bias), instead of ES/SE verbs. 13 Examples are in Table 9. Twenty-four AP and AE verbs were 13 The semantic role of Evocator is, semantically, equivalent to Patient. The term ‘Evocator’ was introduced in the implicit causality literature (e.g., Au, 1986; Rudolph & Forsterling, 1997; Ferstl et al., 2011) to distinguish agent-patient verbs that elicit IC1 bias (Agent-Patient) from those that elicit IC2 bias (Agent-Evocator). Thus, for our purposes, all of the verbs have an Agent argument 66 selected based the norms collected by Hartshorne & Snedeker (2013) and Ferstl et al. (2011) to ensure that the AP verbs are indeed subject-biased (Mean subject bias=67.7%, SD=9.16) and that the AE verbs are object-biased (Mean object bias=72.1%, SD=5.53). Other than this verb change, Experiment 3 used the same 24 targets and the same 36 fillers (including 11 items that function as catch trials) as Experiment 2. Verb Type Referential structure Examples IC1 (AP) Two-Pronoun Henry cheated Kevin because he daxed him Pronoun+Name Henry cheated Kevin because he daxed Tom. IC2 (AE) Two-Pronoun Henry criticized Kevin because he daxed him. Pronoun+Name Henry criticized Kevin because he daxed Tom. Table 9. A set of example stimuli in Experiment 3 2.7.3. Procedure Experiment 3 was conducted following the same procedure as Experiment 1 and Experiment 2. Examples of the displays are given in Figure 9. and a Patient argument. 67 Figure 9a. Example Two-Pronoun item from Experiment 3 Figure 9b. Example Pronoun+Name item from Experiment 3 2.7.4. Data processing and analysis Experiment 3 was analyzed like Experiments 1 and 2. Of all the data, 0.3% of the responses were removed (marked as N/A) because the subject-position pronoun or object-position pronoun was not interpreted as referring to either the preceding subject or object. In total, 99.7 % of the data were submitted for analysis. The data were statistically analyzed in the same way as Experiment 1 and Experiment 2, using a generalized linear mixed logit model (glmer). Models included the same dependent variable as in the previous experiments (the proportion of object interpretations), and fixed effects of referential structure (contrast-coded, Pronoun+Name = 0.5, Two-Pronoun = −0.5), verb type (contrast-coded, IC1 = -0.5, IC2 = 0.5), and the referential structure x verb type interaction. In addition, random effects were determined as in Experiments 1 and Experiments 2. 2.6.5. Predictions If the effects of referential structure generalize across verb classes, the predictions are same as for Experiments 1 and 2. 68 2.7.6. Results Figure 10 shows the proportion of trials where the subject-position pronoun is interpreted as referring to the object of the preceding clause. Figure 10. Experiment 3: The proportion of trials where the subject-position pronoun refers to the preceding object. (error bars represent ±1 S.E.) Experiment 3 reveals main effects of both referential structure and IC verb type, in the expected directions, and no interaction (Table 10). Thus, Experiment 3 replicates the expected IC effects (more object interpretations with object-biased AE verbs than with subject-biased AP verbs). As regards referential structure, Figure 10 shows that subject-position pronouns exhibit a stronger object preference in Pronoun+Name conditions than in Two-Pronoun conditions both with AP verbs (Pronoun+Name: 82.2% vs. Two-Pronoun: 68.3%) and AE verbs (Pronoun+Name: 85.8% vs. Two-Pronoun: 70.8%). The absence of an interaction shows that this preference is equally strong with AE verbs and AP verbs. This ‘across-the-board’ increase in object choices in the Pronoun+Name condition relative to the Two-Pronoun condition is exactly in line with the predictions of the Dependence View of referential structure hypotheses. 69 Fixed effect Estimate Std. Error z value Pr(>|z|) (Intercept) 2.3126 0.3803 6.082 1.19e-09 *** Ref.st type 1.6993 0.3706 4.585 4.54e-06 *** IC type 0.3937 0.1821 2.162 0.0306 * Ref.st type:IC type 0.3233 0.3624 0.892 0.3723 Table 10. Experiment 3: results of the glmer model 2.7.7. Discussion Experiment 3 replicated and extended the findings of Experiments 1 and 2, which tested Experiencer-Stimulus (ES) and Stimulus-Experiencer (SE) verbs, to agentive verb classes, namely Agent-Patient (AP) and Agent-Evocator (AE) verbs. Crucially, Experiment 3 again finds effects of referential structure: The preceding object was more likely to be interpreted as the antecedent of a subsequent subject-position pronoun in the Pronoun+Name than in the Two-Pronoun configuration. These results are in line with the Dependence Views on referential structure hypotheses which state that a subject-position pronoun is more likely to be interpreted as referring to the preceding object when the pronoun-containing clause only mentions one referent from the prior sentence. These findings support the Dependence Views that assume that resolving anaphoric dependencies initiated from one of the pronouns in the Two-Pronoun configuration influences how the dependency for the other pronoun is resolved. (We take a closer look at the three subtypes of the Dependence View – the Syntax-level Dependence View, the Discourse-level Dependence View, and the Syntax-Discourse Dependence View – in Chapter 3.) Combined with Experiments 1 and 2, Experiment 3 demonstrates the robustness of referential structure effect: It occurs with verbs that denote agentive, voluntary actions (AP/AE verbs) in addition to verbs that denote mental state (SE/ES verbs). 70 Furthermore, in line with prior work on implicit causality, Experiment 3 shows that subject- position pronouns are more likely to be interpreted as referring to the preceding subject with AP (IC1) verbs than AE (IC2) verbs. Thus, Experiment 3 extends the results of Experiments 1 and 2 by further corroborating effects of implicit causality and shows that these effects can be reliably detected with our experimental paradigm and in the presence of nonce verbs. Overall, the results of Experiment 3 confirm that, ultimately, pronoun resolution is guided not only by the IC information in the preceding clause but also by the referential structure of the entire pronoun-containing clause, indicating that referential structure effects are strong enough to be observed with verbs of different semantic classes. 2.7.8. Comparing the three experiments So far, we have seen evidence from three experiments supporting the Dependence Views – i.e., the idea that pronoun interpretation is ultimately sensitive to whether a clause mentions only one or both of the referents from the preceding clause. To the best of my knowledge, this work is the first experimental demonstration of this effect. Orthogonal to this key point, it is interesting to note that, at least visually, a comparison of Figure 6, 8, and 10 suggests that the rate of object interpretations is higher in Experiment 3 (with agentive action verbs) than in Experiments 1 and 2 (with ES/SE psych verbs). Indeed, statistical analyses (glmer) comparing the proportion of object interpretations with AP (IC1) verbs (Experiment 3) relative to SE (IC1) verbs (Experiments 1 and 2) confirm that AP verbs elicit higher rates of object interpretations than SE verbs (Exp3 vs. Exp1: beta = -2.22, SE = .46, z = -4.74, p<.001; Exp3 vs. Exp2: beta = -1.74, SE = .44, z = -3.85, p<.001). Thus, in our experiments, AP (IC1) verbs elicit higher rates of object interpretations than SE (IC1) verbs. This is the case even 71 though based on corpus norms, the subject bias strengths of the AP verbs (mean subject bias=67.7%, SD=9.16) and SE verbs (mean subject bias=67.4%, SD=13.6) do not differ significantly (t(46)=0.3558, p>.7). In fact, a similar asymmetry was reported by Stevenson et al. (1994), who used a variety of verb classes and connectives in fragment-completion studies (e.g., Ken admired Geoff ./and/because/so he …). Crucially, this difference in object bias strength between AP and SE verbs is orthogonal to the main claims I am making, especially since I make no predictions about (and nothing hinges on) absolute differences between verb types’ object bias strength. What matters is that both Experiments 2 and 3 show effects of verb bias as well as effects of referential structure. 2.8. General discussion Despite the large body of psycholinguistic work on reference resolution, most prior investigations have focused on the production and comprehension of a single pronoun, with little attention given to the potential presence or consequences of pronouns occurring later in the same clause. In this chapter, I highlight the importance of looking at effects of referential structure in pronoun interpretation, which is assumed to be resulted from the interactions between multiple referential dependencies initiated from different pronouns. As an initial step towards ultimate understanding of the interactions between the coreferential operations, Chapter 2 tested two hypotheses regarding referential structure: the Independence View vs. the Dependence Views. (The more detailed question of how the interaction of coreferential operations is affected by factors at different levels of representation, namely Syntax-level, Discourse-level, and Syntax-Discourse Dependence View is addressed in Chapter 3). The Dependence Views are different from the Independence View in the presence or 72 absence of the interactions between the referential dependencies. The Independence View predicts that there will be no differences in subject-position pronoun interpretation by referential structure type clause (e.g., she verbed (Kate) vs. she verbed her) because the anaphoric dependencies constructed for the two different pronouns are resolved fully independently according to the PFS (Parallel Function Strategy); on the contrary, the Dependence Views predict that subject pronouns in clauses that only mention one of the referents from the preceding clause are more likely to be interpreted as referring to a preceding object than subject pronouns in clauses that mention both referents from the preceding clause, because of interactions between multiple referential dependencies. To look for effects of referential structure, I compared (i) configurations where the second clause only mentions one of the two entities from the preceding clause with a subject-position pronoun (e.g., because he daxed vs. because he daxed Tom) to (ii) configurations where the pronoun-containing clause mentions both entities from the preceding clause with pronouns (e.g., he daxed him). Three experiments were conducted using both non-agentive IC verbs (SE/ES verbs in Experiments 1 and 2) and agentive IC verbs (AP/AE verbs in Experiments 3) in the preceding clause to assess the robustness of referential structure effects in different contexts. The findings from Experiment1 through Experiment 3 show that comprehenders’ ultimate interpretation of pronouns is modulated by the referential structure of the entire pronoun-containing clause, supporting the Dependence Views. 2.8.1. Referential structure effects All three experiments show overall effects of referential structure on pronoun interpretation: a pronoun in the subject position is less likely to be interpreted as referring back to the preceding 73 object in configuration with two pronouns (e.g., she verbed her) than in configurations with only one pronoun (e.g., she verbed, she verbed Kate). These findings are in line with Kaiser’s (2009) production results – where referential structure and preceding verb type were not explicitly manipulated – which suggest that if a preceding object is referred to by the subject position of a following clause (by means of a subject-position pronoun), the preceding subject is less likely to be mentioned later in the same clause. More specifically, in Experiment 2 and Experiment 3, referential structure effects are observed with both IC1 and IC2 verbs, while in Experiment 1, these effects are only observed with IC2 verbs (not IC1 verbs). I suggest that the lack of referential structure effects with IC1 verbs in Experiment 1 likely stems from the fact that Experiment 1 compared transitive and intransitive clauses, whose subjects differ in agentivity. Once this potential confound was addressed in Experiments 2 and 3, I found referential structure effects with both IC1 and IC2 verbs. These results were obtained with clauses that have the same surface syntactic structure and differ only in their referential properties. In other words, regardless of whether the verb has a baseline bias towards the preceding subject (IC1) or the preceding object (IC2) in explanation contexts like the ones I tested, the likelihood of a subject-position pronoun being interpreted as referring to the preceding subject vs. object is influenced by the presence or absence of another pronoun in the same clause. These results are compatible with the Dependence Views, but do not support the Independence View. The results suggest that multiple-pronoun resolution process is fundamentally different from single-pronoun resolution due to the interactions between the anaphoric dependencies of the two different pronouns. Contrary to the Independence View, the data suggest that the referential dependencies constructed for the two different pronouns are not resolved fully 74 independently, indicating that a purely syntax-based version of the PFS cannot straightforwardly capture referential structure effects. Since Chapter 1 aims for the initial step of testing the Dependence vs. the Independence View, it is not clear which factors are a source of the interactions between the referential dependencies: whether it is because the stronger dependency suppresses the other (weaker) dependency linked to the same antecedent (i.e., Syntax-level Dependence View) or because the referential properties of clauses contribute to discourse coherence (i.e., Discourse- level Dependence View) or both (i.e., Syntax-Discourse Dependence View). In the next chapter, I further investigate this issue by detangling the different subtypes of Dependence View from each other. As a whole, the results of all three comprehension experiments indicate that referential properties of the entire pronoun-containing clause are utilized during pronoun resolution. These findings also have methodological implications for existing work using the sentence-completion paradigm. In this widely-used paradigm, participants are typically given a clause (e.g., Lisa impressed Tom), with or without a connective and/or a pronoun prompt at the end (e.g., Lisa impressed Tom because he/she), and asked to write a completion. It is currently standard practice to focus only on the subject of the second clause. The question of whether the other referent from the initial clause is re-mentioned in a later non-subject position in the continuation is typically not investigated. However, the current results suggest that the rest of the continuation should also be analyzed, and that focusing only on the subject position may lead researchers to overlook meaningful differences between conditions (in terms of whether the other referent is or is not mentioned). 75 2.8.2. Implicit Causality effects Another aim of this chapter was to test whether well-known implicit causality (IC) effects are replicable with a novel method (the picture-writing task). I manipulated verb semantics in the preceding clause by using different classes of IC verbs, and compared IC verbs known to have a subject bias (IC1 verbs) to IC verbs known to have an object bias (IC2 verbs). Experiment 1 and Experiment 2 used non-agentive IC1 and IC2 verbs with Stimulus and Experiencer arguments (known as ‘psych verbs’). Experiment 3 used IC1 and IC2 verbs with agentive subjects. In all three experiments, a significant effect of verb bias was observed: Subject-position pronouns consistently showed a stronger object preference with IC2 verbs than IC1 verbs. These results replicate prior work (e.g., Bott and Solstad, 2014; Caramazza et al., 1977; Ferstl et al., 2011; Hartshorne et al., 2015; Hartshorne & Snedeker, 2013; McKoon et al., 1993) and thus confirm that the picture- writing task, even with nonce verbs in the second clause, yields meaningful data regarding pronoun interpretation. Thus, this study corroborates, using a novel task, previous findings that verb cues encountered before the pronoun are utilized for pronoun resolution. 2.8.3. Object shift in IC bias and subjectivity In Experiment 2 and Experiment 3, overall IC biases are shifted in the direction of the preceding object. The mean proportion of the object interpretations from Experiment 2 and Experiment 3 are higher (Obj-bias: M=78.1% (ES verbs); Obj-bias M=78.4% (AE verbs)) than the mean IC bias of the IC2 verbs (Obj-bias: M=77.2% (ES verbs); M=72.1% (AE verbs)) based on the norms collected by Hartshorne & Snedeker (2013) and Ferstl et al. (2011). On the contrary, the mean proportion of the subject interpretations from Experiment 2 and Experiment 3 are lower (Sbj-bias: M=48.2% (SE verbs); M=24.4% (AP verbs)) than the mean IC bias of the IC1 verbs (Sbj-bias: 76 M=67.4% (SE verbs); M=67.7% (AP verbs)) from the norms. I assume that the object-shift in IC bias results might be a task-based effect, mostly driven by subjectivity of because-clause and the nonce verbs used in it. Let us first consider the subjectivity of because-clause. It has been argued that there are two types of because-clause depending on the degree of the involvement of a thinking subject’s opinion/belief (e.g., speaker, author), which is called subjectivity (e.g., Finegan, 1995; Traxler et al., 1997; Pander Maat & Sanders, 2001; Sanders et al., 2009; Sanders & Spooren, 2015; Wei et al., 2019). The first type is often called a ‘subjective because’, as in the example (29a). Here, the semantic relation of the two clauses connected by because is involved with a high degree of subjectivity such that an opinion/belief is provided from the point of view of the thinking subject. The second type is ‘objective because’ as in the example (29b). This involves with a very low degree of subjectivity such that the cause is located in the real world, describing a situation of outside world (see Wei et al., 2019 for more discussion). (29) a. Heidi could imagine and create things because she won first prize at the art show. b. Heidi felt very proud and happy because she won first prize at the art show. (Traxler et al., 1997, p.483) Applying these ideas to my experiment, it’s interesting to note that interpreting the subject- position pronoun as referring to the subject antecedent (e.g., Henry in (30a)) is more likely to trigger the ‘subjective because’ than interpreting it as referring to the object antecedent (e.g., Kevin in (30b)). When the subject pronoun is co-referential with the preceding subject as in (30a), the because-clause is more likely to be interpreted as a redescription of the previous event (e.g., the 77 daxing event itself constitutes the hating event). On the contrary, when the pronoun is co- referential with the preceding object as in (30a), the because-clause is more likely to be interpreted as the cause of the previous event (e.g., the daxing event is the cause of the hating event such as hitting). (30) a. Henryi hated (IC2) Kevinj because hei daxed (hated>hit) Tom. [Subjective because] b. Henryi hated (IC2) Kevinj because hej daxed (hated<hit) Tom. [Objective because] Recall that when the clause with an IC verb is followed by the because-clause, an explanation relation (effect-cause sentence) is most likely to be inferred to hold between the two clauses (e.g., Bott & Solstad, 2014). Based on the subjectivity type, the typical explanation relation (effect-cause relation) can be established under the objective because type. Moreover, it has been reported in the previous studies (e.g., Traxler et al., 1997; Traxler et al., 1998) that sentences with the ‘subjective because’ have greater difficulties in processing than those with the ‘objective because’ since the ‘subjective because’ needs “an inference to establish the nature of the causal consequence” (Traxler et al., 1997:482). Thus, for the causal interpretation under the IC context and for the processing ease, this subjectivity factor is assumed to exert an influence on reference resolution in such a way that the subject pronoun is co-referenced with the preceding object, resulting in object-shift in IC bias. Let us take a closer look at how the subjectivity factor affects pronoun resolution with each IC verb type. In Experiments 2 and 3, the mean proportion of object antecedent interpretations was higher than the norms in the IC2 verb condition while it was lower than the norms with the IC1 verb condition. I assume that the subjectivity factor influences how strongly the IC bias effect 78 occurs in a preferred direction (i.e., the subject preference of IC1 verbs vs. the object preference of IC2 verbs). In the context of the IC2 verb, the subjectivity factor (objective because) would exert its influence in the same direction as the IC2 verb bias, showing a stronger preference for the object antecedent than the subject antecedent. However, in the IC1 context, the subjectivity factor (objective because) is assumed to exert its influence in the opposite direction. Due to the cause/reason interpretation or processing difficulties, the pronoun would prefer to be co-referenced with the preceding object, while the IC1 verb is biased to the preceding subject. As only the IC1 context shows a conflict in referential preference, it seems that the subjectivity effects provide an explanation for why the object antecedent preference in the IC2 context is stronger than the subject antecedent preference in the IC1 context. In terms of my current experiments, it is worth emphasizing that the difference in IC bias strength resulting from the subjectivity does not seem to modulate the effects of referential structure, which is the main aim of my dissertation research. While the current chapter focuses on how the interpretation of pronouns differs between the One-Pronoun structure and the Two- Pronoun structure, the subjectivity factor exerts its influence (i.e., object preference) in the same way on both of the structure type. Thus, the potential effects of subjectivity are orthogonal to my key conclusion that pronoun resolution is guided by referential structure due to presence or absence of the interactions between the anaphoric dependencies from different pronouns. 2.8.4. Thematic role effects In this chapter, I found an unexpected interaction between verb type and referential structure with SE (IC1) / ES (IC2) verbs (Experiments 1,2), but not with AP (IC1) / AE (IC2) verbs (Experiment 3). While sentences with SE (IC1) verbs were less susceptible to referential structure effects than 79 those with ES (IC2) verbs, sentences with AP (IC1) and AE (IC2) verbs were equally sensitive to referential structure effects. Although this asymmetry is not central to or problematic for our claims (given that both SE and ES verbs still showed referential structure effects), let us nevertheless briefly consider what could be causing it. I suggest that this asymmetry may stem from the thematic roles associated with the different verb classes. First, let us consider how this might yield relatively weaker referential structure effects with ES verbs than SE verbs. Experiencer is known to be inherently more topical/salient than Stimulus, because Experiencer is prototypically animate and highly sentient, while Stimulus is not prototypically animate or sentient (e.g., Verhoeven, 2009, 2014). It has been argued that that animate referents are more topical than inanimate referents (e.g., Givon,1983; Taylor, 2000) and sentient referents are more topical than non-sentient referents (e.g., Fedriani, 2014). Recall that I propose the two discourse-related Dependence Views on referential structure (Syntax-Discourse/Discourse-level Dependence View). As predicted by Centering Theory (CT), if referential structure effects are related to the discourse-level co-referential mechanisms, then it is expected that re-mentioning the higher-salient referent (preceding subject) in the less salient position (object-position pronoun) yields a less coherent transition than re-mentioning it in the salient position (subject-position pronoun) in the Two-Pronoun configuration (i.e., RETAIN vs. CONTINUE). Considering that Experiencer is more topical than Stimulus, there could be differences in the consequences of demoting the subject to the less salient position in clauses with SE verbs (Stimulus-subject) vs. in clauses with ES verbs (Experiencer-subject). Specifically, demoting a topical/salient Experiencer referent could yield an even less coherent transition than demoting a 80 less topical Stimulus referent. According to CT, “measuring coherence in transition is based on an estimate of a hearer’s inference load, relative to other choices a speaker had as to how to realize the same propositional content” (Walker et al., 1998:5). Since RETAIN corresponds to a situation in which the speaker signals a topic shift in the next utterance by realizing the current center in a lower salient position, it would be harder to infer for the hearer to demote the more topical entity compared to demoting the less topical entity. Thus, this would result in ES verbs exhibiting greater sensitivity to referential structure effects than SE verbs – which is indeed what we found. In contrast, the thematic role of the high-salience referent does not differ between AP (IC1) verbs and AE (IC2) verbs – the subject is an Agent in both cases. Accordingly, there are no differences in topicality/salience of the arguments of AP (IC1) and AE (IC2) verbs, and thus I do not expect any asymmetry in the strength of referential structure effects. This matches what I found in Experiment 3. Thus, I suggest that the verb type x referential structure interaction may stem from meaningful differences in thematic roles. This fits with prior claims that the salience of discourse entities is influenced by thematic roles and topicality as well as by grammatical roles. 2.9. Conclusion Taken together, the three experiments reported here provide novel evidence for the effects of referential structure, namely that whether subject-position third-person pronouns are interpreted as referring to the preceding subject or object depends on whether or not the clause contains another third-person pronoun referring to the other referent in the preceding clause. Thus, the current results show that pronoun interpretation is sensitive to the referential properties of the entire pronoun-containing clause, supporting the Dependence View: Multiple-pronoun resolution process is fundamentally different from single-pronoun resolution due to the interactions between 81 the anaphoric dependencies of the two different pronouns. Since Chapter 1 aims for the initial step of testing the Dependence vs. the Independence View, the three types of the Dependence View – the Syntax-level, the Discourse-level, and the Syntax-Discourse Dependence View – predicted the same patterns of the referential effects. However, what still remains an open question concerns the levels of representation that the interactions between the dependencies take place. In the next chapter, I further investigate this issue by detangling the different Dependence Views from each other. 82 Chapter 3: Source of the dependency interactions in multiple pronoun resolution 3.1. Introduction The experiments reported in Chapter 2 revealed that referential structure of the pronoun-containing clause guide pronoun interpretation: how subject-position third-person pronouns are interpreted depends on whether or not the clause contains another third-person pronoun referring to the other referent in the preceding clause. To be specific, Experiments 1 to 3 found that the subject-position pronoun is more likely to be interpreted as referring back to the preceding object in configurations with one pronoun (e.g., she verbed, she verbed Kate) than in configurations with two pronouns (e.g., she verbed her). These results provide initial evidence against the Independence View on referential structure but supporting evidence for the class of Dependence View. These results suggest that the process of reference resolution of multiple pronouns is different from that of a single pronoun in that multiple pronoun resolution is additionally guided by interactions between the anaphoric dependencies initiated from the different pronouns. Recall that Chapter 2 introduced three types of the Dependence View (Syntax-level, Discourse-level, and Syntax-Discourse Dependence View). These make the same descriptive predictions but make different claims about the underlying mechanisms. More specifically, they make different claims about how the interaction of coreferential operations is affected by factors at different levels of representation (syntax and/or discourse): Referential structure effects are assumed to arise in the course of searching for morpho-syntactically matching antecedents (i.e., cue-based Parallel Function Strategy) according to the Syntax-level Dependence View. In contrast, under the Discourse-level Dependence View, referential structure effects arise as a consequence of maximizing the coherence of transitions between utterances in discourse (i.e., 83 Centering Theory (CT)). Finally, the Syntax-Discourse Dependence View hypothesizes that both syntactic and discourse factors affect the interaction of coreferential operations. Since Chapter 2 focused primarily on the presence or absence of overall effects of referential structure (i.e., compared the Independence View to the class of Dependence Views), it is still unclear what factors at the different levels of representation contribute to referential structure effects. Thus, Chapter 3 aims to shed light on which of the three types of Dependence View best captures the nature of referential structure effects. To answer this question, the three types of the Dependence View are tested by disentangling discourse-coherence related factors from syntactic parallelism related factors. Experiment 4 manipulates topicality in the pre-final clause by re-mentioning (pronominalizing) the preceding referent of a context question in either the subject position (31a: Subject-Topic condition) or the object position (31b: Object-Topic condition). (31) a. One person asks: What about Henry? Another person answers: He cheated Kevin because he daxed {Tom/him}.[Subject-Topic] b. One person asks: What about Kevin? Another person answers: Henry cheated him because he daxed {Tom/him}.[Object-Topic] In the Subject-Topic condition (31a), the topic of the pre-final clause Henry is realized in the subject position by being pronominalized as he; on the other hand, in the Object-Topic condition (31b), the topic Kevin is realized in the object position by being pronominalized as him. By detangling information status (topic or non-topic) from a syntactic function (subject or object), this topicality manipulation allows us to compare the three types of Dependence View hypotheses. Moreover, just as in Experiments 1 to 3, referential structure is manipulated by comparing 84 clauses with two pronouns (Two-Pronoun structure) to those with one pronoun in the subject position and a proper name in the object position (Pronoun+Name structure). 3.2. Experiment 4 3.2.1. Participants Sixty-five U.S.-born native English-speaking adults participated and were paid $4. Four participants were excluded from data analysis for being non-native speakers of English or non- U.S.-born speakers. Two participants were excluded for giving random answers throughout the task. Three participants were excluded from data analysis because they failed to pass the minimum catch trial threshold. (This experiment had six catch trials; participants had to get at least eight correct). Thus, 56 participants were included in the final analysis. 3.2.2. Materials and design Experiment 4 employs a 2 × 2 design and manipulates referential structure in the final clause and topicality in the pre-final clause. This is illustrated in Table 11. Referential structure is manipulated, just as in Experiment 2 and Experiment 3, by comparing clause with two pronouns in the subject and the object position (Two-Pronoun condition) to clauses with one pronoun in the subject position and a new name in the object position (Pronoun+Name condition). 85 Topic- hood Referential structure Examples Subject -Topic Two-Pronoun One person asks: What about Henry? Another person answers: He cheated Kevin because he daxed him Pronoun +Name One person asks: What about Henry? Another person answers: He cheated Kevin because he daxed Tom. Object -Topic Two-Pronoun One person asks: What about Kevin? Another person answers: Henry cheated him because he daxed him. Pronoun +Name One person asks: What about Kevin? Another person answers: Henry cheated him because he daxed Tom. Table 11. A set of example stimuli in Experiment 4 Crucially, in order to test the source of referential structure effects, Experiment 4 manipulates topicality in the pre-final clause, which is preceded by a context question What about X?. I used an about-phrase in the context frame What about X? to manipulate topicality without being asymmetrically affected by any potential event or aspectual information. It also has been found in previous studies that the information structural status of a referent can be successfully manipulated by about phrases (e.g., Burmester et al., 2014; Burmester et al., 2018; Cowles, 2007; Cowles & Ferreira, 2012; Lee, 2022). The referent of the context question What about X is re- mentioned (pronominalized) in either the subject position (Subject-Topic condition) or the object position (Object-Topic condition 14 ) of the pre-final clause. This manipulation can impact topicality because according to CT (Centering Theory), either the subject or the object becomes the topic (Cb, a backward-looking center in CT terms) by being pronominalized in different positions. Thus, the topicality manipulation in Table 11 means that the topical element and the subject-position 14 The previous studies have shown that a topic can be successfully referred to by the object-pronoun in many languages, including English (e.g., English: Chien and Wexler 1990, McKee, 1992; Dutch: Hendriks et al., 2011, Rij et al., 2016). 86 element can be distinct and allows us to compare the three types of Dependence View hypothesis. Moreover, just as in the previous experiments, referential structure effects are tested under the implicit causality (IC) context. However, Experiment 4 does not include an implicit causality (IC) manipulation as a series of experiments in Chapter 2, which replicated IC effects with a novel picture writing task. The pre-final clause contains only the object-biased IC2 (Agent-Evocator) verbs that were used in Experiment 3 (mean object bias=72.1%, SD=5.53). Moreover, referential structure effects are tested under explanation relations (signaled by ‘because’) with nonce verbs to minimize semantic variability in the final clause. This allows us to look at syntactic parallelism in the absence of semantic parallelism, in contrast to previous work on parallelism which typically tested sentences involving both syntactic and semantic parallelism at the same time (e.g., Smyth, 1994; Chambers & Smyth, 1998). The current study included 24 targets and 36 fillers. As in Experiment 1 to Experiment 3, the fillers consist of three types such as riddle type, ambiguous relativized sentences, and negative stripping sentences. However, in Experiment 4, each filler sentence starts with a context clause (e.g., what happened in/at/on … ?, Is it true that … ?, and Tell me about X. ) like the target sentences. The filler items were pseudo-randomly interspersed between the target items, and they were presented using a Latin-Square design. The complete list of target sentences can be found in Appendix A. 3.2.3. Procedure The task and the procedure were the same as in Experiments 1 to 3 except that the participants were instructed that the experiment consisted of short conversations between pairs of people. Examples of the displays are in Figure 11. 87 Figure 11a. Example Subject-Topic item from Experiment 4 Figure 11b. Example Object-Topic item from Experiment 4 3.2.4. Data processing and analysis Experiment 4 was analyzed like Experiments 1 to 3. 0.3% of the responses were excluded from the entire data for the subject-position or object-position pronoun not being interpreted as referring to either the preceding subject or object. Thus, 99.7% of the data were submitted for final statistical analysis. For the statistical analyses, a generalized linear mixed logit model (glmer) was used with the proportion of object interpretations as the dependent variable, and referential structure type (contrast-coded, Pronoun+Name = 0.5, Two-Pronoun = −0.5), topicality type (contrast-coded, Sbj- topic= 0.5, Obj-topic = −0.5), and the referential structure type x topicality type interaction as fixed effects. Models were estimated using the same R packages as in Experiment 1 to Experiment 3. Random effects were determined as in Experiment 1 to Experiment 3 except for by-subject and by-item random slopes being entered for the effects of topicality type instead of IC verb type. 3.2.5. Predictions Experiment 4 aims to replicate referential structure effects found in Experiment 1 to Experiment 3 and to further investigate what factors at the different levels of representation (syntax and/or 88 discourse) contribute to the interactions between anaphoric dependency-building operations of different pronouns. If referential structure effects are replicated in Experiment 4, we should see differences between the Two-Pronoun and Pronoun+Name conditions in interpreting the following pronouns, just as in Experiments 1 to 3. However, the patterns of pronoun resolution by the referential structure type are predicted to be different depending on the level of representation that the interactions between pronominal dependencies are involved with (i.e., syntactic factor and/or discourse factor) due to the topicality manipulation. thus, now the three subtypes of the Dependence View make distinct predictions: First, the Syntax-level Dependence View predicts that the effects of referential structure will not be affected by topicality because this view hypothesizes that only morpho-syntactic factors (e.g., a gender, number, person and grammatical role) affect the interaction between the anaphoric dependencies of the subject and the object pronoun in the Two-Pronoun structure. According to the Cue-based PFS (Parallel Function Strategy), it is predicted that both in the Subject-Topic and the Object-Topic condition, the object antecedent of the pre-final clause will be more likely to be retrieved for the following subject-position pronoun in the Pronoun+Name structure, relative to the Two-Pronoun structure. This is because the interactions between the dual dependencies initiated from the two different pronouns can be realized in the form of competition, where a stronger dependency (parallel grammatical roles) suppresses the other dependency that is linked to the same antecedent (non-parallel grammatical roles). Thus, this competition leads to differences in pronouns resolution by referential structure because it only exists in the process of Two-Pronoun resolution, not in single pronoun resolution. However, there should be no difference by topicality type since the strength of the pronominal dependency is determined by syntactic features, but not by topicality-related features. 89 Second, the Discourse-level Dependence View predicts that referential structure effect will be affected by the topicality of antecedents because this view posits that the factors at the level of discourse affect how the anaphoric dependencies of the two different pronouns are resolved. Recall that Centering Theory (CT) views pronoun resolution as a part of holistic process that forms information structural coherence, which is achieved by the most salient discourse entity being held constant across utterances. Among three main types of coherence transition – CONTINUE is the most coherent, RETAIN is the second most coherent and SOFT/ROUGH-SHIFT is the least coherent – pronoun interpretation that maximizes coherence is preferred over other interpretations. According to CT, it is predicted that if the subject of the pre-final clause is a topic and pronominalized (Subject-Topic condition), the following subject-position pronoun will be more likely to be interpreted as referring to the object antecedent in the Two-Pronoun condition than in the Pronoun+Name condition. On the other hand, if the object of the pre-final clause is a topic and pronominalized (Object-Topic condition), it is predicted that the following subject pronoun will be more likely to refer to the object antecedent in the Pronoun+Name condition than in the Two- Pronoun condition. The different predictions come from the degree of transition coherence. As illustrated in Table 12, when the final subject-position pronoun he refers to the ‘non-topical’ preceding object 15 (Subject-Topic condition), the topic is shifted to Kevin without another pronoun (a: SOFT-SHIFT), but it still keeps the same topic Henry with the presence of the object pronoun (b: RETAIN). On the flip side, when the final subject pronoun refers to the ‘topical’ preceding object (Object-Topic 15 As the other interpretations that the subject antecedent (e.g., Henry) is referred to yield the same type of transition coherence without regard to the referential structure type (CONTINUE in the Subject-Topic condition and SHIFT in the Object-Topic condition), the differences by the referential type stem from the object-antecedent interpretations. 90 condition), the same topic continues as Kevin without another pronoun (c: CONTINUE), but it is shifted to Henry with another pronoun in the object position (d: ROUGH-SHIFT). Thus, the transition coherence in discourse is influenced by the topicality of the re-mentioned entity. CT Transition Topic Ref. structure The final subject-position pronoun referring to the preceding object SOFT-SHIFT Sbj- topic Pronoun+Name a. What about Henry1? He1 cheated Kevin2 because he2 daxed Tom. RETAIN Two-Pronoun b. What about Henry1? He1 cheated Kevin2 because he2 daxed him1. CONTINUE Obj- topic Pronoun+Name c. What about Kevin2? Henry1 cheated him2 because he2 daxed Tom. ROUGH- SHIFT Two-Pronoun d. What about Kevin2? Henry1 cheated him2 because he2 daxed him1. Table 12. Transition type predicted by CT in Experiment 4 Lastly, the Syntax-Discourse Dependence View hypothesizes that how we resolve anaphoric dependencies from one of the two pronouns affects dependency formation of the other pronoun, and both factors at the level of syntax and discourse guide the process of dependency construction. Thus, this view predicts that the overall referential structure effects driven by either mechanism for anaphoric dependency resolution (the Cue-based PFS or CT) will be modulated by the other. Specifically, it is predicted that the difference in object antecedent interpretation between the Pronoun+Name structure and the Two-Pronoun structure will be smaller in the Subject-Topic condition than in the Object-Topic condition because the predictions between the Syntax-level Dependence View (i.e., the Cue-based PFS) and Discourse-level Dependence View (i.e., CT) are pitted against each other only in the Subject-Topic condition but not in the Object-Topic condition. In the Subject-Topic condition, the Syntax-level Dependence View predicts that the object 91 antecedent bias will be weaker in the Two-Pronoun condition compared to the Pronoun+Name condition in resolving the subject-position pronoun; on the other hand, the Discourse-level Dependence View predicts that the object antecedent bias will be stronger in the Two-Pronoun condition. However, in the Object-Topic condition, both of the Dependence Views predict that the object antecedent bias will be weaker in the Two-Pronoun condition than in the Pronoun+Name condition. 3.2.6. Results Figure 12 shows the proportion of trials where the subject-position pronoun in the final clause is interpreted as referring to the object of the pre-final clause. Figure 12. Experiment 4: The proportion of trials where the subject-position pronoun refers to the preceding object. (error bars represent ±1 S.E.) Three numerical patterns are clearly visible in Figure 12. First, the results replicate the effects of referential structure found in Experiments 1 to Experiment 3 – the percentage of object interpretations was markedly reduced in the Two-Pronoun structure compared to the Pronoun+Name structure. Specifically, the same pattern of referential structure effects was found 92 both in the Subject-Topic condition (Two-Pronoun = 72.6%, Pronoun+Name = 83.0%) and the Object-Topic condition (Two-Pronoun = 77.9%, Pronoun+Name = 92.5%). Second, the difference between the Pronoun+Name and the Two-Pronoun structure is smaller in the Subject-Topic condition (difference = 10.4%) than in the Object-Topic condition (difference = 14.6%). Crucially, the two patterns are found in the direction predicted by the Syntax-Discourse Dependence View. This shows that the syntax-level cue-based parallelism underlies the overall effects of referential structure, which is also modulated by topicality. Lastly, Figure 12 shows that topicality affects overall pronoun interpretation – the object-position topic in the pre-final clause markedly increases the proportion of object interpretations compared to the subject-position, both in the Two-Pronoun structure and the Pronoun+Name structure. To assess these effects statistically, I fit a generalized linear mixed logit model. Statistical analyses are reported in Tables 13 and Table 14. As can be seen in Table 13, there are main effects of referential structure type and topicality type as well as a Referential structure type × Topicality type interaction. To see if there are differential effects of referential structure in Subject-Topic and Object-Topic conditions, planned comparisons were conducted. As shown in Table 14, an effect of referential structure was found both in the Subject-Topic and Object-Topic conditions. Thus, the statistical results confirm that the both the syntax-level cue-based parallelism and the discourse- level coherence mechanism (CT) guide pronoun resolution by contributing to the interactions between anaphoric dependencies of the two different pronouns (Syntax-Discourse Dependence View), while the syntax-level parallelism exerts more influence on reference resolution than the discourse-level coherence. 93 Fixed effect Estimate Std. Error z value Pr(>|z|) (Intercept) 2.3594 0.2701 8.734 < 2e-16 *** Ref.st type 1.2826 0.3379 3.796 0.000147 *** Topicality type 0.8235 0.1824 4.515 6.33e-06 *** Ref.st type:Topicality 0.8622 0.3645 2.366 0.017998 * Table 13. Experiment 4: results of the glmer model Topicality Fixed effect Estimate Std. Error z value Pr(>|z|) Sbj Topics (Intercept) 1.9446 0.2881 6.750 1.48e-11 *** Ref.st type 0.8658 0.2268 3.817 0.000135 *** Obj Topics (Intercept) 4.4644 0.8918 5.006 5.55e-07 *** Ref.st type 5.0978 1.7035 2.993 0.00277 ** Table 14. Experiment 4: planned comparisons 3.3. Discussion Experiment 4 investigated what factors at the different levels of representation involve the interactions between anaphoric dependency-building operations (e.g., subject pronoun vs. object pronoun dependency), which result in referential structure effects in pronoun interpretation. The central aim of this study was to investigate which of the three types of Dependence View (Syntax- level, Discourse-level, and Syntax-Discourse Dependence View) best fits the data. More generally, this will elucidate whether the source of the interactions between the anaphoric dependencies lies on the level of syntax, discourse or both. More specifically, Experiment 4 investigate whether referential structure effects arise in the course of searching for syntactically matching antecedents (i.e., Syntax-level Dependence View) or in the course of maximizing the transition coherence between utterances (i.e., Discourse-level Dependence View) or both (i.e., Syntax-Discourse 94 Dependence View). The findings from Experiment 4 show that referential structure effects are driven by both syntactic and discourse-level factors, supporting the Syntax-Discourse Dependence View. 3.3.1. Source of referential structure effects Referential structure effects were tested by disentangling discourse-coherence related factors from syntactic parallelism related factors. Experiment 4 manipulated topicality in the pre-final clause, by re-mentioning (pronominalizing) the preceding referent of a context question in either the subject position (Subject-Topic condition) or the object position (Object-Topic condition). This manipulation allowed us to disentangle a discourse-related factor (i.e., topical element) from a syntax-related factor (i.e., subject-position element) and elicited the three different predictions from each type of the Dependence View. Before turning to the results, let us revisit the predictions: The Syntax-level Dependence View predicts that object antecedent preference will be stronger in the Pronoun+Name structure than the Two-Pronoun structure both in the Subject-Topic and the Object-Topic condition. Under this view, referential structure effects should be insensitive to topicality because the strength of the pronominal dependency is determined by morpho- syntactic features, not by topicality-related features. On the contrary, the Discourse-level Dependence View predicts that the object antecedent preference will be stronger in Pronoun+Name than Two-Pronoun structures in the Object-Topic condition, whereas it will be weaker in Pronoun+Name than Two-Pronoun structure in the Subject- Topic condition. Under this view, referential structure effect will be affected by topicality because anaphoric dependencies of the two different pronouns interact in the process of establishing transition coherence in discourse and the topical elements occur in different syntactic positions by 95 the topicality type. Lastly, the Syntax-Discourse Dependence View posits that the two different dependencies are influenced by factors at the level of syntax and discourse. This view predicts that referential structure effects will be weaker in the Subject-Topic condition compared to the Object-Topic condition. This is because patterns of pronoun resolution resulted from the Syntax-level and Discourse-level Dependence View are pitted against each other in the Subject-Topic condition but not in the Object-Topic condition. Now, let us revisit the results. Experiment 4 shows that when patterns of pronoun interpretation were compared between configurations where the final clause only mentions one of the two preceding entities (Pronoun+Name condition) and configurations where the pronoun- containing clause mentions both of the preceding entities (Two-Pronoun condition), I find clear effects of referential structure. Specifically, the results show that these effects are driven by both syntactic and discourse-level factors, supporting the Syntax-Discourse Dependence View. These results suggests that both syntax-level cue-based parallelism and a discourse-level coherence mechanism (i.e., CT) guide pronoun resolution by contributing to the interactions between anaphoric dependencies of the two different pronouns. Taken together, the results of Experiment 4 break new ground on pronoun resolution by finding the effects of referential structure, and showing that it is driven by the dependency interactions that take place both at the level of syntax and discourse. 3.3.2. Syntax-level parallelism over discourse-level coherence Experiment 4 provides evidence that referential structure effects resulted from syntax-level parallelism being modulated by topicality. The overall patterns of pronoun interpretation by 96 referential structure type (Pronoun+Name vs. Two-Pronoun structure) are guided by the syntax- level cue-based parallelism – stronger object-antecedent preference in the Pronoun+Name structure than in Two-Pronoun structure – while the strength of referential structure effects was affected by topicality type. These findings indicate the syntax-level parallelism exerts more influence on reference resolution than discourse-level coherence. Similar patterns were reported by Chambers and Smyth (1998), who tested whether single pronoun resolution is guided by the syntactic parallelism and/or the centering model of discourse coherence. In their Experiment 1, they manipulate topicality by realizing a topic character in either subject or non-subject position and manipulate a grammatical role of a pronoun by containing either a subject or non-subject pronoun in the following sentence, as shown in (32). Overall, they found the syntactic parallelism effect dominated pronoun interpretation, but the parallelism effect was reduced when the non-parallel antecedent was a previously mentioned topic. (32) a. Leonard and his friends were sharing their lunches. Leonard handed Michael a sandwich. Then he passed Carla an apple. [Subject-Topic] b. Leonard and his friends were sharing their lunches. Michael handed Leonard a sandwich. Then Carla passed him an apple. [Non-subject-Topic] (Chambers and Smyth, 1998, p.597) What is common between Experiment 4 and Chambers and Smyth (1998) is that the pronoun-containing sentence has a parallel syntactic structure as its preceding sentence (e.g., transitive or ditransitive structures), and thus the pronoun shares the identical grammatical role with its potential antecedent. As mentioned in Chapter 2, other work shows that a lack of full structural parallelism between the two sentences decreases parallel interpretation of the pronoun 97 (e.g., Smyth, 1994; Stevenson et al., 1995). Given these previous findings, I assume that the stronger effects of the syntax-level parallelism found in Experiment 4 result from the specific parallel contexts used in the current experiments. It is well-known that various factors are weighted differently in pronoun resolution depending on the context. Thus, I expect that various patterns of referential structure effects would emerge depending on the contextual circumstances where different factors operate in with a different weight. In different context, the effects of the discourse factors could be stronger than the syntactic parallelism with regard to referential structure effects. In addition, it should be noted that the current experiment found the parallelism effects in single or multiple pronoun resolution in the contexts that are syntactically parallel but semantically non-parallel (i.e., do not have resemblance relations with similar verbs). This indicates that parallelism effects on pronoun interpretation can occur independently of the semantic parallelism. 3.3.3. Topicality effects In Experiment 4, I observed that topicality of potential antecedents guides overall pronoun interpretation – the proportion of object antecedent interpretations was significantly higher in the Object-Topic condition compared to the Subject-Topic condition. Centering Theory (CT) can explain these results because transitions are more coherent when pronouns refer to topics vs non- topics. Specifically, when a subject-position pronoun refers back to a preceding object, this is a more coherent transition when that object is a topic than when it is not a topic (Object-Topic condition: Pronoun+Name = CONTINUE, Two-Pronoun = ROUGH-SHIFT vs. Subject-Topic condition: Pronoun+Name = SOFT-SHIFT, Two-Pronoun = RETAIN). Moreover, these results are in line with the previous work on topicality effects in reference resolution (e.g., Ariel, 1990; Arnold, 1998, 1999; Colonna et al., 2010; Cowles et al., 2007; Ellert, 98 2013; Givón, 1983). A sizeable body of psycholinguistic research has suggested that topic referents occupy a cognitively salient status in our mental models of discourse, which makes them more accessible and preferred by a subsequent pronoun, because the salient referents tend to be subsequently referred back to by means of reduced anaphoric expressions like the pronouns. It has been attested by many previous studies that pronouns prefer antecedents realized as the topic element than the non-topic element (e.g., Arnold, 1998, 1999; Colonna et al., 2010; Cowles et al., 2007; Ellert, 2013; Kaiser, 2011). Thus, the topicality effects found in Experiment 4 corroborates previous findings that topicality of the potential antecedents guides pronoun resolution. 3.4. Conclusion Experiment 4 further investigates the effects of referential structure and provides novel evidence that factors at different levels of representation contribute to the effects of referential structure by focusing on three subtypes of the Dependence View (Syntax-level, Discourse-level, and Syntax- Discourse Dependence View). The results show that referential structure effects result from syntax- level parallelism being modulated by topicality, supporting the Syntax-Discourse Dependence View. These findings indicate that both syntax-level cue-based parallelism and a discourse-level coherence mechanism (Centering Theory) guide pronoun resolution by contributing to the interactions between anaphoric dependencies of the two different pronouns but the syntax-level parallelism exerts more influence on reference resolution than discourse-level coherence. However, it is important to note that I do not expect that referential structure effects would surface in the same way in all contextual circumstances. Although the syntax-level parallelism has a stronger influence on reference resolution than discourse-level coherence in the structurally parallel context of Experiment 4, I assume that it would surface in different ways depending on 99 context and the pronouns appear in as well as linguistic forms of the pronouns (overt pronouns vs. null pronouns). Since different factors weigh in differently from one context to another during pronoun resolution, I assume that various patterns of referential structure effects would emerge by different factors operating in with a different weight. 100 Chapter 4: The effects of referential structure with null pronouns 4.1. Introduction So far, this dissertation has focused on the nature of multiple pronoun resolution and how it is different from single pronoun resolution. A series of experiments in the previous chapters show that the differences in referential structure (One-pronoun vs. Two-Pronoun configurations) guide pronoun resolution. The results suggest that interactions between the dual pronominal dependencies in the Two-Pronoun configuration play a key role in eliciting different patterns of pronoun interpretation as a function of referential structure type. However, the effects of referential structure found in the previous chapters are based on one type of anaphoric form – namely, overt pronouns in English. It is important to broaden the scope of referential structure effects to other types of anaphoric forms such as null pronouns, because it is widely agreed that different anaphoric forms can show different interpretational patterns (e.g., Alonso-Ovalle et al, 2002; Frana, 2008; Carminati, 2002; Filiaci, Sorace & Carreiras, 2013; Kaiser & Trueswell, 2008; Kwon & Sturt, 2013). According to Kaiser and Trueswell’s (2008) Form-Specific Account, anaphoric forms can differ in their sensitivity to different properties of potential antecedents, and the multiple factors which guide reference resolution can carry different weight for different types of anaphoric form. Thus, this chapter aims to investigate whether and how referential structure effects are influenced by different anaphoric forms, by means of null pronouns (pro) in Korean. As regards the nature of null arguments in Korean, it is an open question whether they are derived as a result of ellipsis (e.g., the DP-ellipsis / the VP-ellipsis) or null pronouns (pro) (e.g., Ahn & Cho, 2010, 2020; Huang, 1984; Kim, 1999; Moon, 2010; Park, 1994; Um, 2011); research 101 on Korean has not reached a consensus as to the status of null arguments because it seems that no one approach can fully capture their various properties (e.g. the interpretational properties of null- argument sentences, such as strict, sloppy, and sloppy-like interpretation, see Ahn & Cho, 2010, 2020 for more discussion). In this dissertation, I do not provide an answer to this debate, because I limit my inquiry to how the interaction of co-referential operations can yield referential structure effects and, moreover, the two basic mechanisms that contribute to referential structure effects (i.e., the cue-based Parallel Function Strategy (PFS) and Centering Theory (CT)) can both be explored independent of the status of the null arguments. For the Cue-based PFS, which makes use of a feature-matching process, a retrieval trigger does not necessarily need to be a pronoun, because a retrieval process is initiated whenever a representation that is currently out of focus of attention is required for interpretation (e.g., Bhatia & Dillon, 2022; Dillon, 2011; Hammerly, et al., 2019; Lewis & Vasishth, 2005; McElree, 2000; Van Dyke & Johns, 2012; Van Dyke & McElree, 2011; Yadav, et al., 2022). For Centering Theory, any elements that have semantic representation can be regarded as realization of a discourse entity – not only null-pronouns but also elided elements (e.g., Walker et al.,1998; Grosz et al., 1995). Thus, the question of whether or not null arguments in Korean are null pronouns does not crucially impact my aims and conclusions. Although the status of null arguments does not impact the hypotheses of this dissertation, for ease of exposition I will call Korean null arguments null pronouns (pro). 4.2. Relationship between anaphoric forms and topicality It is well-known that the type of anaphoric form is correlated with the referent’s degree of accessibility/topicality in discourse – the more reduced the anaphoric form is, the more topical its 102 referent is (e.g., Ariel 1990, 1994; Chafe, 1994; Givon, 1983; Gundel et al., 1993). In other words, when a referent is highly accessible in the hearer’s mental discourse, the speaker prefers to use the reduced anaphoric forms over the fuller forms. The assumption behind this idea is that the referent which is referred back to by more reduced anaphoric forms must be highly accessible/topical in discourse in order for the hearer to identify its identity despite insufficient information from the reduced forms. For example, Givón (1983) proposed a Topic-Continuity Approach based on cross- linguistic corpus data, and the approach involves an accessibility/topicality scale that ranks different kinds of anaphoric forms according to topicality of their referents. The scale is shown in Figure 13. Most continuous/accessible topic zero anaphora unstressed/bound pronouns or grammatical agreement stressed/independent pronouns R-dislocated DEF-NPs neutral-ordered DEF-NPs L-dislocated DEF-NPs Y-moved NPs ( ‘contrastive topicalization ’) cleft/focus constructions referential indefinite NPs Figure 13. Topic-Continuity Hierarchy (Givón, 1983, p.17) In Figure 13, various anaphoric forms are ordered on this scale from null (zero) anaphora 103 (most continuous/accessible topic) to referential indefinite NPs (least continuous/accessible topic). Givón’s approach predicts that null pronouns tend to be interpreted as an indication of topic- continuity more often than (unstressed) overt pronouns. To experimentally elucidate the relationship between anaphoric forms and their referents’ level of topicality, a number of psycholinguistic studies have studied the languages that allow both overt and null pronouns, such as Italian, Spanish, and Korean (e.g., Almor et al., 2017; Alonso- Ovalle et al, 2002; Frana, 2008; Filiaci, Sorace & Carreiras, 2013, Fedele, 2016; Gelormini- Lezama, & Almor, 2011; Kwon & Sturt, 2013; Runner & Ibarra, 2016). For example, Frana (2008) tested null and overt pronouns in Italian to investigate the effects of topicality on null vs. overt pronoun interpretation. She conducted a questionnaire study with sentences where the topicality of the antecedent and pronominal form were manipulated by means of pronouns vs. names, as in (33): in one condition, the discourse topic la signora Rossi ‘Mrs. Rossi’ from the context sentence is re-mentioned in the pre-final clause of the next sentence with an object clitic pronoun (33a). Here, re-mentioning la signora Rossi ‘Mrs. Rossi’ by means of a pronoun can be taken as a signal that she is a topical referent. However, in another condition, la signora Rossi ‘Mrs. Rossi’ is mentioned in the pre-final clause of the second sentence by using a repeated name (33b) – thus, there is no linguistic cue (no pronominal form cue) signaling her status as a potentially topical referent. The final clause of the second sentence contains either a null pro or an overt pronoun lei ‘she’, as shown in (33a) and (33b). 104 (33) Context: La signora Rossi è una persona moto maleducata che non merita alcom riguard. ‘Mrs. Rossi is a very rude person that doesn’t deserve any regard.’ a. Clitic: Quando Maria la incontra per strada, {pro/lei} fa sempre finta di non vederla. ‘When Maria her-sees in the street, {pro/she} pretends not to see her.’ b. Full DP: Quando Maria incontra la signora Rossi per strada, {pro/lei} fa sempre finta di non vederla. ‘When Maria sees Mrs. Rossi in the street, {pro/she} pretends not to see her.’ After reading these kinds of two-sentence sequences, participants were asked questions probing their interpretation of the null/overt pronouns in the final clause. The results reveal that only null pronouns are sensitive to the (form-signaled) topicality status of the antecedents, but the overt pronouns are not. Overt pronouns showed a preference for the object-position antecedent la signora Rossi ‘Mrs. Rossi’ regardless of whether it was realized with a pronoun or a full name. On the other hand, null pronouns preferred the object antecedent la signora Rossi ‘Mrs. Rossi’ over the subject-position antecedent Maria only when the object antecedent was realized with a clitic pronoun, functioning as a topical antecedent (e.g., la ‘her’ referring to ‘Mrs. Rossi’ in (a)). Thus, these results suggest that null pronouns in Italian prefer prominent discourse entities (topics) more strongly than overt pronouns do. Similar results come from null pronouns in Korean. Although Kwon and Sturt (2013) did not directly compare null pro vs. overt pronouns, they investigated whether the reference of the null pronoun could be shifted from a discourse topic to a new intra-sentential antecedent based on morpho-syntactic cues, just as overt pronouns in English could (see Liversedge & Van Gompel, ms.). Kwon and Sturt (2013) conducted a self-paced reading study by manipulating presence vs. 105 absence of a context sentence (34a) and presence vs. absence of the reflexive caki ‘self’ in the Clause 1, which contains a subject-position null pronoun (34b). The null pronoun pro in Clause 1 has two potential referents: the matrix subject na ‘I’ and the embedded subject kanhosa ‘nurse’ in Clause 2, which is always identical to the discourse topic in the context sentence, if present. Crucially, since a reflexive caki ‘self’ can be only bound by a third-person singular referent, the subject pro that is binding the reflexive caki ‘self’ in the same clause cannot be co-referencial with the following matrix subject na ‘I’ due to a person mismatch. Thus, this study predicted that if the parser attempts to form a dependency with a following matrix subject (e.g., na ‘I’) even when pro has already formed a dependency with a discourse topic (e.g., kanhosa ‘nurse’), reading times (RTs ) of the caki condition will be slower than those of the No-Caki condition at the matrix subject region. (34) a. I kanhosai-nun maywu chincelhata. This nursei-top very kind [Context] b. proi enceyna {cakii /aphun} hwancatul-ul cengsengsulepkey kanhohay-se, proi always {selfi /sick} patients-acc with care took care-as [Clause 1] na j-nun ipen sisangsik-eyse kanhosa i-ka choywusu cikwonsang-ul patayahanta-ko cwucanghayssta I j-top this ceremony-at nurse i -nom best worker-award-acc be awarded-comp claimed [Clause 2] ‘(This nursei is very kind.) Because proi/j took care of {self’si /sick} patients with much care, Ij claimed that the nursei should be awarded the best employee award at this ceremony.’ (Kwon & Sturt 2013, p.381) The results show that the RTs in the caki condition were slower than those in the No-caki condition only when there was no discourse topic due to lack of the context sentence. This indicates 106 that the null pronoun only forms a dependency with the topic kanhosa ‘nurse’ in the context sentence, when it is present, and did not shift its referent to the intra-sentential antecedent na ‘I’ in Clause 2. These findings suggest that null pronouns in Korean have a stronger preference to refer to prominent discourse entities (topic) than overt pronouns in English. It should be noted that Korean, like Spanish and Italian, has overt pronouns as well as null pronouns. However, while Korean does have a system of overt pronouns similar to other null- pronoun languages (e.g., distinguished by number and person), it is well-known that this is “an incomplete and less robust system, characterized by item-specific arbitrariness in actual usage” (Han, 2004:38) and use of overt pronouns is not a wide-spread linguistic phenomenon in Korean (see Lim & Chang, 1995; Lee et al., 1997 and Kim, 1997 for more details). Especially the third person overt pronouns are hardly used in spoken discourse in Korean, as they are a recently- developed linguistic form that started being used from the early 20th century (see Park, 2014). Due to the under-development and under-use of overt pronouns in Korean, there has been little comparative research with null pronouns in Korean. Nevertheless, Kweon (2011) conducted a questionnaire study to find out how strongly null and overt pronouns prefer preceding subjects vs. objects, as in example (35), testing the assumption that there is a special association between the null pronouns and the prominent antecedent (i.e., subject-position referent). (35) Inhoi-ka Sunggij-lul chingchanhal-ttay {ku-ka/ pro} mikwuk-ey iss-e-yo. Inhoi-Nom Sunggij-Acc praise-when {he-Nom/ pro} Amerika-in be-Pres-Dec ‘When Inhoi praises Sunggij, {he/pro} is in America.' (Kweon, 2011, p.7) As expected, the results showed that Korean speakers have a stronger preference for 107 interpreting pro as referring to the preceding subject, compared to overt pronouns (see Ueno & Kehler (2016) for similar results on subject preference for null pronoun over overt pronouns in Japanese). Although prominence is defined as a syntactic notion in this study, rather than a discourse-related notion, the results are consistent with previous findings, suggesting that the more reduced the anaphoric form, the more prominent its referent becomes (e.g., Ariel 1990, 1994; Chafe, 1974; Givon, 1983; Gundel, Hedberg, & Zacharski, 1993). As a whole, work on null pronouns supports the general idea that different anaphoric forms can assign different weights to the factors that guide pronoun resolution – null-form pronouns are more sensitive to the discourse prominence of their potential referents relative to overt-form pronouns. 4.3. Korean null pronouns As discussed, Korean allows null pronouns. Null pronouns occur very frequently in various syntactic positions. In a corpus study, Kim (2000) reported that 60.1% of all subject-position elements were null and 45.6% of all object-position elements were null in spoken conversation data from seven people. Similar patterns obtain for narratives: Kim (2003) examined utterances from a corpus of nine folktales and reported that 59.9% of all subject-position elements were null pronouns and 36% of all object-position elements were null pronouns, from among the utterances where the subject and object elements were given information. Contrary to other null-pronoun languages that mark agreement between the subject and the verb with verbal inflections, such as Spanish and Italian, Korean lacks an overt verbal agreement system. However, Korean has morphological case marking on nouns, including nominative -i/-ka, accusative -ul/-lul, and dative -hanthey/-ekey. Due to the presence of the case markers on overt 108 nouns, the grammatical role of the null pronouns can be easily identified. As shown in (36), the presence of a missing argument in the object position is clear because the overt noun Jiye is marked nominative with -ka. (36) Yeri-kai nolay-lul pwulless-nuntey, Jiye-ka proobj/i pokoissess-ki ttaymwun-i-ya. Yeri-Nom song-Acc sang-Con, Jiye-Nom proobj was-looking.at-Nml reason-be-Dec ‘Yerii sang a song because Jiye was looking at proi.’ Moreover, multiple null pronouns can co-occur in the same clause, as in (37). Here, the second sentence contains both a subject pro and an object pro: the preceding subject Kim is referred to by the subject pro and the preceding object cha ‘car’ is referred to by the object pro. (37) Kim-ii ecey cha-lulj sa-ass-ta. prosbj/i onul na-hanthey proobj/j poyecwu-ess-ta. Kim-Nom yesterday car-Acc buy-Past-Dec. prosbj/i today I-Dat proobj/j show-Past-Dec. ‘Kim bought a car yesterday. prosbj/i showed proobj/j to me today.’ (Lee, 2019, p.32) It is worthy of note that some null-argument languages such as Japanese and Chinese show an asymmetry related to the grammatical role of the null pronoun – object pros cannot be co- referential with potential antecedents in the matrix clause while the subject pros can (e.g., Kuroda 1965, Huang 1984, Hasegawa 1985). 16 However, a number of previous studies suggest that this 16 According to Huang (1984), the null object is a variable that is locally A’-bound by the zero-topic; the null subject can function as the topic-bound variable most of the time, but also can function as a null pronoun when it is c-commanded by the arguments in the matrix clause. 109 does not hold for Korean: In Korean, not only null subjects but also null objects can be co- referential with matrix-clause arguments and, moreover, the co-referentiality of null objects with the matrix-clause arguments is as natural as that of the null subjects (e.g., Lee, 1987; Moon, 2010; Kim, 2006). (38) a. Johni-i [proi/j Mary-lul ttaly-ess-ta] ko sayngkakha-n-ta. [Subject pro] Johni-Nom [proi/j Mary-Acc hit-Past-Dec] Comp think-Pres-Decl ‘John thinks that pro i/j hit Mary.’ b. Johni-i [Mary-ka proi/j -lul ttaly-ess-ta] ko sayngkakha-n-ta. [Object pro] Johni-Nom [Mary-Nom proi/j hit-Past-Dec] Comp think-Pres-Decl ‘John thinks that Mary hit proi/j.’ (Lee 1987, p.270) As shown in (38), both the subject pro (38a) and the object pro (38b) can refer to the matrix subject, John, as well as someone in the discourse. This suggests that Korean null pronouns do not show a grammatical role asymmetry in terms of their interpretational properties. 4.4. Experiment 5 Experiment 5 tests whether and how referential structure effects are influenced by different anaphoric forms, by testing null pronouns (pro) in Korean. As we saw in Chapter 3, in English referential structure effects with overt pronouns appear to stem from both syntax-level parallelism and discourse-level coherence mechanism, with syntax-level factors being stronger than discourse- level factors (see Experiment 4 for more discussion). Given that (i) the multiple factors which guide reference resolution can carry different weight for different types of anaphoric forms (Kaiser 110 and Trueswell’s (2008) Form-Specific Account) and (ii) more reduced anaphoric form tend to be associated with more topical referents (e.g., Givón’s (1983) Topic-Continuity Approach), we may find that, in Korean null pronoun interpretation, discourse-level coherence might exert a stronger influence than syntax-level parallelism, which would contrast with what I found with English overt pronouns. To explore this idea, Experiment 5 tests referential structure effects on Korean null pronouns by manipulating referential structure type in the final clause and topicality type in the pre-final clause, adapting the design of Experiment 4 (the English overt pronoun study). 4.4.1. Participants Sixty-nine native adult speakers of Korean, born in South Korea, completed this study over the internet via Qualtrics (Provo, UT). Eighteen participants were compensated by electronic gift cards (each worth approximately $8) distributed via lottery system with 25% chance of winning. I excluded one participant because they provided random answers throughout the task. Thus, 68 participants were included in the final analyses. 4.4.2. Materials and design Experiment 5 manipulates (a) referential structure type (One-pro vs. Two-pro) and (b) topicality in the pre-final clause (Subject-Topic vs. Object-Topic). As illustrated in (39), the Two-pro structure has both the subject and object null pronouns in the final clause. In the One-pro structure, the final clause contains a null pronoun in the subject position and a new name in the object position. Moreover, I manipulated whether the subject (39a) or object (39b) of the pre-final clause (e.g. Eunju or Younghi ) is topical. This is done by mentioning one of these names in the lead-in 111 clause (e.g., Speaking of Eunju…). Thus, (39a) shows the Subject-Topic version and (39b) shows the Object-Topic version. (39) a. Eunju-ey-tayhay malhacamyen, proEunju Younghi-lul sokyess-nuntey, [Subject-Topic] Eunju-about speaking.of, proEunju Younghi-Acc deceived-Con, prosbj {Jieuni-lul/ proobj } cwuoyhayss-ki ttaymwun-i-ya. prosbj {Jieuni -Acc/ proobj} cwuoyed-Nml reason-be-Dec ‘Speaking of Eunju, pro deceived Younghi because pro juakyed {Jieuni / pro}.’ b.Younghi-ey-tayhay malhacamyen, Eunju-ka proYounghi sokyess-nuntey, [Object-Topic] Younghi-about speaking.of, Eunju-Nom proYounghi deceived-Con, prosbj {Jieuni-lul/ proobj } cwuoyhayss-ki ttaymwun-i-ya. prosbj {Jieuni -Acc/ proobj} cwuoyed-Nml reason-be-Dec ‘Speaking of Younghi, Eunju deceived pro because pro juakyed {Jieuni / pro}.’ To be maximally comparable to the English studies (Experiments 1 to 4), referential structure effects are tested in explanation relation contexts signaled by a connective ‘because’, which allows us to disentangle effects of semantic parallelism (e.g., resemblance relation signaled by ‘and’) from those of syntactic parallelism in pronoun resolution. Moreover, Implicit Causality (IC) verbs are used in the pre-final clause as in the Experiments 1 to 4. However, only IC1 (subject- biased) verbs in Korean are used in the pre-final clause because most IC2 (object-biased) verbs in Korean show very strong preference to the preceding object referent according to the norms collected by Park (2009) and Kim & Grüter (2018). 17 Having too big a difference in the verbs’ 17 Among all responses, the percentage of subject referent choices was less than 20% for 38 IC2 verbs out of 40 based on the 112 biases would make the English and Korean results hard to compare, so I opted not to use IC2 verbs. In order to control the strength of each verb’s IC1 bias so that it is similar to the English verbs used in the previous experiments (IC bias towards the preceding subject between 60% and 82%), eight IC1 (Agent-Patient (AP)) verbs were selected from the Korean norms by Park (2009) and Kim & Grüter (2018) (AP verbs: Sbj-bias M=69.33%, SD=7.12). Another sixteen IC1 (AP) verbs were selected from the English IC verb lists of Hartshorne & Snedeker (2013) and Ferstl et al. (2011) (AP verbs: Sbj-bias M=62.81%, SD=12.12). These English IC verbs were first translated into Korean using the NA VER English–Korean dictionary (http://dic.naver.com) and then edited by two Korean-English bilinguals to ensure that they express the most natural interpersonal IC meaning. In addition, none of the IC verbs in this experiment contain the causative suffix –key. I chose to not use any verbs with this suffix because explicit causality marking in Korean is known to induce a stronger IC bias compared to IC verbs without causality marking (e.g., Kim, 2018; Kim & Grüter, 2018). In the final clause, Korean nonce verbs are used, in order to minimize semantic variability. They are designed to be pronounceable in Korean and presented with the active-voice verb ending -ha to prevent a passive interpretation. In addition, the gender of the names used in each sentence was matched (both female or both male). Twenty-four target items and 36 filler sentences were created. The fillers were of various types, including ambiguous relative-clause structures and negative ‘stripping’ structures (e.g., X verbed Y but not Z), as in the previous English experiments (Chapters 2 and 3). The filler items were intermixed with the target items, which were presented using a Latin-Square design. The Korean norm by Kim & Grüter (2018) and also for 31 IC2 verbs out of 32 based on the Korean norm by Park (2009). 113 complete list of target sentences is in Appendix A. 4.4.3. Procedure The task and the procedure were the same as in Experiments 1 to 4, except that all materials were in Korean. Examples of the displays are in Figure 14. I used a picture-writing task, where participants saw a sentence coupled with a schematized picture depicting the underlined part of the sentence. For target items, the underlined part was always the null pronoun-containing clause. The participants’ task was to type the names of the characters into the textboxes. This reveals how they interpret the null pronoun(s). The study was conducted over the internet using Qualtrics (Provo, UT). Figure 14a. Example Subject-Topic item from Experiment 5 Figure 14b. Example Object-Topic item from Experiment 5 4.4.4. Data processing and analysis The data coding was conducted as in the same way as for Experiments 1 to 4, using a generalized linear mixed logit model (glmer). The dependent variable is the proportion of subject interpretations. I included fixed effects of referential structure (contrast-coded, One-pro = 0.5, 114 Two-pro = −0.5), topicality type (contrast-coded, Subject-Topic = -0.5, Object-Topic = 0.5), and the referential structure x topicality type interaction. In addition, random effects were determined as in Experiments 1 to 4. Of all the data, 2.6% of the responses were excluded because participants responses indicated that they failed to interpret the subject-position pro or object-position pro as referring to either the preceding subject or object. Thus, 97.4% of the data were submitted for statistical analysis. 4.4.5. Predictions Experiment 5 investigates whether patterns of multiple pronoun resolution are different depending on anaphoric forms by looking at null pronouns in Korean. Overall, referential structure effects are predicted to occur in Experiment 5 as well – in other words, Two-pro and One-pro conditions should pattern differently, echoing what was found for English (Chapters 2 and 3). Recall that in English, I argued that referential structure effects with overt pronouns stem from syntactic and discourse level factors, with syntactic factors playing a stronger role. In Experient 5, I use Korean to test whether (a) referential structure effects generalize to another language and whether (b) the relative important of syntactic vs. discourse level factors might be different between Korean and English. This idea is rooted in the observations that (i) the multiple factors which guide reference resolution can carry different weight for different types of anaphoric forms (Kaiser & Trueswell 2008) and (ii) more reduced anaphoric form tend to be associated with more topical referents (e.g., Givón’s 1983). Thus, it might be that when looking at referential structure effects with Korean null pronouns, we may find a stronger effect of discourse compared to what we found in English – or we could find the same pattern as in English, where syntactic factors matter more. In what follows, I consider both of these possibilities in more depth. 115 4.4.5.1. Syntax-over-Discourse Dependence View First, let’s consider the idea that referential structure effects are driven by the dependency interactions guided by both syntax-level cue-based parallelism and discourse-level coherence mechanisms, and that syntactic parallelism has a stronger impact than discourse coherence (as we found for English). Let’s call this the Syntax-over-Discourse Dependence View. Under this view, it is expected that the overall referential structure effects in Korean are modulated by topicality, but with syntax having a stronger effect than discourse-level topicality. In sum, for Korean, this view predicts that the effects of referential structure type will be smaller in the Subject-Topic condition than in the Object-Topic condition. More specifically, the syntax-level cue-based parallelism predicts that the object antecedent of the pre-final clause (e.g., Younghi in example (39)) will be more likely to be retrieved for the following subject-position pro in the One-pro structure, compared to the Two-pro structure. In other words, in the example (39), focusing on the null pronoun conditions only, the prediction is that participants should be more likely to interpret the null pro as referring to the preceding object (Younghi) in the One-pro condition than in the Two-pro condition. This is because, under this view, each pronoun forms a stronger dependency with the grammatically parallel antecedent (other morpho-syntactic features being equal) and the interactions between the dual dependencies that arise in the Two-pro structure trigger competition. I assume, according to the cue-based version of the PFS that I am exploring, that (a) dependencies to grammatically parallel antecedents are stronger than dependencies to grammatically non- parallel antecedents, and that (b) stronger dependencies can suppress weaker dependencies. Moreover, under this view, I do not expect to find any effect of the topicality manipulation: The referential structure effects in the Subject-Topic and the Object-Topic conditions should be alike, 116 if it’s the case that the strength of the pronominal dependency is determined only by morpho- syntactic features, not by discourse features. However, if it’s the case that syntax matters more than discourse but that discourse also plays some role, the predictions are different. In particular, according to Centering Theory (CT), referential structure effects of the Subject-Topic condition should show opposite patterns from the Object-Topic condition: if the subject of the pre-final clause is a topic (Subject-Topic condition), the following subject-position pro in the Two-pro structure will be more likely to be interpreted as referring to the object antecedent (e.g., Younghi in example (39)) than that in the One-pro structure. But in the Object-Topic condition, the subject-position pro in the One-pro structure is expected to be more likely to refer to the object antecedent than that in the Two-pro structure. This is because CT argues that pronouns are interpreted in such a way that maximizes coherence of discourse transitions, which is achieved by the most salient discourse entity being held constant across utterances. To see how this yields the prediction that (i) in the Subject-Topic condition, a subject pro should show a stronger object preference in the Two-pro than in the One-pro structure while (ii) in the Object-Topic condition, a subject pro should show a stronger object preference in the One-pro structure than in the Two-pro structure, let’s take a closer look at CT. Four main types of coherence transition have been proposed in CT: CONTINUE (the most coherent), RETAIN (the second most coherent) and SOFT/ROUGH-SHIFT (the least coherent). For Experiment 5, the specific CT predictions are illustrated in Table 15. Here, I consider the transitions for the scenario where the subject-position pro in the final clause refers to the preceding object (e.g., Younghi). I do not focus on the other interpretations that the preceding subject (e.g., Eunju) is referred to by the pro because they yield the same type of transition coherence regardless of referential structure type (CONTINUE in the Subject-Topic condition and SHIFT in the Object- 117 Topic condition). Thus, the effects of referential type stem from object-antecedent interpretations. CT Transition Top Ref. structure The final subject-position pronoun referring to the preceding object SOFT-SHIFT RETAIN Sbj- top Clause1/2 Eunju-ey-tayhay malhacamyen, pro Eunju Younghi-lul sokyess-nuntey, Eunju-about speaking.of, pro Eunju Younghi-Acc deceived-Con, Speaking of Eunju, pro deceived Younghi Clause3 (One-pro) a. prosbj Jieuni-lul cwuoyhayss-ki ttaymwun-i-ya. prosbj Jieuni -Acc cwuoyed-Nml reason-be-Dec because pro sbj(Younghi) cwuoyed Jieuni. Clause3 (Two-pro) b. prosbj proobj cwuoyhayss-ki ttaymwun-i-ya. prosbj proobj cwuoyed-Nml reason-be-Dec because pro sbj(Younghi) cwuoyed pro obj(Eunju). CONTINUE ROUGH-SHIFT Obj -top Clause1/2 Younghi -ey-tayhay malhacamyen, Eunju-ka pro Younghi sokyess-nuntey, Younghi -about speaking.of, Eunju-Nom pro Younghi deceived-Con, Speaking of Younghi, Eunju deceived pro Clause3 (One-pro) c. prosbj Jieuni-lul cwuoyhayss-ki ttaymwun-i-ya. prosbj Jieuni -Acc cwuoyed-Nml reason-be-Dec because pro sbj(Younghi) cwuoyed Jieuni Clause3 (Two-pro) d. prosbj proobj cwuoyhayss-ki ttaymwun-i-ya. prosbj proobj cwuoyed-Nml reason-be-Dec because pro sbj(Younghi) cwuoyed pro obj(Eunju). Table 15. Transition type predicted by CT in Experiment 5 Let’s first consider the Subject-Topic conditions. Here, in the One-pro version (null pro only in subject position and interpreted as referring to the preceding object, ‘Younghi’), the topic changes from ‘Eunju’ to ‘Younghi’ in the final clause (a: SOFT-SHIFT). However, in the Two-pro version (null pros in both subject and object position, with subject position pro referring to preceding object, ‘Younghi’), the same topic ‘Eunju’ can now be maintained in the final clause, thanks to it being mentioned via the object-position pro (b: RETAIN). Now, let us turn to the Object-Topic conditions. In the One-pro version (null pro only in 118 subject position and interpreted as referring to the preceding object), the preceding topic ‘Younghi’ can be maintained in the final clause (c: CONTINUE) by not mentioning the salient entity in the preceding clause (the subject ‘Eunju’ = preferred center, Cp in Clause 2). On the other hand, in the Two-pro version (null pros in both subject and object position, with subject position pro referring to preceding object), the topic changes from ‘Younghi’ to ‘Eunju’ in the final clause (d: ROUGH- SHIFT). In sum, if the null pro in subject position is interpreted as referring to the preceding object, the Subject-Topic and Object-Topic conditions are predicted by CT to yield opposite effects of referential structure type. Recall that the Syntax-over-Discourse Dependence View predicts that the overall referential structure effects are guided by the syntax-level cue-based parallelism such that the One- pro condition will have a stronger object antecedent preference than the Two-pro condition, in both Subject-Topic and Object-Topic conditions. However, discourse-level topicality modulates these effects, leading to a stronger effect of referential structure in the Object-Topic condition compared to the Subject-Topic condition. This is because both CT and syntax-level cue-based parallelism predict a stronger object antecedent preference in the One-pro structure than the Two-pro structure in the Object-Topic condition, but in the Subject-Topic condition, the predictions from the two mechanisms conflict (CT predicts a stronger object antecedent preference in the Two-pro structure). As a result, the effects of referential structure are predicted to be smaller in the Subject-Topic condition than in the Object-Topic condition. 4.4.5.2. Discourse-over-Syntax Dependence View Second, let us consider the possibility that referential structure effects are guided by both syntax- level cue-based parallelism and discourse-level coherence mechanisms, with the latter having a 119 greater influence. Let’s call this the Discourse-over-Syntax Dependence View. According to this view, discourse coherence plays a dominant role in shaping overall referential structure effects, which are modulated by syntactic parallelism. It is predicted that the overall patterns of referential structure effects will vary depending on the topicality type: in the Object-Topic condition, a stronger object antecedent preference is expected in the One-pro structure compared to the Two-pro structure, while the opposite pattern is expected in the Subject- Topic condition. However, since these overall patterns will be modulated by syntactic parallelism, referential structure effects are predicted to be smaller in the Subject-Topic condition than in the Object-Topic condition. This is because the predictions from the CT and those from the syntactic PFS are pitted against each other only in the Subject-Topic condition, but not in the Object-Topic condition. 4.4.5.3. Syntax-level / Discourse-level Dependence View The final possibility is that null pronominal dependencies are resolved by interacting either at the level of syntax or at the level of discourse (Syntax-level Dependence or Discourse-level Dependence View). In this case, it is predicted that the overall referential structure effects driven by either mechanism for anaphoric dependency resolution (the syntactic Cue-based PFS or discourse-coherence CT) will not be modulated by the other. Rather, the patterns of pro interpretation will follow the prediction either from the syntactic Cue-based PFS or from the discourse-coherence CT, but will not be sensitive to both. 120 4.4.6. Results Figure 15 shows the proportion of trials where the subject-position pronoun is interpreted as referring to the object of the preceding clause. Figure 15. Experiment 5: The proportion of trials where the subject-position pro refers to the preceding object. (error bars represent ±1 S.E.) Visually, Figure 15 shows that (i) referential structure influences null pronoun interpretation in the direction predicted by the syntactic parallelism, and that (ii) referential structure effects are modulated by topicality. To be specific, in the Object-Topic condition, the presence of an object-position null pronoun reduces the likelihood that participants interpret the subject-position null pronoun as referring to the preceding object (Two-pro condition: 64%), relative to sentences with an object-position full NP (One-pro condition: 83.3%). However, in the Subject-Topic condition, interpretation of the subject-position null pronoun in the One-pro condition (41.5%) does not markedly differ from its interpretation in the Two-pro condition (31.9%). In sum, Figure 15 suggests that null pronoun interpretation is less sensitive to the referential structure manipulation in the Subject-Topic condition than in the Object-Topic condition. For statistical analysis, I fit a generalized linear mixed logit model, using R (R 121 Development Core Team, 2020). Overall, there are main effects of both referential structure and topicality type, and significant interaction (see Table 16). The main effects of referential structure indicate that syntactic parallelism has a stronger impact than discourse coherence on null pronoun interpretation in Korean. However, planned comparisons (Table 17) show that the effect of referential structure is further modulated by topicality – referential structure effects are only significant in the Object-Topic condition, not the Subject-Topic condition. The absence of referential structure effects in the Subject-Topic condition diverges from the English experiment, which yielded referential structure effects in the Subject-Topic and Object-Topic condition (Chapter 3). I assume that this indicates that with null pronouns, effects of discourse-coherence are strong enough to reduce effects of syntactic parallelism. I return to this in the Discussion section. Thus, these results support the Syntax-over-Discourse Dependence View. Fixed effect Estimate Std. Error z value Pr(>|z|) (Intercept) -0.2866 0.1053 -2.722 0.00649 ** Ref.st type -0.7805 0.1460 -5.348 8.92e-08 *** Topicality type -1.7952 0.1232 -14.569 < 2e-16 *** Ref.st :Topicality -0.6099 0.2451 -2.488 0.01283 * Table 16. Experiment 5: results of the glmer model Topicality Fixed effect Estimate Std. Error z value Pr(>|z|) Sbj-Topic (Intercept) 1.0101 0.2948 3.427 0.000611 *** Ref.st type -0.2866 0.3923 -0.730 0.465093 Obj-Topics (Intercept) -1.5114 0.1986 -7.612 2.70e-14 *** Ref.st type -1.4509 0.3540 -4.099 4.16e-05 *** Table 17. Experiment 5: planned comparisons 122 4.5. Discussion This chapter broadened the scope of inquiry beyond reference resolution of overt pronouns in English to see whether there are differences in referential structure effects depending on anaphoric form. Experiment 5 focused on null pronouns in Korean, and is motivated by previous findings that the type of anaphoric form is correlated with the referent’s degree of topicality in discourse – the more reduced the anaphoric form is, the more topical its referent is (e.g., Givon, 1983; Ariel 1990, 1994). While discourse coherence factors (e.g., as predicted by Centering Theory) have been shown to exert a weaker influence on building interactions between anaphoric dependencies of the overt pronouns in English, relative to syntactic factors (i.e., cue-based parallelism) (see Chapters 2 and 3), how much the discourse coherence factors influence the pronominal dependency-building operation could differ for null and overt pronouns. In this chapter, I tested four hypotheses that make different claims about how referential structure effects are impacted by syntactic and discourse level factors. First, the Syntax-over- Discourse Dependence View assumes that the syntactic parallelism has a stronger impact than discourse coherence and predicts that the overall referential structure effects driven by the syntactic parallelism will be modulated by discourse coherence. Second, the Discourse-over-Syntax Dependence View predicts, conversely, that discourse coherence has a stronger impact than syntactic parallelism, and thus predicts that the overall referential structure effects resulting from the discourse coherence will be modulated by syntactic parallelism. Lastly, the Syntax-level Dependence and Discourse-level Dependence Views predict that pronominal dependencies involving null pronouns are resolved either purely at the level of syntax or at discourse, and thus these two accounts predict that effects of one factor on referential structure effects will not be modulated by the other factor. 123 In order to test these hypotheses, Experiment 5 used three-clause sequences and manipulated the referential structure type (One-pro vs. Two-pro) in the final clause and topicality in the pre-final clause (Subject-Topic vs. Object-Topic), as in Experiment 4. In Experiment 5, in the final clause, the Two-pro structure contains null pronouns in both subject and object position, while the One-pro structure contains a null pronoun in subject position and a new name in object position. In the pre-final clause, topicality is manipulated by the referent of a preceding context sentence being realized as a null form in either subject position (Subject-Topic condition) or in object position (Object-Topic condition). 4.5.1. Stronger syntactic parallelism effects in null pronoun resolution Experiment 5 replicated referential structure effects found in the English overt pronoun studies (Experiments 1 to 4), indicating that referential structure effects generalize across anaphoric form type in different languages. The Korean results show that null pronoun interpretation is sensitive to referential structure type: a null pronoun in subject position is more likely to be interpreted as referring back to the preceding subject in a configuration with two null pronouns than in a configuration with only one null pronoun. This shows that the syntactic cue-based parallelism plays a key role in the overall referential structure effects found in Korean. Moreover, given that the parallelism effects in single/multiple pronoun resolution are found in the contexts that do not involve semantic parallel relations (i.e., do not have resemblance relations), this suggests that parallelism effects on pronoun interpretation can occur independently of the semantic parallelism even with null pronouns, as well as overt pronouns. Crucially, Experiment 5 reveals that the effects of referential structure type are smaller in the Subject-Topic than in Object-Topic conditions. This indicates that the strength of referential 124 structure effects is affected by topicality type. These results are compatible with the Syntax-over- Discourse Dependence View. Thus, these findings suggest that syntactic parallelism has a stronger influence on establishing null pronominal dependencies than discourse coherence in Korean. 4.5.2. Stronger topicality effects: null pronouns vs. overt pronouns Given that null pronouns are known to be sensitive to topicality, one might be surprised to find that discourse-coherence factors did not exert a stronger influence on referential structure effects of null pronouns relative to syntactic parallelism factors in Experiment 5. However, in this section I show that detailed crosslinguistic analyses by topicality type providing evidence that the discourse-coherence factors do have a stronger impact on null pronoun resolution compared to overt pronoun resolution in English. Recall that discourse-coherence factors exert an influence in the opposite direction from the syntactic factors only in the Subject-Topic condition. With Korean null pronouns, referential structural differences were found only in the Object-Topic condition, but not in the Subject-Topic condition, while the English overt pronoun study (Experiment 4) showed differences both in the Subject-Topic and the Object-Topic condition. I assume that these cross-linguistic differences result from the influence of discourse-coherence factors with null pronouns being strong enough to completely offset the effects of syntactic parallelism when the two forces are pitted against each other. In the Subject-Topic condition, the discourse coherence factor removes differences in pronoun resolution by referential structure type (One-pro(noun) vs. Two-pro(noun)) guided by the syntax-level cue-based parallelism, only with the null pronouns, but not with the overt pronouns. In other words, I suggest that the discourse-coherence factor in null pronoun resolution is strong 125 enough to completely reduce the syntactic parallelism effect. Thus, this cross-linguistic difference suggests that the discourse coherence factor has a stronger impact on null pronoun resolution (in Korean) than overt pronoun resolution (in English). Moreover, the topicality effects found in Experiment 5 replicate and extend those found in Experiment 4. The interpretation of null pronouns is guided by the topicality of potential antecedents, with a significantly higher proportion of subject antecedent interpretations of subject- pros in the Subject-Topic condition compared to the Object-Topic condition. It is noteworthy that, at least visually, a comparison of the topicality effects between the English overt pronoun study (Experiment 4) and the Korean null pronoun study (Experiment 5) suggests that effects of topicality type are larger in Experiment 5 than in Experiment 4, as illustrated in Figure 16. Indeed, statistical analyses (glmer) comparing the proportion of object interpretations with the overt pronouns (Experiment 4) to that with null pronouns (Experiments 5) confirm that null pronouns elicit larger differences between the Subject-Topic and the Object-Topic condition than overt pronouns (Topicality type X Experiment type: beta = -1.6991, SE = .4311, z = -3.941, p<.001). Thus, these results clearly indicate that discourse-coherence factors play a more significant role in null pronoun resolution than in overt pronoun resolution. 126 Figure 16. Experiment 4 vs. Experiment 5: The proportion of trials where the subject-position pro refers to the preceding object. (Two-Pro(noun) and Pronoun+Name/One-Pro conditions are collapsed) (error bars represent ±1 S.E.) In sum, these findings show that null pronouns are more strongly guided by the discourse coherence (topicality) factor than are overt pronouns. This supports previous research on the relationship between anaphoric forms and topicality (e.g., Ariel, 1990; Arnold, 1998, 1999; Givón, 1983), which posits that null-form pronouns are more sensitive to the discourse prominence of their potential referents than are overt pronouns. However, it’s also worthwhile to consider the possibility of the differences between overt pronoun resolution in English and null pronoun resolution in Korean perhaps being a result of language differences rather than anaphoric form differences. To further investigate this, one direction for future work would be to compare overt pronoun resolution with null pronoun resolution in the same language, such as Korean and Chinese, which allow both overt and null form arguments in subject and object positions. For example, in Korean, referential structure effects with different anaphoric forms could be tested by using null pronouns and overt ‘quasi’-pronouns such as kyay, which consist of a demonstrative being followed by a generic noun (e.g., Oh, 2007, 2010; Kim, 2009). Before 127 continuing, it’s important to note that third-person overt pronouns in Korean (e.g., ku ‘he’, kunye ‘she’ ) are highly restricted in their distribution and rarely used in spoken language (Section 4.2), but that another class of elements – so-called overt ‘quasi’-pronouns – are known to be “productively used in naturally-occurring conversations, constituting an important reference type in contemporary Korean” (Oh, 2010, p.1224). The third-person overt ‘quasi’-pronoun kyay, in particular, has been claimed to be frequently used in natural speech contexts and to be in its later stage towards evolving into a genuine pronoun. (See e.g., Oh, 2007, 2010; Kim, 2009). Moreover, according to native speaker judgments, the ‘quasi’-pronoun kyay can occur in either subject position (40a), object position, or both (40b). (40) a. Hannai-ka Sarahj-lul sokyess-nuntey, kyayi/j-ka Joe-lul silhehayss-ki ttaymwun-i-ya. Hanna-Nom Sarah-Acc deceived-Con, Quasi.Pro-Nom Joe-Acc hated-Nml reason-be-Dec ‘Hannai deceived Sarahj because shei/j hated Joe.’ [One-Pronoun structure] b. Hannai-ka Sarahj-lul sokyess-nuntey, kyayi/j-ka kyayi/j-lul silhehayss-ki ttaymwun-i-ya. Hanna-Nom Sarah-Acc deceived-Con, Quasi.Pro-Nom Quasi.Pro-Acc hated-Nml reason-be-Dec ‘Hannai deceived Sarahj because shei/j hated heri/j.’ [Two-Pronoun structure] Therefore, given that the ‘quasi’-pronoun kyay is compatible with the referential structure manipulation as shown in the one-pronoun structure (40a) and the two-pronoun structure (40b), it would be informative to compares interpretation patterns between null pronouns and the ‘quasi’- pronoun kyay in Korean in order to clarify the ultimate source of referential structure effects with different anaphoric forms. 128 4.6. Conclusion To the best of our knowledge, Experiment 5 on Korean is the first experimental study on referential structure effects in null pronoun interpretation. Together with Experient 4 on English, it shows that referential structure effects are present across anaphoric form type, indicating that the interactions between multiple referential dependencies exist with both null and overt pronouns, and that discourse-coherence effects were stronger with null pronoun resolution than overt pronoun resolution. Note that while the results showed that discourse-coherence factors have a stronger impact on null pronoun resolution than overt pronoun resolution, the discourse-coherence factor was not strong enough to ‘flip’ the overall patterns of referential structure effects with null pronouns in Experiment 5 (i.e., Discourse-over-Syntax Dependence View). It is possible that the use of fully syntactically parallel structures in the experiment (i.e., transitive target clauses) may have reduced the impact of discourse-coherence factors (e.g., Smyth, 1994; Stevenson et al., 1995). As the effects of various factors on pronoun resolution may vary depending on the context, it is possible that discourse factors could play a greater role in different structures. 129 Chapter 5: Referential structure effects during real-time processing 5.1. Introduction The experiments reported in Chapter 2 to 4 provide evidence for the effects of referential structure in pronoun resolution. The results indicate that referential structure effects result from the presence or absence of the interactions between referential dependencies initiated by the two different pronouns. However, Experiments 1 to 5 examined how referential structure guides pronoun interpretation through the lens of end-state focused offline tasks (using a picture-writing task). Thus, these findings do not shed light on how referential structure affects pronoun resolution during real-time processing, which is the focus of this chapter. It is well-known that language comprehension occurs incrementally, including pronoun interpretation (e.g., Arnold et al., 2000; Cooper, 1974; Eberhard et al., 1995; Kaiser, 2011; Tanenhaus et al., 1995). Thus, this chapter explores (a) how our mental models of referential dependencies are dynamically updated, as cues about co-reference become available and (b) whether the patterns of pronoun resolution are different depending on referential structure type during real-time processing. From the perspective of incremental processing, what is crucially different between single pronoun resolution and two-pronoun resolution is the presence or absence of another pronoun’s dependency information when the comprehender encounters a pronoun that appears later in the sentence. As in (41), at the point when the comprehender encounters the sentence-final indirect object pronoun him, another referential dependency – involving the preceding subject-position pronoun he – already exists in Two-Pronoun sentences, as in the example (41b). However, no such preceding dependency is present in One-Object-Pronoun sentences, as in the example (41a). Both 130 examples include nonce words, I discuss those more later. (41) a. Tommy encouraged Max after the winter vacation because Hanna had wengered the neend to him. [One-Object-Pronoun] b. Tommy encouraged Max after the winter vacation because he had wengered the neend to him. [Two-pronoun] For brevity, in this chapter, I mostly refer to him as an object pronoun, not an indirect object pronoun. However, I acknowledge that indirect object pronouns may not pattern the same way as the direct object pronouns tested in Experiments 1 to 4. I return to this issue more in the discussion section of this chapter. For now, I am largely focusing on the distinction between subject pronouns vs. non-subject pronouns, i.e., pronouns in any kind of object position. Among the various cues that guide pronoun resolution, the preceding pronoun’s dependency information is a cue that is only available in the Two-Pronoun configuration, not in the One-Object-Pronoun configuration. Based on the offline results from Experiments 1 to 5, it is reasonable to expect that (a) the process of building referential dependencies for the object pronoun can be influenced by the preceding subject pronoun’s dependency information and that (b) the initial interpretation of the subject pronoun may get reanalyzed during the process of resolving the object pronoun. While the preceding (subject) pronoun’s dependency information becomes available as soon as a comprehender encounters the subsequent (object) pronoun in the same sentence, what remains an open question is when this information begins to be used in pronoun resolution. In other words, during real-time processing, are there different processes happening at him in the 131 One-Obj-Pronoun vs. Two-Pronoun conditions and if so, what is the time course of these processes? While most factors that guide reference resolution (e.g., morpho-syntactic or IC verb cues) have been argued to be utilized immediately after the anaphoric expression in an incremental manner (e.g., Greene & McKoon, 1995; Koornneef & van Berkum, 2006), it cannot be simply assumed that the preceding pronoun’s dependency cue will be used immediately in the same way as the other factors. This is because in this situation, we are testing whether comprehenders integrate information about the referential patterns of prior expression (the subject pronoun) into their processing of the current element (the object pronoun). This two-component process is arguably more complex than immediate use of cues from the prior sentence (e.g., IC verb bias) to resolve subject pronouns, a process where dependencies of another anaphoric element are not being considered. Figure 17a. One-Pronoun structure Figure 17b. Two-Pronoun structure Figure 17 illustrates the referential dependencies initiated from each pronoun. So far in this dissertation, I have argued that the subject pronoun he is more strongly linked to the subject antecedent (Henry) than to the object antecedent (Kevin) in the Two-Pronoun configuration. This is because grammatical role parallelism and discourse coherence operate within the structure where the same antecedents are shared by both the following object pronoun him and the preceding subject pronoun he (see Chapter 1 Cue-based PFS and Centering Theory for details). But what happens during real-time processing, at the moment that a comprehender Figure 1a. One-pronoun structure Figure 1b. Two-pronoun structure 132 encounters the second pronoun him? Let us consider two possible situations: The first scenario is the Immediate Activation Hypothesis, which posits that upon encountering the object pronoun (e.g., him) in a Two-Pronoun sentence, a comprehender activates both potential referents in the preceding clause (e.g., Henry and Kevin) and builds dependencies between him and both Henry and Kevin. Recall that there already exist dependencies from the subject pronoun he to potential referents, as the subject pronoun has already been encountered. According to the Immediate Activation Hypothesis, the interpretation of the object pronoun him immediately triggers dependency-building to potential referents and, crucially, is also immediately sensitive to any already-built dependencies that were triggered by the subject pronoun he. In other words, whatever happens at the point of hearing him is, right away, sensitive to the referential properties of the earlier parts of the sentence. Alternatively, the Delayed Activation Hypothesis assumes the situation that effects of pre- existing dependencies do not kick in until later. Perhaps, when encountering the object pronoun him, the comprehender again retrieves the representation of the potential referents in the preceding clause (e.g., Henry and Kevin) but now forms initial dependencies between him and the potential referents Henry and Kevin without paying any attention to what happened earlier in the sentence. In other words, under this view the object pronoun him is initially processed in an ‘encapsulated’ manner, separate from any pre-existing dependencies the comprehender may have built for the subject pronoun he. After the comprehender has built the initial dependencies for him, the pre- existing dependencies for he are then retrieved and the dependencies associated with each pronoun may be revised and re-analyzed. In other words, under this view, the presence of an earlier pronoun in the same clause initially has no effect on how the object pronoun is processed; the effects of a pre-existing pronoun only kick in at a late stage. 133 Thus, the crucial difference between the Immediate Activation Hypothesis and the Delayed Activation Hypothesis lies in the question of at what moment does the pre-existing dependency information (involving subject pronoun) exert its influence on the interpretation of the current (object) pronoun. The Immediate Activation Hypothesis predicts that referential structure effects (One-Pronoun vs. Two-Pronoun) will emerge early on during resolution of the object pronoun. But the Delayed Activation Hypothesis predicts that referential effects will emerge late during object pronoun resolution. Thus, the difference between the two hypotheses is a relative difference in terms of timing and it may be hard to distinguish between these two hypotheses, especially when dealing with webcam-based eye tracking whose timing properties are less well-understood than in-lab eye-tracking. To test these hypotheses, I conducted a webcam-based visual-world eye-tracking study, using PCIbex (Zehr & Schwarz, 2018) and the webhazer.js plug-in (e.g., Papoutsaki et al., 2016). This experiment focuses on how rapidly the preceding dependency information is utilized and integrated into forming the current pronominal dependency (e.g., indirect object pronoun), as well as revising the preceding pronominal dependency (e.g., subject pronoun). Since the visual-world eye-tracking method – where people hear linguistic stimuli while looking at images – relies on the observation that what people look at reflects how they interpret what they hear (e.g., Tanenhaus et al., 1995), this experiment will allow us to tap into the time course of referential structure effects. 5.2. Recency effects in pronoun interpretation Experiments 1 to 5 found that the effects of referential structure are driven by both syntactic and discourse level factors. However, the results do not speak to the question of whether these are the only factors that contribute to referential structure effects and whether they have the same weight 134 in different contexts. Indeed, prior work on reference resolution more generally shows that different factors exert their effects differently from one context to another. For example, Stevenson et al (1995) tested the effect of parallelism in pronoun interpretation (i.e., Parallel Function Strategy (PFS)) under the syntactically parallel contexts (42a) and non-parallel contexts (42b). However, both involved semantically parallel coherence relations, signaled by the connective and. The results of their forced choice task for the interpretation of the non-subject pronoun (e.g., her) demonstrated that the preceding subject (a non-parallel grammatical role) was preferred over the preceding object in the non-parallel structure (42b), but not in the parallel structure (42a). This indicates that the effect of parallelism in pronoun resolution can be hard to find under the structurally non-parallel context, providing supporting evidence for the claim that not all factors weigh in equally in reference resolution. (42) a. Jean asked Penelope to make an effort and Roger told her to have a break. b. Mary helped Priscilla change the wheel and Graham talked to her without interest. (Stevenson et al 1995, p. 401) In the current experiment (Experiment 6), I test sentences like those in (43), with the aim of exploring real-time referential structure effects using eye-tracking. Experiment 6 uses more complex and non-parallel structure with increased distance between the potential antecedents and pronouns, due to the demands of the visual-world eye-tracking method. 135 (43) a. Tommy encouraged Max after the winter vacation because Hanna had wengered the neend to him. [One-Object-Pronoun] b. Tommy encouraged Max after the winter vacation because he had wengered the neend to him. [Two-pronoun] As shown in (43), (i) the two clauses are not parallel in structure: a transitive matrix clause is followed by a ditransitive because-clause and (ii) the potential antecedents are mentioned further ‘upstream’ from the pronoun(s). This greater distance comes from an adverbial phrase (e.g., after the winter vacation) at the end of the matrix clause and the direct object (e.g., neend) before the indirect object pronoun. These elements are included because, when testing pronoun interpretation in visual-world eye-tracking, it is important to have look-aways that attract eye-gaze to a neutral position shortly before the critical pronoun(s) are encountered. Moreover, the sentences are also more ambiguous due to the use of (iii) two nonce words (e.g., wengered the neend) in the because-clause, along with a neutral (non-IC-biased) matrix verb (e.g., encourage) that does not favor subject or object as the antecedent of the subject-position pronoun. In Experiment 6, since referential structure effects are investigated in a more ambiguous context and with the length of the referential dependency increased, relative to Experiments 1 to 4; I assume that the different contextual circumstances may change how different factors guide pronoun resolution. In particular, the idea of recency now becomes very relevant. Among the factors that affect pronoun resolution, it has been claimed that the most recently mentioned element is favored as the antecedent for a pronoun. This is usually called a recency effect (e.g., Arnold, 1998; Clark & Sengul, 1979; Cunnings et al., 2014; Ehrlich & Rayner, 1983; Hall, 2021; Streb et al., 2004). Regarding 136 recency effects, it should be noted in that most of the cases, the effects were found in contexts where potential antecedents were separated from the pronoun by intervening sentences (i.e., quite a long distance) or where pronoun interpretation is highly ambiguous and complex due to multiple potential antecedents or complex sentence structure. In light of these kinds of findings, recency effects are often attributed to degrading of representations in memory over time. When representations decay, this makes retrieval harder, and thus more recent, less-degraded information is easier to retrieve (e.g., Streb et al., 2004; Van Dyke & Johns, 2012). In essence, as the length of the pronominal dependency increases, the more difficult it becomes to retrieve the potential antecedents. (44) a. The squadron paraded through town. Every soldieri who knew that Jamesj was watching was convinced that hei<j should wave as the parade passed. b. The squadron paraded through town. It looked to Jamesi that every soldierj was completely convinced that hei<j should wave as the parade passed. (Cunnings et al., 2014, p. 49) For example, Cunnings et al., (2014) investigated whether a quantified noun (e.g., every soldier) is preferred over a non-quantified noun (e.g., James) as the antecedent of a pronoun, using an offline questionnaire and eye-tracking study. As shown in (44), the pronoun is separated from the potential antecedents by being deeply embedded in a complex sentence. Although Cunnings et al. expected to find an interpretational difference depending on type of noun phrase (initial preference to the quantified noun), what they actually found was a recency effect: the linearly closest antecedent was preferred as the antecedent of the pronoun, regardless of the 137 quantificational status of the noun phrases. In light of these previous findings, the following sections explore the possibility of recency playing a role in referential structure effects, alongside the previously-discussed syntactic and discourse factors. 5.3. Experiment 6 Experiment 6 is a web-based visual-world eye-tracking experiment that examines the effect of referential structure during real-time processing. In this study, we look for referential structure effects by analyzing the proportion of looks to images that correspond to the potential antecedents of the pronouns. Experiment 6 is primarily different from Experiment 1 to 5 in three aspects. First, Experiment 6 tests sentences with pronouns in subject position and indirect object position, instead of subject position and direct object position. This change is due to the nature of the visual-world eye-tracking method: Testing indirect object pronouns allows for a look-away object to be mentioned in direct object position right before the pronoun, thus attracting eye-gaze to a neutral position. Second, the key dependent variable in Experiment 6 is looks to and interpretation of the pronoun in indirect object position, not in subject position. This change is due to the nature of the visual-world eye-tracking paradigm: it is easier to collect offline data for sentence-final pronouns. Furthermore, focusing on the indirect object pronoun means that comprehenders already know whether they are dealing with a One-Pronoun or Two-Pronoun context, so this is the best position to look for potential effects of referential structure. Third (and related to the second point), the task is different. The offline picture-writing 138 task used in Experiments 1 to 5 asks participants to write down antecedents for the pronouns while looking at a sentence coupled with a schematized picture. Experiment 6 uses eye-tracking coupled with a last-mentioned selection task: participants hear sentences auditorily while their eye-gaze is tracked, and participants click on the picture depicting whoever they think was mentioned last in the sentence. If referential structure effects are found in Experiment 6, despite the different syntactic position of the critical pronoun and different methodology, it would be strong supporting evidence for referential structure effect. Web-based eye-tracking is an emerging methodology that has been recently introduced in the field of psycholinguistics (e.g., Degen et al., 2021; Lee, 2022; Slim & Hartsuiker, 2021; Storbeck, 2022; V os et al., 2022). Thus, the results of Experiment 6 will contribute to validating this novel method. 5.3.1 Participants Ninety-eight native English-speaking adults, recruited on the internet via Prolific, completed this experiment. They were compensated $12. Data from five participants were excluded because they were either non-native English speakers or born outside the U.S. Moreover, I excluded one participant for reporting that they have a hearing impairment and five participants for reporting that they were not wearing headphones. 18 Additionally, four participants were excluded for poor accuracy either on attention check catch-trials (correct answers fewer than two out of six) or on practice trials (correct answers fewer than one out of six); one participant was excluded due to 18 Use of headphones ensures participants are able to hear the critical sentences clearly. 139 giving random answers throughout the experiment. Twelve further participants were excluded due to poor calibration scores (mean calibration score before targets < 40). All exclusion criteria were determined prior to data analysis. Thus, 70 participants were included in the final data analyses. 5.3.2 Materials and design As a visual-world eye-tracking study, every item in Experiment 6 included multiple sub-images (combined into a visual display) and an auditory stimulus. For each target item, participants listened to the two clauses connected by because (see examples in (45)). The first clause contains the two potential antecedents (e.g., Tommy and Max) for the following pronoun(s) and a transitive Implicit Causality (IC) verb that is equi-biased between the subject and the object antecedents. Since the equi-biased IC verbs do not create a strong bias to either antecedent under the explanation relation, this context can minimize the reference resolution cues towards either antecedent that come from the verb or the connective. Twenty-four equi-biased IC verbs were selected based on prior IC norming studies (e.g., Ferstl et al., 2011, Hartshorne & Snedeker, 2013), such that each verb’s average degree of IC bias as reported in prior norming data was between 40% to 60% or its degree of IC bias was between 40% to 55% at least in one norming data (equi-biased IC verbs (Stimulus-Experiencer and Agent- Patient type): preferred-bias M=48.0%, SD=17.4). By using the connective because in all target items, the coherence relation of every target sentence is established to be explanation and thus any potential confounds related to coherence relation variability can be avoided (see Hobbs, 1979; Kehler, 2002, 2008 on the importance of coherence relations). In addition, the two potential antecedents for the pronouns matched in gender, which was balanced across items (i.e., 12 male names and 12 female names). 140 (45) Sample item a. Tommy encouraged Max after the winter vacation because he had wengered the neend to Hanna. [One-Subject-Pronoun] b. Tommy encouraged Max after the winter vacation because Hanna had wengered the neend to him. [One-Object-Pronoun] c. Tommy encouraged Max after the winter vacation because he had wengered the neend to him. [Two-pronoun] In the second clause, referential structure type is manipulated by comparing a structure with pronouns in the subject and indirect object position (45c: Two-Pronoun condition) to a structure with only one pronoun in the either the subject position (45a: One-Subject-Pronoun condition) or the indirect object position (45b: One-Object-Pronoun condition). The eye movement patterns of the One-Pronoun conditions will serve as baselines for the Two-Pronoun condition (e.g., he/him vs. Hanna). It is worthy of note that the second clauses in all target items describe ditransitive events, which are structurally non-parallel to the preceding clause (a transitive clause). Since parallel grammatical roles between a potential antecedent and a pronoun function as a cue to guide pronoun resolution (Parallel Function Strategy), Experiment 6 tests the potential emergence of referential structure effects in a more complex and ambiguous context than in Experiments 1 to 5. Moreover, the direct object nouns (e.g., jerv, neend, chise) and ditransitive verbs (e.g., frimoused, plooned, wengered) are nonce words to exclude potential effects of nominal/verbal semantics in the process of pronoun resolution. Each nonce verb was only used once. In addition, the gender of the new name in the because-clause (e.g., Hanna) is differs from the gender of the 141 pronoun in the same clause to avoid being considered as a potential referent of the pronoun. Crucially, each clause contains phrases that serves as look-aways, attracting eye-gaze to a neutral position before the subject pronoun and indirect object pronoun – the time or location PPs (e.g., after the winter vacation) in the first clause appears before the subject pronoun, and the direct object in the because-clause (e.g., neend) appears before the indirect object pronoun. In addition to 24 target items, the experiment included 36 filler items. The fillers were of three types, including sentences with homophones, similar national flags or confusing landmarks and ambiguous PP-attachment structure (e.g., Natalie painted the boy with a brush). Six of these fillers were unambiguous catch trials that function as attention check trials. Targets and fillers were randomly intermixed and presented using a Latin-Square design. 5.3.3. Auditory (sentence) stimuli All the items were recorded by a male native speaker of American English, using the Praat software (Boersma & Weenink, 2021). During recording, the speaker repeated each target and filler item multiple times in randomized order. The sound files were edited using Praat as follows: To make the target sound files, for a given target item, the pronoun-containing clauses with because were extracted and then spliced with the first clause that was held constant across conditions. In other words, each target sentence shared the same first clause across conditions. After the target sound files had been assembled as described above, I marked the four time-alignments for further analyses with eye-gaze estimates, using Praat TextGrids: (i) the onset of speech/sentence onset (i.e. onset of the matrix-clause subject), (ii) the offset of the second clause subject noun/pronoun, and (iii) the onset of the second clause indirect object noun/pronoun, and (iv) the offset of the second clause indirect object noun/pronoun. 142 5.3.4. Visual stimuli Participants are presented with scenes that correspond to the narratives that they listen to. For target items, each scene contains five images as in Figure 18: three human characters corresponding to the preceding subject, object, and the new name in the pronoun-containing because clause (e.g., Tommy, Max and Hanna)); two look-away images corresponding to time or location PPs (e.g., after the winter vacation) in the first clause and the direct object in the because clause (e.g., neend). The five images are placed in different regions of the screen – upper-left, upper-right, bottom-left, bottom-right and center. The preceding subject image (e.g., Tommy) is randomly assigned to one of the four corners and the preceding object image (e.g., Max) is positioned on the opposite diagonal corner from the subject image (e.g., upper-left corner vs. bottom-right corner). Thus, the images for the critical subject and indirect object characters are always on opposite sides of the screen. The look-away image of the location/time phrase (e.g., after the winter vacation) is placed between the preceding subject and object in an either clockwise or counter-clockwise direction in order to divert eye gazes to an equi-distanced location from the potential antecedents (e.g., Tommy and Max) before encountering the following subject pronoun (e.g., he). The new name in the pronoun-containing because clause (e.g., Hanna) is placed in the remaining position among the four corners. Thus, this randomization design allows the positions of the three human characters and the location/time look-away image to be counterbalanced. In addition, the other look-away image in the because-clause (e.g., neend) is always centered on the screen to be placed in the equi-distanced location from the potential antecedents (subject and indirect object characters, who are on opposite sides of the screen) before encountering the following indirect object pronoun (e.g., him). 143 Figure 18. An example screen layout for the target sentence: Tommy encouraged Max after the winter vacation because he/Hanna had wengered the neend to Hanna/him. Fillers are ambiguous as regards the interpretations of the last-mentioned referent, except for the catch trials, which are always unambiguous passive sentences (e.g., The spider was found by Dorothy in Figure 19). I included different types of ambiguous filler sentences – semantically ambiguous sentences with a homophone, as in Figure 20 (e.g., While getting a perm in the beauty salon, Joanna read the book about the tortoise and the hare), structurally ambiguous sentences with multiple attachment sites for a prepositional phrase, as in Figure 21 (e.g., At the stroony trindle, Isaac sang to the girl with a microphone), and sentences with a name of a landmark or national flag with multiple confusing images, as in Figure 22 (e.g., While touring Europe, the students went on a churby trip to London and saw Big Ben). Some of the filler items contain nonce word(s) (e.g., stroony, trindle, churby) just as in target items. The fillers have two to five images for each item and their screen layout is more varied than the target items. In addition to the five areas on the screen that are used for the targets – upper- left, upper-right, bottom-left, bottom-right and center – the other remaining areas such as middle- left, middle-right, and upper-center are also used for the filler images, to prevent participants 144 becoming accustomed to a particular layout throughout the experiment. Figure 19. An example screen layout for the catch trial: The spider was found by Dorothy Figure 20. An example screen layout for the homophone type: While getting a perm in the beauty salon, Joanna read the book about the tortoise and the hare Figure 21. An example screen layout for the PP-attachment type: At the stroony trindle, Isaac sang to the girl with a microphone Figure 22. An example screen layout the landmark type: While touring Europe, the students went on a churby trip to London and saw Big Ben All the images are clip-art images in color adopted from Clipart.com (https://clipart.com/), except for the look-away nonce images. The look-away nonce images are photographs of novel objects adopted from The Novel Object and Unusual Name (NOUN) Database by Horst and Hout (2016). To select the maximally novel images from the database, I selected 24 images based on how familiar each object is (familiarity score below 40%) and how spontaneously the same name 145 can be come up with for each object (name-ability score below 50%). 19 5.3.5. Procedure The web-based visual world eye-tracking experiment was run remotely over the internet, using PennController IBEX (Zehr & Schwarz, 2018), which implements the eye-tracking JavaScript library Webgazer.js library (Papoutsaki et al., 2016). The Webgazer.js library enables the participants’ eye gaze information gathered through their webcams to be tracked, analyzed and recorded. The eye-tracking experiment begins with an initial calibration phase. During this phase, participants were asked to fixate on a series of green circles that sequentially appeared and disappeared in the center and around the screen. After a final circle on the screen, the initial calibration score was given to them out of 100%. The participants had five calibration attempts in total to score at least 60%. 20 If the score was below 60%, they repeated the calibration procedure until the maximum number of attempts with some tips on how to improve their calibration score. However, the participants who were not able to score at least 60% until the maximum attempts still proceeded to the main experiment. After the initial calibration was done, participants were instructed that they would listen to sentences while looking at the pictures of people and things about the sentences. The task was to click on the picture of whatever they thought was mentioned last in the sentence they heard. Before advancing to the main experiment, participants were given six practice items with up to three 19 In the familiarity and name-ability scores, the lower the score is, the more novel the object is. 20 This calibration parameter was determined based on previous web-cam based eye-tracking experiments conducted in the USC Language Processing Lab, including Storbeck (2022). 146 attempts per item. For each practice item, they were provided with text-based feedback, depending on their click selections and the number of attempts they had used. When the click selection was correct, a message appeared confirming that they chose the right answer, as shown in Figure 23a (e.g., That's right! It was "Last Friday the dax was bitten by the squid."). When the click selection was incorrect, the feedback became more and more explicit as to which one was the last-mentioned referent, as reaching the maximum number of attempts (e.g., First attempt: Not quite! It was "Last Friday, the dax was bitten by the squid." Please click on whatever is mentioned last., Second attempt: One more chance! What the dax was bitten by?, Third attempt: Sorry! The squid is mentioned last). Moreover, two practice items contained non-anaphoric expressions and four practice items contained subject-position or object-position pronouns as the last-mentioned referent, to avoid misunderstanding about the task – both anaphoric and non-anaphoric expressions can be the last-mentioned referent. I used this detailed feedback phase to ensure that participants understood the task; Because the study was done entirely remotely, there was no opportunity for the experimenter to observe the participant to see if they understood. Crucially, the practice items were designed so as not to bias participants’ responses on the critical target trials in the main experiment. 147 Figure 23a. An example screen of the practice item after success Figure 23b. An example screen of the practice item after failure (1st attempt) In the main experiment, each trial began by checking participants’ calibration. During the pre-trial calibration checks, participants fixated on a green circle in the center of the screen. Although the pre-trial calibration procedure was presented on the screen for every trial, the calibration score was recorded only for the targets but not for the fillers or practice items because the eye-gaze estimates were not recorded during the filler and practice items. For the target items, if the calibration score was lower than 40%, participants repeated the calibration procedure they had done initially up to two times to score at least 40%. Even if their re-calibration scores did not meet the minimum threshold (at least 40%) after the two attempts, they still proceeded to the trial. Crucially, the participants whose initial calibration score was lower than 40% or whose pre-trial re-calibration attempts exceeded eight times in total were not asked to re-calibrate in the pre-trial calibration phase, regardless of their scores. This measure can reduce time and effort taken from the participants whose pre-trial calibration score would be very likely to be low, considering their previous calibration scores. In addition, data from those participants whose mean pre-trial calibration scores were below 40% were excluded from subsequent analysis. After the pre-trial calibration checks, a central fixation cross appeared on the screen for 148 500ms and then the visual scene was presented. The audio play started 4000ms after the visual scene was displayed to let participants have time to inspect the images. After the audio finished playing, participants were given 10000 ms to click on the ‘last-mentioned element’. Upon clicking or after 10000 ms after the audio stopped, the trial ended and proceeded to the next trial. The experiment lasted approximately 40 to 60 minutes. Participants’ offline mouse click selections and timings for the last-mentioned referents were recorded along with the real-time eye movements that were tracked from the onset of the fixation cross until they click on the image up to 5000 ms after the offset of the audio play to avoid potential data loss. 5.3.6. Predictions In the visual-world eye-tracking experiment, two different types of data are measured: participants’ click selections for the last-mentioned referent of the sentence and their real-time eye-gaze patterns before the click selections. As the two types of data provide different sort of information regarding the effects of referential structure, in this section, I discuss predictions for (i) the last-mentioned referent selections, which are made after the end of each sentence and (ii) the real-time eye-gaze data during pronoun interpretation. 5.3.6.1. Predictions about last-mentioned selections As discussed in the Materials and design section, the last-mentioned entity corresponds to the different types of referents depending on target conditions: in the One-Subject-Pronoun condition as in (46a), the last referent mentioned is a new name in the because-clause, whereas in the One- Object-Pronoun as in (46b) and Two-Pronoun condition as in (46c), it is the antecedent of an indirect-object pronoun. Thus, for the One-Subject-Pronoun condition, it is predicted that the last- 149 mentioned referent is more likely to be interpreted as referring to the new name in the same clause compared to the other two conditions. (46) Sample item a. Tommy encouraged Max after the winter vacation because he had wengered the neend to Hanna. [One-Subject-Pronoun] b. Tommy encouraged Max after the winter vacation because Hanna had wengered the neend to him. [One-Object-Pronoun] c. Tommy encouraged Max after the winter vacation because he had wengered the neend to him. [Two-pronoun] For the One-Object-Pronoun and Two-Pronoun condition, it is expected that referential structure effects will be observed, showing different selection patterns from each other. In the previous chapters, referential structure effects were tested in structurally parallel contexts (e.g., Jane respected Mary because she daxed her/Lisa) and affected by both factors at the level of syntax and discourse. However, as discussed in section 5.2, I assume that now referential structure effects could emerge in different ways because the different contextual circumstances may cause different factors to guide pronoun resolution – Experiment 6 uses more ambiguous context with the length of the referential dependency increased, relative to the previous experiments. Regarding the syntax-level parallelism factor (cue-based Parallel Function Strategy), I predict that it will play a very little role in referential structure effects contrary to Experiment 1 to 5. Given that each target sentence consists of structurally non-parallel clauses and do not have connectives indicating a parallel relation (e.g., Hobbs, 1979; Kehler, 2002; Kehler et al., 2008), I 150 presume that grammatical role parallelism may not be actively used as one of the retrieval cues. Since all the retrieval features are equal between the two potential antecedents (e.g., Tommy and Max) except for grammatical role (one is a subject, one is a direct object) and neither antecedent is parallel to the following indirect object pronoun in terms of grammatical role, it is predicted that syntactic parallelism cue (i.e., parallel grammatical roles) may not contribute to eliciting referential structure effects – namely, there would be no differences in indirect object pronoun interpretation (e.g., to him) between the One-Object-Pronoun and the Two-Pronoun configuration. However, the predictions are different for the discourse-level coherence factor. In line with predictions inspired by Centering Theory, I expect that local discourse coherence may contribute to the interaction between the referential dependencies initiated from two different pronouns in the current target sentence structure. It is because Centering Theory (CT) views pronoun resolution as a part of holistic process that forms information structural coherence between utterances. Thus, I predict that different patterns of indirect pronoun resolution will emerge depending on referential structure type. To be specific, the subject in the preceding clause should be selected more often as the antecedent for the indirect object pronoun in the One-Object- Pronoun condition relative to in the Two-Pronoun condition. According to Centering Theory (CT), the most coherent transition in the Two-Pronoun condition can be established when the indirect object pronoun refers to the object antecedent (CONTINUE in (c)), not the subject antecedent (RETAIN in (d)), as illustrated in Table 18. In contrast, in the One-Object-Pronoun condition, the second most coherent transition (RETAIN in (a,b)) obtains regardless of which antecedent is referred to by the indirect object pronoun. In sum, since CT claims that pronouns are resolved in such a way that maximizes coherence of the discourse transition, I predict that the subject antecedent preference will be weaker in the 151 Two-Pronoun condition than in the One-Object-Pronoun condition. CT Transition Ref. structure Examples RETAIN One-Object- Pronoun a. Tommyi encouraged Maxj after the winter vacation because Hanna had wengered the neend to himi. b. Tommyi encouraged Maxj after the winter vacation because Hanna had wengered the neend to himj. CONTINUE Two- Pronoun c. Tommyi encouraged Maxj after the winter vacation because he1 had wengered the neend to himj. RETAIN d. Tommyi encouraged Maxj after the winter vacation because hej had wengered the neend to himi. Table 18. Transition type predicted by CT in Experiment 6 Lastly, I predict that the new factor of antecedent recency may come into a play in pronoun resolution in this experiment, contrary to the previous experiments. As discussed in section 5.2, recency has been reported to guide pronoun resolution, in particular, in highly complex and ambiguous contexts where potential antecedents are distant from the pronouns (e.g., Arnold, 1998; Clark & Sengul, 1979; Cunnings et al., 2014; Ehrlich & Rayner, 1983; Hall, 2021; Streb et al., 2004). Considering that the current target structure is complex and ambiguous, I expect to see referential structure effects being influenced by recency. More concretely, it is predicted that the subject in the preceding clause will be more likely to be selected as the antecedent in the indirect object pronoun in the Two-Pronoun condition relative to in the One-Object-Pronoun condition. This is because there is competition for the object antecedent (most recently mentioned) between the subject and indirect object pronoun in the Two-Pronoun condition, while there is no competition for it in the One-Object-Pronoun condition. 152 5.3.6.2. Predictions about eye-gaze patterns during pronoun interpretation The eye-gaze data provides information about how the process of pronoun resolution is dynamically updated in real time, as available co-referential cues become available. I focus here on eye gaze patterns during the interpretation of the object-position pronoun. Among the cues that were manipulated for pronoun interpretation in this study, I assume that the preceding pronoun’s dependency information is the cue that critically influences referential structure effects, in the sense that only Two-Pronoun configurations contain such a preceding dependency; One-Pronoun configurations with only an object pronoun obviously do not. Accordingly, participants’ looking behaviors as they hear the object position pronoun are the principal domain of analysis. (For completeness, I also report eye-gaze patterns during the subject-position pronoun, though I do not expect to see effects of referential structure at this pronoun, simply because the crucial manipulation – absence vs. presence of a second pronoun – has not yet been encountered at that point.) Gaze patterns during subject pronoun interpretation: As mentioned above, I do not expect to find effects of referential structure at the subject-position because at this point, the One- Subject-Pronoun and Two-Pronoun do not yet differ in regards to our key manipulation (presence of a second pronoun). I do not expect the two conditions to show different gaze patterns during processing of the subject-position pronoun. What about verb bias? I used equi-biased IC verbs in the matrix clause, so as far as verb- driven effects are concerns, participants are not expected to show a preference for either the preceding subject or object when they process the subject-position pronoun in the because-clause. Although the IC verb bias is one of the cues that has been known to have a very strong influence on pronoun interpretation (e.g., Cozijn et al., 2011), other cues could come into play as 153 well. If the discourse-level coherence factor contributes to pronoun interpretation, I predict that the proportion of looks to the subject will be slightly higher than the proportion of looks to the object due to a preference to interpret pronouns in a way that keeps the most discourse central entity constant across utterances; on the contrary, if the recency factor exerts its influence, I predict that the proportion of looks to the object (the most recently mentioned referent) will be slightly higher than that of looks to the subject. In contrast, I expect to see a high proportion of looks to the new named character (e.g., Hanna in (46)) in the because-clause in the One-Object-Pronoun condition, where an unambiguous name occurs in subject position. This condition serves as a baseline for the ambiguous pronoun conditions. Gaze patterns during indirect object pronoun interpretation: When participants in the Two-Pronoun condition encounter the indirect object pronoun, I propose that the fact that they have already encountered a subject-position pronoun and constructed a referential dependency for it will impact their processing of the indirect object pronoun. It is also possible that they will reanalyze the dependency they initially built for the subject-position pronoun and link that pronoun to a different referent than what they initially considered. However, in the One-Object-Pronoun structure, there is no pre-existing pronominal dependency to worry about. Here, I present two competing predictions regarding how quickly the preceding dependency information exerts its influence during the interpretation of the indirect object pronoun. They make different predictions about when effects of referential structure emerge. First, the Immediate Activation Hypothesis predicts that referential structure effects will emerge early on during indirect object pronoun resolution (as soon as the indirect object is heard). As illustrated in Figure 24, under this approach, the representations of the potential referents in the 154 preceding clause (e.g., Lucy and Ivy) are assumed to be immediately retrieved along with the dependencies linked to the preceding subject pronoun (e.g., she). As soon as the indirect object pronoun (e.g., her) starts to establish its referential dependencies, the initial dependencies of the subject pronoun begin to be reanalyzed using available cues. During this pronominal resolution process, the subject pronoun’s referential dependencies and related information are immediately utilized into building referential dependencies with the indirect object pronoun, also leading to the preceding pronominal dependencies being interactively reanalyzed. Thus, in this case, emergence of referential structure effects (Two-Pronoun vs. One-Object-Pronoun condition) is predicted to be immediate upon hearing the indirect object. Figure 24. A visualization of the Immediate Activation Hypothesis In contrast, the Delayed Activation Hypothesis predicts that referential structure effects emerge relatively later during the process of interpreting the indirect object pronoun. Under this view, the initial interpretation of the indirect object pronoun (e.g., her) occurs without regard for the possible existence of prior dependencies constructed with the other pronoun (e.g., she) in the same clause. As illustrated in ‘the 1 st stage’ in Figure 25, a comprehender initially interpret indirect object pronoun in an ‘encapsulated’ way, without regard to how the subject pronoun was interpreted. However, it is only during a second stage of resolution process (the ‘2 nd stage’ in Figure 25) that pre-existing pronominal dependencies come into play and constrain and/or are constrained 155 by the processing of the indirect object pronoun. Thus, during this second stage of processing, the fact that there exist referential dependencies already built between the subject pronoun and the preceding subject and object (e.g., Lucy and Ivy) will impact how the comprehender establishes dependencies between the indirect object pronoun and the preceding referents (which interactively impact reanalysis of the subject pronoun’s dependencies). In this case, emergence of referential structure effects (Two-Pronoun vs. One-Object-Pronoun condition) is predicted to be delayed and unable to find the effects immediately upon hearing the indirect object. 1 st stage: 2 nd stage: Figure 25. A visualization of the Delayed Activation Hypothesis 5.4. Offline mouse-click results 5.4.1. Data processing and analysis To analyze referent selections for the last-mentioned pronouns, I focus on the One-Object-Pronoun vs. Two-Pronoun conditions (The One-Subject-Pronoun condition is not relevant as it ends in a name, i.e., fully unambiguous.). In my analyses, I look at how often the indirect object position pronoun was interpreted as referring to the preceding subject. Thus, subject selections were coded as 1 and object selections as 0. The dependent variable was the proportion of subject selections, and the independent variable (referential structure) was contrast-coded (Object-Pronoun = 0.5, Two-Pronoun = −0.5). However, when participants clicked on something other than the preceding 156 subject or object, the trial was marked as N/A (2.5% of all responses) and excluded from analysis. In addition, I compared participants’ referent selections for the last-mentioned pronouns with those of the last-mentioned names, as a ‘sanity check’ to confirm that the expected patterns emerged with the novel webcam-based eye-tracking study. The referents of the last-mentioned names are unambiguous, contrary to the last-mentioned pronouns, and thus provide an unambiguous baseline. The responses of the One-Subject-Pronoun condition (where the object is an unambiguous name) were compared with those of the One-Object-Pronoun condition (where the object is a pronoun) and with those of the Two-Pronoun condition (where the object is also a pronoun), respectively, and the data was coded as either 1 (name selections) or 0 (preceding subject or object selections). The dependent variable was the proportion of selections of the because-clause name, and referential structure type was included in models as fixed effects (One-Object-Pronoun = 0.5 vs. One-Subject-Pronoun = −0.5; Two-Pronoun = −0.5 vs. One-Subject-Pronoun = 0.5). Moreover, the trials where the because-clause name or matrix-clause referents were not selected were excluded from final analysis (I excluded 2.4% of the data from the Object-Pronoun and the Subject-Pronoun condition, and 2.3% of the data from the Two-Pronoun and the Subject-Pronoun condition). The final data were analyzed with a generalized linear mixed logit model (glmer), using R statistical software (R Development Core Team, 2019) and the lme4 package (Bates et al., 2015) and lmertest (version 3.1.1) (Kuznetsova et al., 2017). As random effects, intercepts for subjects and items were entered as well as by-subject and by-item random slopes for the effects of referential structure type. A maximum random effect structure was used when justified by model comparison. The initial random effect structure that was fully crossed and fully specified with by- subject and by-item effects of referential structure type was reduced if needed due to 157 nonconvergence (starting with by-item effects). Only random effects that show a significant contribution to the model (p < 0.05) were included (Baayen et al., 2008). 5.4.2. Last-mentioned referent selections Figures 26 and 27 show participants’ mouse-click selections of the last-mentioned referent. The percentage of trials on which the indirect object position pronoun was interpreted as referring to the preceding subject is visualized in Figure 26; the proportion of trials where the indirect object- position pronoun was interpreted as referring to the new name in the pronoun-containing because clause is visualized in Figure 27. Figure 26. How often did participants interpret the last-mentioned referent as referring back to the preceding subject? (error bars represent ±1 S.E.) 158 Figure 27. How often did participants interpret the last-mentioned referent as referring back to the name in the because clause (error bars represent ±1 S.E.) When visually comparing the two conditions that end in a pronoun (One-Object-Pronoun vs. Two-Pronoun), Figure 26 shows that referential structure type influences pronoun interpretation: The presence of another pronoun in the because clause (Two-Pronoun condition) markedly increases the percentage of subject interpretations (60.2%) relative to configurations with only one pronoun in the indirect object-position pronoun (One-Object-Pronoun condition) (46.8%). This suggests that pronoun interpretation is sensitive to referential structure type and antecedent recency. In addition, when we look at how often participants select the picture of the new named referent (e.g., Hanna) as the last-mentioned referent (Figure 27), we see that in conditions with last-mentioned names (e.g., he had wengered the neend to Hanna), participants correctly clicked on the picture of Hanna 96.1% of the time, as expected, and very rarely in the pronoun-final conditions, again as expected. This is reassuring, as it shows that participants correctly understood the task. For statistical analysis, I fit a generalized linear mixed logit model (glmer) and found main effects of referential structure when referent selections for the last-mentioned pronouns were 159 compared (One-Object-Pronoun vs. Two-Pronoun in Table 19), and also when referent selections for the last-mentioned names were compared with the last-mentioned pronouns (One-Subject- Pronoun vs. One-Object-Pronoun in Table 20; One-Subject-Pronoun vs. Two-Pronoun in Table 21). Crucially, the significant subject preference found in the Two-Pronoun condition compared to the One-Object-Pronoun condition points to effects of antecedent recency. It indicates that competition for the most recently mentioned antecedent during pronoun interpretation results in stronger subject-antecedent preference in the Two-Pronoun condition relative to the One-Object- Pronoun condition. Fixed effect Estimate Std. Error z value Pr(>|z|) (Intercept) 0.3230 0.1638 1.971 0.048699 * Ref. structure type -0.7647 0.2297 -3.329 0.000872 *** Table 19. Experiment 6: results of the glmer model comparing One-Object-Pronoun vs. Two- Pronoun condition Fixed effect Estimate Std. Error z value Pr(>|z|) (Intercept) -0.6379 0.3581 -1.782 0.0748 . Ref. structure type -10.1311 1.0313 -9.824 <2e-16 *** Table 20. Experiment 6: results of the glmer model comparing One-Subject-Pronoun vs. Object- Pronoun condition Fixed effect Estimate Std. Error z value Pr(>|z|) (Intercept) -1.3009 0.7341 - 1.772 0.0764 . Ref. structure type 14.8156 3.5668 4.154 3.27e-05 *** Table 21. Experiment 6: results of the glmer model comparing One-Subject-Pronoun vs. Two- Pronoun condition 160 5.5. Online eye-tracking results 5.5.1. Eye-gaze patterns during the matrix clause Before analyzing participants’ gaze patterns during pronoun resolution, the gaze patterns during the matrix clause were visually inspected to check whether participants showed the expected looking behavior. Since the web-based eye-tracking is a new method that has only recently begun to be used, if this method works well as planned, I expect to see three distinct peaks in looks to each referent mentioned in the matrix clause (the subject, object, and time/location look-away) corresponding to the order that each referent is mentioned. During the matrix clause period, I also expect to see the same gaze patterns across conditions for each item (One-Object-Pronoun, One- Subject-Pronoun, and Two-Pronoun), since the matrix clause was constant within items. Figure 28 shows the gaze patterns that are aligned to the onset of the matrix-clause subject. This figure includes three regions of interest (the subject, object, and time/location look-away) for each condition. The proportions of looks to each referent were calculated within 200 ms time bins. Figure 28. Eye movements relative to the onset of the matrix-clause subject (each facet shows looks to the three regions of interest within a condition) 161 Figure 28 confirms that participants’ looking patterns during the matrix clause pattern as expected. In all three conditions, the temporally initial peak in looks arises for the matrix subject, the following peak for the matrix object, and the final peak in looks arises for the look-away. This indicates that the novel web-cam method can yield interpretable data. Having checked this, I now provide the statistical analysis of the gaze patterns during pronoun resolution in the following sections. 5.5.2 Data processing and analysis for the because-clause For statistical analyses, I aligned eye-gaze data to points of interest, determined the duration of time windows and calculated subject advantage scores. To look at eye-movements triggered by pronouns, I looked at eye-gaze estimates aligned to the offset (end) of the because-clause subject (e.g., because he || had wengered the need to Hanna/him) and onset of the indirect object in the because-clause (e.g., because he/Hanna had wengered the need to || him). In most prior eye-tracking studies, temporal analyses start at the onset of a critical word. However, in the current study, I looked at the offset of the critical pronoun (e.g., because he || had wengered the need to Hanna/him). This was done due to the difficulty in identifying the onset of the pronoun when the final segment of because is followed by she (assimilation for place of articulation). Thus, for sake of consistency, and following Storbeck (2022), I opted to align the gaze patterns to the end (offset) of the subject pronoun in all conditions, as the offset could be identified more reliably. It is worth noting that this issue does not impact the central analyses focusing on the object pronoun, as those are aligned to the onset of the sentence-final pronoun or name (e.g., because he/Hanna had wengered the need to || him). For time windows of analysis, eye-movements were binned into 500 ms time windows. 162 These are longer than the 200 ms time windows often used in lab-based eye-tracking studies (e.g., Altmann & Kamide, 2004; Do & Kaiser, 2019; Hwang & Kaiser, 2014; Salverda, Kleinschmidt, & Tanenhaus, 2014), which is a choice I made due to the lower and more variable sampling rates associated with web-cam gaze tracking. 21 Following prior work (e.g., Arnold et al., 2000, 2007; Kaiser, 2011), I calculated subject advantage scores by subtracting the proportion of looks to the object antecedent from those to the subject antecedent. Thus, the subject advantage scores allow us to assess how the strength of the preference for the subject antecedent over the object antecedent is affected by referential structure type (e.g., One-Object-Pronoun vs. Two-Pronoun or One-Subject-Pronoun vs. Two-Pronoun). Positive advantage scores indicate that participants look more at the subject than the object, while negative advantage scores indicate that participants look more at the object than the subject. The subject advantage scores were analyzed with a linear mixed logit model (lmer), using R statistical software (R Development Core Team, 2019) and the lme4 package (Bates et al., 2015) and lmertest (version 3.1.1) (Kuznetsova et al., 2017). Referential structure (contrast-coded, One- Subject-Pronoun = -0.5 vs. Two-Pronoun = 0.5 or One-Object-Pronoun = -0.5 vs. Two-Pronoun = 0.5) was included in models as fixed effects, and intercepts for subjects and items were entered as well as by-subject and by-item random slopes for the effects referential structure type as random effects. A maximum random effect structure was used when justified by model comparison. The initial random effect structure that was fully crossed and fully specified with by-subject and by- item effects of referential structure type (starting with by-item effects) was simplified via model comparison. Only random effects that show a significant contribution to the model (p < 0.05) were 21 In the current study, the median sampling rate was approximately 22 Hz, while lab-based eye-trackers have sampling rates of 500-1000 Hz (e.g., Hwang & Kaiser, 2014; Xu et al., 2015). 163 included (Baayen et al., 2008). 5.5.3. Eye-gaze patterns during the subject in the because-clause Before turning to the main analysis – examining where people look when they hear the object pronoun (e.g., …because he/Hanna had wengered the need to him) – I first consider the looking patterns triggered by the subject pronoun (e.g., because he had wengered the need…). Figure 29 plots the proportions of looks to the four regions of interest (the matrix subject, matrix object, time/location look-away, and because-clause name in the subject-position), relative to the offset of the subject in the because-clause (pronoun-containing clause (because he || had wengered the need…). Thus, 0ms in Figure 29 is the offset of the subject. The 500 ms time windows begin 500 ms prior to the offset of subject and continue for 1500 ms after it. Figure 29. Eye movements relative to the offset of the because-clause subject (each facet shows looks to the four regions of interest within a condition) 164 In Figure 29, only in the One-Object-Pronoun condition (top graph) shows a sharp increase in looks to the picture of the named character (e.g., Hanna). This is expected, because only that condition mentions a name in subject position the other conditions contain a pronoun in that position. The One-Subject-Pronoun and Two-Pronoun conditions (two lower graphs) show similar gaze patterns. This is expected, because at the point where participants hear the subject pronoun, the preceding input in these two conditions is the same. Indeed, statistical analyses confirm that the two conditions pattern alike: Statistical analyses over the subject advantage scores were conducted using linear mixed models on four 500 ms time-slices (-500 ms to 1500 ms). The analyses (Table 22) reveal that the two conditions are not significantly different from each other during any time slices. In sum, the looks triggered by the subject pronoun in the One-Subject- Pronoun and Two-Pronoun conditions do not differ. This is expected, given that the referential structure manipulation has not yet been encountered. Descriptively, as regards the raw patterns of looks to the preceding subject and object, the visualization (Figure 29) suggests that – in both conditions – before people hear the subject of the because-clause, they tend to look more at the preceding object (black circles) than the preceding subject (open circles), probably simply because the object was just recently mentioned. After participants hear the subject of the because-lause (e.g., because he had wengered the need…), the object preference weakens but numerically there are more looks to the object than to the subject throughout the time window shown. This pattern is compatible with recency effects combined with semantic effect from the equi-biased IC verbs. 165 Timing Fixed effect Estimate Std.Error df t value Pr(>|t|) -500 – 0ms (Intercept) -0.033123 0.008035 47.020006 -4.122 0.000151 *** Ref. structure -0.005322 0.015160 689.080352 -0.351 0.725658 0 – 500ms (Intercept) -0.011734 0.007355 20.699629 -1.596 0.126 Ref. structure 0.006953 0.012252 688.624270 0.567 0.571 500 - 1000ms (Intercept) -0.005021 0.009460 48.894791 -0.531 0.598 Ref. structure -0.006071 0.014998 691.839206 -0.405 0.686 1000- 1500ms (Intercept) -0.005263 0.012609 35.282799 -0.417 0.679 Ref. structure -0.008903 0.015364 687.946032 -0.579 0.562 Table 22. Summary of fixed effects for the models predicting the strength of the subject advantage scores at time intervals aligned to the offset of the because-clause subject 5.5.4. Eye-gaze patterns during the indirect object in the because-clause Let us now consider the key question: Where do people look as they hear the indirect object pronoun? This analysis is more informative with respect to the hypotheses being tested because the object pronoun is the point where the effects of referential structure can be detected: here, the One-Object-Pronoun condition differs from the Two-Pronoun condition, in terms of the presence vs. absence of another pronominal dependency in the same clause. Figures 30 and 31 present visualizations of the eye-gaze patterns aligned to the onset of the indirect object of the because-clause (e.g., … had wengered the need to || him/Hanna). 166 Figure 30. Eye movements relative to the onset of the because-clause indirect object (each facet shows looks to the three regions of interest) Figure 31. Eye movements relative to the onset of the pronoun in the because-clause indirect Figure 30 plots the proportions of looks to three regions within each condition (the matrix subject, matrix object, and because-clause name in the indirect object position, while Figure 31 only plots the proportions of looks to the preceding subject and object for the One-Object-Pronoun 167 and Two-Pronoun conditions. The 500 ms time windows begin 500 ms prior to the onset of the because-clause indirect object (0 ms) and continue for 4000 ms after it. 22 As expected, different gaze patterns are observed at the indirect object position in the pronoun conditions and the new name condition (the One-Object-Pronoun and Two-Pronoun conditions vs. One-Subject-Pronoun condition, e.g., …had wengered the need to him/Hannah). First, note that when participants hear a name in object position (e.g., Hanna), they look at the picture of that character: This is clearly visible in the middle graph in Figure 30, which shows a high proportion of looks to the picture of the named character (open triangles). In contrast, when participants hear a pronoun in object position (e.g., him), they look at the picture of either the subject or object character in the matrix clause (e.g., Tommy or Max). This is evident in the top and bottom graphs in Figure 30, which display a high proportion of looks to the picture of the preceding subject (open circles) and object character (black circles). Overall, there appears to be competition between the preceding subject and object in both of these conditions. However, at least visually, people look more at the preceding subject than the preceding object in the Two-Pronoun condition than the One-Object-Pronoun condition, starting from around 500 ms until 3100 ms after the onset of the indirect object. To assess the strength of this subject preference in the One-Object-Pronoun vs. Two- Pronoun conditions, statistical analyses on the subject advantage scores were carried out using linear mixed models on nine 500 ms time-slices (-500 ms to 4000 ms). Note that no models converged during the negative time window from -500 ms to 0 ms. Thus, the time window from 22 The analysis windows were determined on the basis of the distribution of selection latencies for the last-mentioned referent – the mean selection latency relative to the indirect object onset was 3291.63ms in the One-Object-Pronoun condition, 3326.38ms in the Two-Pronoun condition, and 2156.51ms in the One-Subject-Pronoun condition. Since eye-tracking data were stopped recording right after the last-referent selection, about 75.7% of all trials had already finished recording the eye-tracking samples 4000ms after the onset of the indirect object. Thus, the 3500 ms to 4000 ms time-slice was chosen as the last analysis window. 168 -400 ms to 0 ms was included in the analyses, instead. A summary of the fixed effects of the models is given in Table 23. Timing Fixed effect Estimate Std.Error df t value Pr(>|t|) -400 – 0ms (Intercept) -0.012572 0.006346 17.448991 -1.981 0.0636 . Ref. structure -0.002761 0.012144 941.929497 -0.227 0.8202 0 – 500ms (Intercept) -0.017871 0.007670 24.871448 -2.330 0.0282 * Ref. structure -0.006413 0.012436 930.436413 -0.516 0.6062 500 - 1000ms (Intercept) -0.003179 0.009973 28.026181 -0.319 0.7523 Ref. structure 0.032794 0.015961 927.520920 2.055 0.0402 * 1000- 1500ms (Intercept) 0.01608 0.01518 33.60102 1.060 0.297 Ref. structure 0.02351 0.02136 918.27989 1.101 0.271 1500- 2000ms (Intercept) 1.497e-02 1.640e-02 6.683e+01 0.913 0.365 Ref. structure 9.304e-03 2.359e-02 8.597e+02 0.394 0.693 2000- 2500ms (Intercept) -0.01911 0.02382 146.62519 -0.802 0.424 Ref. structure 0.03071 0.02839 754.88281 1.082 0.280 2500- 3000ms (Intercept) 0.003108 0.023269 30.651733 0.134 0.895 Ref. structure 0.031582 0.036943 53.438925 0.855 0.396 3000- 3500ms (Intercept) -0.003586 0.019195 21.038271 -0.187 0.854 Ref. structure 0.020894 0.044611 47.458715 0.468 0.642 3500- 4000ms (Intercept) 0.01918 0.02064 45.39079 0.929 0.358 Ref. structure 0.01820 0.04065 386.20274 0.448 0.655 Table 23. Summary of fixed effects for the models on the subject advantage scores in eight 500 ms time-slices and one 400 ms time-slice aligned to the onset of the indirect object (significant effects are shaded) A significant effect of referential structure – i.e., more looks to the subject than the object in the Two-Pronoun condition compared to the One-Object-Pronoun condition – was only found 169 during the time window from 500 ms to 1000 ms. However, the gaze patterns visualized in Figure 30 and Figure 31 showed an overarching subject preference, starting from around 500 ms until 3100 ms after the onset of the indirect object – the subject preference was stronger in the Two- Pronoun condition than the One-Object-Pronoun condition. Because the lack of the significant effects in the time slices after 1000 ms might be due to insufficient samples in a window relative to the effect size, I created a bigger time window from 1000 ms to 3000 ms and conducted the same statistical analysis. The fixed effects of the models are summarized in Table 24. Timing Fixed effect Estimate Std.Error df t value Pr(>|t|) -400 – 0ms (Intercept) -0.012572 0.006346 17.448991 -1.981 0.0636 . Ref. structure -0.002761 0.012144 941.929497 -0.227 0.8202 0 – 500ms (Intercept) -0.017871 0.007670 24.871448 -2.330 0.0282 * Ref. structure -0.006413 0.012436 930.436413 -0.516 0.6062 500 - 1000ms (Intercept) -0.003179 0.009973 28.026181 -0.319 0.7523 Ref. structure 0.032794 0.015961 927.520920 2.055 0.0402 * 1000- 3000ms (Intercept) 0.01686 0.01471 42.85110 1.146 0.258 Ref. structure 0.03846 0.01842 920.02021 2.089 0.037 * 3000- 3500ms (Intercept) -0.003586 0.019195 21.038271 -0.187 0.854 Ref. structure 0.020894 0.044611 47.458715 0.468 0.642 3500- 4000ms (Intercept) 0.01918 0.02064 45.39079 0.929 0.358 Ref. structure 0.01820 0.04065 386.20274 0.448 0.655 Table 24. Summary of fixed effects for the models on the subject advantage scores in four 500 ms time-slices and one 2000 ms time-slice aligned to the onset of the indirect object (significant effects are shaded) This time, a significant effect of referential structure was observed during the time window from 1000 ms to 3000 ms. This indicates that object pronouns in Two-Pronoun conditions exhibit 170 a stronger subject preference than in One-Object-Pronoun condition. This effect begins to emerge around 500 ms after pronoun onset and persists until 3000 ms. The gaze patterns during 500 ms- 3000 ms match the patterns found in the last-mentioned selection data (Figure 26), where the subject-antecedent preference was stronger in the Two-Pronoun than the One-Object-Pronoun condition. Thus, these findings imply that antecedent recency plays a significant role in guiding pronoun interpretation in the real-time processing as well. In sum, the result support the Immediate Activation Hypothesis, indicating that referential structure effects found in the offline click data can be also detected during real-time processing early on. 5.6. Discussion Experiment 6 investigated how our mental models of referential dependencies are dynamically updated as co-referential cues become available and whether the different patterns of pronoun resolution by referential structure type can be detected during real-time processing. From the perspective of incremental processing, the crucial difference between single pronoun resolution (e.g., One-Object-Pronoun condition) and Two-Pronoun resolution was assumed to be the presence vs. absence of a preceding (subject) pronoun’s dependency information when the comprehender encountered an object pronoun later in the same sentence. Although the offline studies in the previous chapters provided supporting evidence that the interactions between referential dependencies initiated from the preceding pronoun and the subsequent pronoun are key to referential structure effects, these studies were not able to answer at what moment the preceding dependency information exerts its influence on pronoun interpretation. Regarding this timing issue, I formulated two competing hypotheses: the Immediate Activation Hypothesis assumes that when the subsequent pronoun is encountered, its potential 171 antecedents are immediately retrieved with the dependencies linked to the preceding pronoun; on the other hand, the Delayed Activation Hypothesis assumes that the retrieval of the preceding pronoun’s referential dependencies is delayed until the initial reference resolution of the subsequent pronoun. Thus, the Immediate Activation Hypothesis predicts that referential structure effects (One-Object-Pronoun vs. Two-Pronoun) will emerge early on upon hearing the subsequent pronoun; whereas the Delayed Activation Hypothesis predicts that the emergence of referential structure effects may not be immediately noticeable upon hearing the subsequent pronoun. To test these hypotheses, I conducted a webcam-based visual-world eye-tracking study by manipulating the referential structure type in three ways: a structure with pronouns in the subject and the indirect object position (Two-Pronoun condition) to a structure with only one pronoun in either subject position (One-Subject-Pronoun condition) or in the indirect object position (One- Object-Pronoun condition). 5.6.1. Immediate activation of preceding pronoun’s referential dependencies Experiment 6 was primarily designed to detect effects of referential structure during indirect object pronoun resolution. Overall analysis of the eye gaze patterns at the indirect object pronoun (One- Object-Pronoun vs. Two-Pronoun condition) supported the Immediate Activation Hypothesis – referential structure effects emerged early on during indirect object pronoun resolution. Specifically, a significant effect of referential structure began to surface around 500 ms after the indirect object pronoun onset and persisted for 3000 ms. This finding supports the idea that when a comprehender encounters the indirect object pronoun, its potential antecedents are immediately retrieved with the dependencies linked to the preceding subject pronoun and the information about the preceding referential dependencies is actively used to interpret the pronouns. If the potential 172 antecedents were activated without being attached to any preceding referential dependencies, the Two-Pronoun sentences should have shown (at least initially) the same eye gaze patterns as the One-Object-Pronoun sentences. However, the fact that different gaze patterns were observed very early on after the onset of the indirect object pronoun provides crucial evidence that the preceding dependency information is immediately utilized and integrated into forming the current object pronominal dependency as well as revising the preceding subject pronominal dependency. Before continuing, it’s worth considering in more detail the finding that the referential structure effects began to emerge not immediately at the onset of the indirect object pronoun (0 ms) but rather 500 ms after its onset. At first glance, this could appear to weaken my conclusion that the results support the Immediate Activation Hypothesis. However, it’s important to note that other relative timing properties can help us to interpret the timing of these referential structure effects – in particular, other looking patterns also emerge relatively late in the eye-tracking record, presumably due to the timing properties of webcam-based eye-tracking. For example, after hearing the subject of the because-clause (e.g., …after the winter vacation because he/Hanna had…), the peak in looks to the first look-away object (e.g., after the winter vacation) appears late, around the end of the following subject noun (e,g, he/Hanna). Furthermore, the peak in looks to the referent of the new name in subject position (e.g., Hanna) does not appear until approximately 500ms after the subject offset (see Figure 29). These two referents are both unambiguous, and the fact that the peaks in looks to them also occur rather late suggests that overall we are probably dealing with an across-the-board timing delay regarding when looks are being recorded in the eye-tracking record. Thus, in light of the observation that these other looking patterns emerge relatively late, I would like to tentatively suggest that once we compare the timing of the referential structure effects to other timing properties, it is not unreasonable to assume that the referential structure effects in 173 fact may be starting almost immediately after indirect object pronoun resolution. In other words, it seems that once we adjust for the timing properties of webcam-based eye-tracking, the effects of referential structure may not be as ‘delayed’ as one might initially think. However, further work is needed to confirm this. Let me elaborate on how the effects of referential structure occur during the indirect object pronoun interpretation. The subject-antecedent preference was found to be significantly stronger in the Two-Pronoun condition than the One-Object-Pronoun condition, and this difference began to emerge 500 ms after the indirect object pronoun onset. The patterns of referential structure effect suggest that antecedent recency is a primary factor that guide pronoun interpretation (other factors might contribute to referential structure effects as well, see section 5.5.2) – the competition for the most recently mentioned antecedent (object antecedent) is expected in the Two-Pronoun condition but not in the One-Object-Pronoun condition. When resolving referential dependencies for the indirect object pronoun in a Two-Pronoun sentence, it will be immediately taken into consideration that the referential dependencies of the subject pronoun are linked to the subject and object antecedents. As soon as the process of indirect object pronoun resolution starts, the retrieval cues such as recency come into play to resolve the indirect object pronoun and simultaneously to reanalyze the initial dependencies of the preceding subject pronoun. This results in recency driving a preference for the most recently mentioned item (object antecedent) not only for the indirect object pronoun but also for the subject pronoun, leading to competition for the object antecedent at the very early stage of the resolution process. 23 However, there is no existing pronominal 23 It could also be the case that the referential dependencies of the subject pronoun were initially more strongly linked to the object antecedent than to the subject antecedent due to recency, which could hinder the establishment of the referential dependency between the object antecedent and the indirect object pronoun. However, further research is necessary to answer this. 174 dependency that interferes the connection to the object antecedent from the indirect object pronoun when there is only one pronoun in a clause (One-Object-Pronoun structure). Thus, Experiment 6 reveals effects of referential structure early on during indirect object resolution in the form of the subject-antecedent preference that is stronger in the Two-Pronoun condition. In regard to the time windows for the statistical analyses, gaze estimates were basically binned into 500 ms windows. When analyzed by windows of 500 ms, referential structure effect reached significance only during the time window from 500 ms to 1000 ms. The gaze estimates from 1000 ms to 3000 ms were found to have a significant effect of the referential structure, when binned into a big single time window. The fact that the windows of 500 ms failed to capture the significant differences might be due to insufficient samples resulted from the low sample rates of webcam-based eye-tracking. Given that the visualization of the gaze patterns showed an overarching subject preference in the Two-Pronoun condition, starting from around 500 ms until 3100 ms; with additional power, referential structure effects might appear by windows of 500 ms in all time slices. Thus, the results of Experiment 6 support the Immediate Activation Hypothesis, which states that the preceding pronoun’s dependency information becomes available as soon as a comprehender encounters a subsequent pronoun in the same clause. It is worth noting why the preceding pronoun’s dependency information is used so rapidly. Further studies are necessary to fully understand this, but one possible explanation lies in the high storage costs in our memory. Due to limited space in our memory, keep dependencies open during processing increases storage costs. To reduce these costs, the parser may be eager to complete the open dependencies as soon as possible. In the current study, it would not have been economical to delay reanalysis of the initial subject pronoun’s dependencies and keep the dependencies 175 incomplete. Similar patterns have been found in cataphora resolution (a pronoun is followed by potential antecedents), where a pronoun has a tendency to establish referential dependencies with the closest referent that follows it (e.g., Fedele, 2016; Fedele & Kaiser, 2014; Kazanina et al., 2007; VanGompel &Liversedge, 2003). Therefore, I speculate that high processing costs may have driven the immediate activation of the preceding dependencies in this study. In summary, the findings from Experiment 6 provide novel evidence for the nature of referential structure effects during real-time processing – the preceding pronoun’s dependency information immediately participates in the process of pronoun resolution. 5.6.2. Recency effects on pronoun resolution Experiment 6 was not primarily intended to explore effects of recency but to explore the real-time processing of referential structure effects. Unlike the target sentence structure of the previous offline studies in this dissertation (Experiment 1 to 5), Experiment 6 used a highly ambiguous sentence structure where the two clauses are structurally non-parallel with increased distance between the potential antecedents and pronouns, to conduct a visual-world eye-tracking study. Previous work has shown that (i) antecedent recency guides reference resolution in contexts where potential antecedents are separated from the pronoun by intervening sentences or where pronoun interpretation is highly ambiguous (e.g., Arnold, 1998; Clark & Sengul, 1979; Cunnings et al., 2014; Ehrlich & Rayner, 1983; Hall, 2021; Streb et al., 2004) and that (ii) different factors can have varying effects from one context to another during pronoun resolution. In light of these previous findings, I explored the possibility that recency could also play a role in referential structure effects as well as a discourse coherence factor (Centering Theory) while syntactic parallelism would not play a role in Experiment 6. 176 The overall results of the last-mentioned selection task (where participants click on the picture of whatever they thought was mentioned last in the sentence they heard) are compatible with a recency effect because subject-antecedent preference was stronger in the Two-Pronoun condition relative to the One-Object-Pronoun condition. Moreover, during the indirect object pronoun resolution, eye-gaze patterns showed that a proportion of looks towards the object was higher than that of looks to the subject in the One-Object-Pronoun condition compared to the Two- Pronoun condition. These results indicate that the factor of recency guides pronoun resolution and contributes to eliciting the different patterns of pronoun interpretation depending on referential structure type. However, the results should not be taken to mean that the recency is the only factor that contributes to referential structure effects under the current context. Taking the last-mentioned selection result of the One-Object-Pronoun condition into consideration in light of a recency effect (although the principal domain of analysis in this study is finding relative differences in pronoun interpretation between the different referential structures), the selection result seems to go against a recency effect because there was no strong object-antecedent preference. However, it can be interpreted as evidence that the factors other than recency came into play together. One possible explanation for the outcome is that the discourse coherence factor (Centering Theory) exerted its influence on pronoun interpretation because Centering Theory predicts the same type of transition status (RETAIN) no matter which antecedent is referred to by the indirect object pronoun in the One-Object-Pronoun condition. Thus, the effect of discourse coherence might have exerted its influence in such a way that weakens a recency effect. In sum, referential structure effects found in Experiment 6 provides further evidence that different factors weigh in differently from one context to another during pronoun resolution and 177 suggest that referential structure effects may surface in different ways depending on context. More importantly, the fact that the convergent evidence on the effects of referential structure were found in different contexts with multiple methods provides strong evidence that referential structure effects can be generalized beyond the particular contexts or method used in this dissertation. 178 Chapter 6: Conclusion 6.1. Overview and summary In this dissertation I have aimed to shed light on the mechanisms underlying multiple pronoun resolution. I propose that the crucial difference between multiple pronoun resolution and single pronoun resolution stems from differences in referential structure – in other words, whether all or only one of the referents in the preceding clause are mentioned with a pronoun in the subsequent clause. I present a series of experiments exploring whether differences in referential structure can influence pronoun interpretation. This dissertation focused on four primary questions: (i) the first question asked whether and how the process of reference resolution of multiple pronouns is different from that of a single pronoun. (ii) The second question focuses on what factors, at different levels of representation, contribute to effects of referential structure (different coreferential mechanism between multiple pronouns and a single pronoun). (iii) The third question looks at whether and how referential structure effects are influenced by differences in anaphoric form, such as null pronouns (pro) in Korean vs. overt pronouns in English. (iv) The last question focuses on the real-time comprehension of multiple pronouns during incremental processing using webcam-based visual- world eye-tracking. In regard to the first question, Experiments 1, 2 and 3 tested two hypotheses regarding potential effects of referential structure: the Independence View and the Dependence View. According to the Independence View, subject-position pronoun interpretation is not sensitive to differences in referential structure type (One-Pronoun vs. Two-Pronoun structure) because the anaphoric dependencies constructed for the different pronouns are resolved fully independently in 179 line with the PFS (Parallel Function Strategy). On the contrary, the Dependence View – which is comprised of three subtypes based on the influence of syntactic and/or discourse factors on referential structure effects (Syntax-level, Discourse-level, and Syntax-Discourse Dependence View) predicts that subject pronouns in the One-Pronoun sentences would be more likely to be interpreted as referring to a preceding object than subject pronouns in the Two-Pronoun sentences, because of interactions between multiple referential dependencies. The results from three offline experiments (Experiments 1 to 3, using a picture-writing task) showed that pronoun interpretation is indeed sensitive to referential structure type. These findings support the Dependence Views: the multiple-pronoun resolution process is fundamentally different from single-pronoun resolution due to interactions between the anaphoric dependencies of the two different pronouns. Experiment 4 further investigated what factors at different levels of representation contribute to effects of referential structure by comparing three types of the Dependence View (Syntax-level, Discourse-level, and Syntax-Discourse Dependence View). The Syntax-level Dependence View posits that referential structure effects arise over the course of searching for morpho-syntactically matching antecedents (i.e., cue-based Parallel Function Strategy). In contrast, according to the Discourse-level Dependence View, referential structure effects arise as a consequence of maximizing the coherence of transitions between utterances in discourse (i.e., Centering Theory). The Syntax-Discourse Dependence View hypothesizes that both syntactic and discourse factors affect the interaction of coreferential operations. The results from an offline picture-writing task support the Syntax-Discourse Dependence View. The results suggest that both syntax-level cue-based parallelism and a discourse-level coherence mechanism (Centering Theory) guide pronoun resolution by contributing to the interactions between anaphoric dependencies of the two different pronouns. 180 Experiment 5 broadened the domain of referential structure effects beyond overt pronouns in English by turning to another type of anaphoric form: namely, null pronouns (pro) in Korean (the third question). Given that the multiple factors that guide reference resolution can carry different weight for different types of anaphoric form (Kaiser and Trueswell’s (2008) Form- Specific Account) and the more reduced anaphoric form has the more topical referent (e.g., Givón’s (1983) Topic-Continuity Approach), Experiment 5 tested whether the discourse-level coherence factor (Centering Theory) exerts a stronger influence on referential structure effects with Korean null pronouns, compared to the syntax-level factor (cue-based Parallel Function Strategy). The results from an offline picture-writing task on Korean indicate that both syntactic and discourse factors contribute to referential structure effects, but crucially, my results suggest that the discourse factor exerts a stronger influence on null pronoun resolution in Korean than on overt pronoun resolution in English. Finally, Experiment 6 shifted from examining referential structure effects through the lens of end-state-focused offline tasks – such as the picture-writing task – to examining them from the perspective of real-time processing, using a novel webcam-based visual-world eye-tracking method (e.g., Papoutsaki et al., 2016). This study tested whether the patterns of pronoun resolution differ depending on referential structure type during real-time processing. This allowed me to explore how rapidly preceding dependency information is utilized and integrated into forming the current pronominal dependency as well as revising the preceding pronominal dependency. I formulated two competing hypotheses with regard to the timing of referential structure effects: the Immediate Activation Hypothesis assumes that when the subsequent pronoun is encountered, its potential antecedents are immediately retrieved with the dependencies linked to the preceding pronoun; on the other hand, the Delayed Activation Hypothesis assumes that the retrieval of the 181 preceding pronoun’s referential dependencies is delayed until the initial reference resolution of the subsequent pronoun. The results showed early emergence of referential structure effects during resolution of the subsequent pronoun, supporting the Immediate Activation Hypothesis. Thus, Experiment 6 suggests that the preceding pronoun’s dependency information becomes available as soon as a comprehender encounters the subsequent pronoun in the same sentence. 6.2. Various patterns of effects of referential structure Effects of referential structure were observed in all of the experiments reported in this dissertation. Thus, the results show that subject position pronouns in sentences that only mention one of the antecedents from the preceding clause (One-Pronoun structure) are interpreted differently than subject position pronouns in sentences that also include a subsequent object-position pronoun (Two-Pronoun structure). These novel findings suggest that resolving one pronominal dependency can influences the formation of the other pronominal dependency. Thus, this work goes beyond most prior experiments on pronoun resolution which – with the exception of work on parallelism effects (e.g., Chambers & Smyth 1998; Stevenson et al., 1993, 1995) – has tended to focus only on subject-position pronouns and has typically not systematically investigated the referential properties of the rest of the clause. The results I report in this dissertation suggest that focusing only on subject-position pronouns without regard for what happens later in the sentence can result in important patterns being overlooked. Furthermore, my results also show that referential structure effects do not always surface in an identical way in all the experiment. Rather, the studies reported here show that the context in which the pronouns appear as well as the linguistic forms of the pronouns (overt pronouns vs. null pronouns) play a role: 182 In Experiments 1 to 4, referential structure effects with overt pronouns in English were tested in a context where (i) the antecedent-containing matrix clause and the pronoun-containing embedded clause were structurally parallel (both contain transitive verbs) and where (ii) the potential antecedents were placed close to the pronoun(s) without additional intervening phrases, such as adverbial phrases. In this context, my results suggest that effects of referential structure are influenced more influenced by syntax-level parallelism than by discourse-level coherence. In Experiment 5, testing for referential structure effects with null pronouns in Korean, I again used two syntactically parallel transitive clauses. In this kind of structurally parallel context, the overall patterns of referential structure effects in Experiment 5 were similar to those found in Experiments 1 to 4: the syntax-level parallelism factor exerted more influence on referential structure effects than the discourse-level coherence factor. However, the results also suggest that relatively speaking, discourse-level coherence had a stronger effect on null pronouns in Korean than overt pronouns in English. After using syntactically parallel clauses in Experiments 1 to 5, in Experiment 6 – using web-cam-based eye-tracking on English – I changed to using a less structurally parallel and more referentially ambiguous configuration, with increased distance between the potential antecedents and pronouns. Under this ambiguous and non-parallel context, referential structure effects were found to be predominantly affected by antecedent recency, not syntactic parallelism. In sum, these findings indicate that the weights of different factors can change from one context to another during pronoun resolution. In a structurally parallel context, the syntactic parallelism factor has a relatively large impact on the process of pronoun resolution – and crucially, this happens even in contexts where the clauses are connected with because and thus do not involve semantic parallelism. However, when the clauses are not strictly structurally parallel (and where 183 potential antecedents were separated from the pronoun by intervening phrases), recency exerts a relatively large effect on the effects of referential structure. Thus, different kinds of referential structure effects emerge depending on the contextual circumstances that different factors operate in with different weights. 6.3. Implications for models of pronoun resolution An important contribution of this dissertation is that it highlights the role of referential structure by looking ‘forward’ towards another pronoun in the same clause. Most existing psycholinguistic approaches to pronoun resolution take a ‘backward-looking’ approach by focusing on what factors influence (a) the salience of potential antecedents in the preceding clause and thus also (b) the likelihood that a pronoun will be interpreted as referring to one of these antecedents (e.g., Caramazza et al., 1977; Gernsbacher & Hargreaves, 1988; Gernsbacher, 1989; Givon, 1983; Gundel, Hedberg, & Zacharski, 1993; Rudolph & Forsterling, 1997). Generally speaking, only Centering Theory (e.g., Grosz et al., 1995; Walker et al., 1998) and coherence-based approaches have taken a more holistic approach that looks at the clause as a whole (e.g., Hobbs, 1979; Kaiser, 2009; Kehler, 2002; Kehler et al., 2008; Winograd, 1972) The findings presented in this dissertation suggest that it is not sufficient for a satisfactory model of pronoun resolution to rely only on preceding information, as suggested by the pioneering work of Kaiser (2009). The effects of referential structure reported in this dissertation show that pronoun interpretation is guided not only by the properties related to the antecedents but also by the presence (or absence) of another pronoun in the same clause. There are interactions between multiple pronominal dependencies, where how one of the pronominal dependencies is resolved interactively influences the formation of the other dependency. In other words, if we assume that, 184 upon encountering a pronoun (e.g., a subject-position pronoun), the language processing system activates a weighted set of potential antecedents whose weights are influenced by information in the preceding clause, these weights can change after the processing system encounters another pronoun in the same clause (e.g., object-position pronoun). Ultimately, the interpretation of both pronouns can be influenced by the presence of the other. Taken together, this dissertation research takes first steps towards the ultimate aim of arriving at a comprehensive model of pronoun resolution by emphasizing the importance of the forward-looking approach and, in particular, the impact of referential structure effects. 6.4 Future directions The studies reported in this dissertation open a number of avenues for future research. One such avenue is to further investigate effects of referential structure using different types of anaphoric forms. Although this dissertation has focused on pronominal forms (overt pronouns in English and null pronouns in Korean), there are a variety of anaphoric forms cross-linguistically that can build referential dependencies with preceding entities. For example, languages like Finnish, German and Dutch also use anaphoric demonstratives for human antecedents (e.g., Bosch & Umbach, 2007; Comrie, 1997; Kaiser, 2003). Simplifying somewhat, these anaphoric demonstratives typically refer to non-topical entities while the personal pronouns refer to more topical entities. By investigating potential referential structure effects involving anaphoric demonstratives, which have different information structural properties from pronouns, we can gain important insights that can help us to better understand the underlying mechanism of interactions between multiple referential dependencies. Another avenue for future research would be to explore the effects of referential structure 185 during real-time processing in terms of processing load. Although Experiment 6 provided temporal information about when the preceding pronominal dependency is utilized in multiple pronoun resolution, it still remains an open question whether or not multiple pronoun resolution requires a higher processing cost compared to single pronoun resolution. In ambiguous contexts where two potential antecedents are present, it might be the case that interpreting two pronouns in the same clause incurs more processing cost compared to single pronoun interpretation because the preceding pronominal dependencies are immediately utilized into building referential dependencies with the subsequent pronoun, according to which the preceding pronominal dependencies are reanalyzed. Then again, it could also be the case that two-pronoun resolution is less costly than single pronoun resolution because the set of antecedents is constrained by the dependency involving the preceding pronoun – in other words, the preceding pronominal dependency may reduce the potential interpretational options for the subsequent pronoun. By further exploring these kinds of questions, future work can help us to gain better insight on the relationship between referential structure effects and our limited processing resources. 6.5. Final conclusion This dissertation explores the mechanisms that underlie pronoun resolution, by investigating the factors that guide the interpretation of multiple pronouns in a clause and how this differs from single pronoun interpretation. Through a series of offline and online studies, this dissertation provides novel evidence that multiple pronoun resolution is different from single pronoun resolution and that these differences stem from differences in referential structure – resolving referential dependency of one pronoun can influence on resolving dependency of the other pronoun. Thus, the findings of this dissertation suggest that a complete understanding of pronoun 186 interpretation must take referential structure into consideration. 187 References Ahn, H.-D., & Cho, S. (2010). More on the absence of CP ellipsis. Studies in Generative Grammar, 20, 137–148. Ahn, H.-D., & Cho, S. (2012). On null arguments and clausal ellipsis in Korean. Language and Linguistics, 57, 95-124. Ahn, H. D., & Cho, S. (2020). A pronoun analysis of null arguments in Korean. Language Research, 56, 193-223. Altmann, G. T. M., & Kamide, Y . (2007). The real-time mediation of visual attention by language and world knowledge: Linking anticipatory (and other) eye movements to linguistic processing. Journal of Memory and Language, 57(4), 502-518. doi:10.1016/j.jml.2006.12.004 Almor, A., de Carvalho Maia, J., Lima, M. L. C., Vernice, M., & Gelormini-Lezama, C. (2017). Language processing, acceptability, and statistical distribution: A study of null and overt subjects in Brazilian Portuguese. Journal of Memory and Language, 92, 98-113. Alonso-Ovalle, L., Fernandes-Solera, S., Frazier, L., & Clifton, C. (2002). Null vs. overt pronouns and the Topic-Focus articulation in Spanish. Rivista di Linguistica, 14(2), 1–19. Ariel, M. (1990). Assessing noun-phrase antecedents. London: Routledge. Ariel, M. (1994). Interpreting anaphoric expressions: A cognitive versus a pragmatic approach. Journal of Linguistics, 30, 3–42. Arnold, J. (1998). Reference form and discourse patterns. Doctoral dissertation, Stanford University. Arnold, J. E., Eisenband, J. G., Brown-Schmidt, S., & Trueswell, J. C. (2000). The rapid use of gender information: Evidence of the time course of pronoun resolution from eyetracking. Cognition, 76(1), B13-B26. Arnold, J. E., & Griffin, Z. M. (2007). The effect of additional characters on choice of referring expression: Everyone counts. Journal of Memory and Language, 56(4), 521-536. Au, T.K. (1986). A verb is worth a thousand words: The causes and consequences of interpersonal events implicit in language. Journal of Memory and Language, 25, 104-122. Bhatia, S., & Dillon, B. (2022). Processing agreement in Hindi: When agreement feeds attraction. Journal of Memory and Language, 125, 104322. Baayen, R. H., Davidson, D. J., & Bates, D. M. (2008). Mixed-effects modeling with crossed 188 random effects for subjects and items. Journal of memory and language, 59(4), 390-412. Baayen, R. H., & Milin, P. (2010). Analyzing reaction times. International Journal of Psychological Research, 3(2), 12-28. doi:10.21500/20112084.807 Badecker, W., & Straub, K. (2002). The processing role of structural constraints on interpretation of pronouns and anaphors. Journal of Experimental Psychology: Learning, Memory, and Cognition, 28(4), 748–769 Bak, J. (2015). An Research on the 3rd Person Pronoun ‘ku’. Hanminjok Emunhak, 71, 104-128. Bates, D., Maechler, M., Bolker, B., & Walker, S. (2015). Fitting Linear Mixed-Effects Models Using lme4. Journal of Statistical Software, 67(1), 1-48. doi:10.18637/jss.v067.i01 Beaver, D. I. (2004). The optimization of discourse anaphora. Linguistics and philosophy, 27, 3- 56. Bock, J. K., & Irwin, D. E. (1980). Syntactic effects of information availability in sentence production. Journal of Verbal Learning and Verbal Behavior, 19(4), 467-484. doi:10.1016/S0022-5371(80)90321-7 Bock, J.K. (1986). Syntactic persistence in language production. Cognitive Psychology, 18, 355- 387. Bosch, P., & Umbach, C. (2007). Reference determination for demonstrative pronouns. ZAS Papers in Linguistics, 48, 39-51. Bott, O., & Solstad, T. (2014). From verbs to discourse: A novel account of implicit causality. In Psycholinguistic approaches to meaning and understanding across languages (pp. 213-251). Brennan, S. E., Walker Friedman, M., & Pollard, C. J. (1987). A Centering approach to pronouns. Proceedings of the 25th Annual Meeting of the Association for Computational Linguistics (ACL), 155-162. Burmester, J., Sauermann, A., Spalek, K., & Wartenburger, I. (2018). Sensitivity to salience: Linguistic vs. visual cues affect sentence processing and pronoun resolution. Language, Cognition and Neuroscience, 33(6), 784-801. Burmester, J., Spalek, K., & Wartenburger, I. (2014). Context updating during sentence comprehension: The effect of aboutness topic. Brain and language, 137, 62-76. Burnsky, J., Keshev, M., Asatryan, M., Hlachova, B., Johnson, K., & Dillon, B. (2022). Look Away! An Object is Coming [Poster presentation]. The 35th Annual Conference on Human Sentence Processing, UC Santa Cruz, USA. Carminati, M.N. (2002). The processing of Italian subject pronouns. PhD dissertation, University of Massachusetts, Amherst. 189 Caramazza, A., Grober, E., Garvey, C., & Yates, J. (1977). Comprehension of anaphoric pronouns. Journal of Verbal Learning and Verbal Behaviour, 16, 601-609. Chafe, W. L. (1974). Language and consciousness. Language, 50(1), 111-133. Chambers, C. G., & Smyth, R. (1998). Structural parallelism and discourse coherence: A test of centering theory. Journal of Memory and Language, 39(4), 593-608. Chien, Y . C., & Wexler, K. (1990). Children's knowledge of locality conditions in binding as evidence for the modularity of syntax and pragmatics. Language acquisition, 1(3), 225- 295. Chmielewski, M., & Kucker, S. C. (2020). An MTurk crisis? Shifts in data quality and the impact on study results. Social Psychological and Personality Science, 11(4), 464-473. doi:10.1177/1948550619875149 Clark, H. H. & Clark, E. V . (1977). Psychology and language: An introduction to psycholinguistics. New York: Harcourt Brace Jovanovich. Clark, S. E., & Gronlund, S. D. (1996). Global matching models of recognition memory: How the models match the data. Psychonomic Bulletin & Review, 3(1), 37–60. Cole, P. (1987). Null objects in Universal Grammar. Linguistic Inquiry 18, 597–612. Colonna, S., Schimke, S., & Hemforth, B. (2010). Le rôle de la structure informationnelle dans l’interprétation d’une anaphore pronominale inter-phrastique. In F. Neveu, V . Muni Toke, J. Durand, T. Klingler, L. Mondada & S. Prévost (Eds.), Actes du Congrès Mondial de Linguistique Française (pp. 1489-1499). Paris, France. Comrie, B. (1997). Pragmatic binding: Demonstratives as anaphors in Dutch. In Annual Meeting of the Berkeley Linguistics Society (V ol. 23, No. 1, pp. 50-61). Cooper, R.M. (1974). Control of eye fixation by meaning of spoken language: New methodology for real-time investigation of speech perception, memory, and language processing. Cognitive Psychology, 6, 84-107. Cowles, H. W. (2007). The influence of'aboutness' on pronominal coreference. ZAS papers in linguistics, 48, 23-38. Cowles, H. W., & Ferreira, V . S. (2012). The influence of topic status on written and spoken sentence production. Discourse Processes, 49(1), 1-28. Cowles, H.W., Walenski, M., & Kluender, R. (2007). Linguistic and cognitive prominence in anaphor resolution: Topic, contrastive focus and pronouns. Topoi, 26, 3–18. Cozijn, R., Commandeur, E., V onk, W., & Noordman, L. G. (2011). The time course of the use of implicit causality information in the processing of pronouns: A visual world paradigm 190 study. Journal of Memory and Language, 64(4), 381-403. Crawley, A., Stevenson, J., & Kleinman, D. (1990). The use of Heuristic Strategies in the interpretation of pronouns. Journal of Psycholinguistic Research, 19, 245-264. Cunnings, I., Patterson, C., & Felser, C. (2014). Variable binding and coreference in sentence comprehension: Evidence from eye movements. Journal of Memory and Language, 71(1), 39-56 Degen, J., Kursat, L., & Leigh, D. D. (2021). Seeing is believing: Testing an explicit linking assumption for visual world eye-tracking in psycholinguistics. Proceedings of the Annual Meeting of the Cognitive Science Society, (43). Retrieved from https://escholarship.org/uc/item/6182t9jb de la Fuente, I. (2015). Putting pronoun resolution in context: The role of syntax, semantics, and pragmatics in pronoun interpretation. Ph.D. Dissertation. Universite Paris Diderot-Paris 7. Dowty, D. (1991). Thematic proto-roles and argument selection. Language, 67(3), 547-619. doi:10.1353/lan.1991.0021 Dillon, B. (2011). Structured access in sentence comprehension. Dillon, B. (2014). Syntactic memory in the comprehension of reflexive dependencies: an overview. Language and Linguistics Compass, 8(5), 171-187 Dillon, B., Mishler, A., Sloggett, S., & Phillips, C. (2013). Contrasting intrusion profiles for agreement and anaphora: Experimental and modeling evidence. Journal of Memory and Language, 69(2), 85-103 Dixon, R. M. (1979). Ergativity. Language, 55(1), 59-138. Philadelphia/Amsterdam: John Benjamins. Do, M. L., & Kaiser, E. (2019). Subjecthood and linear order in linguistic encoding: Evidence from the real-time production of wh-questions in English and Mandarin Chinese. Journal of Memory and Language, 105, 60-75. Eberhard, K., Spivey-Knowlton, M., Sedivy, J., & Tanenhaus, M. (1995). Eye movements as a window into real-time spoken language comprehension in natural contexts. Journal of Psycholinguistic Research, 24, 409–436. Ehrlich, K., & Rayner, K. (1983). Pronoun assignment and semantic integration during reading: Eye movements and immediacy of processing. Journal of verbal learning and verbal behavior, 22(1), 75-87 Ellert, M. (2013). Resolving ambiguous pronouns in a second language: A visual-world eye- tracking study with Dutch learners of German. International Review of Applied Linguistics in Language Teaching, 51(2), 171–197. 191 Fedele, E. (2016). Discourse level processing and pronoun interpretation. Doctoral dissertation, University of Southern California. Fedele, E., & Kaiser, E. (2014). Looking back and looking forward: Anaphora and cataphora in Italian. University of Pennsylvania Working Papers in Linguistics, 20(1), 10. Fedriani, C. (2014). 2 The Experiential Situation and Its Components. In Experiential Constructions in Latin (pp. 15-51). Brill. Ferreira, F. (1994). Choice of passive voice is affected by verb type and animacy. Journal of Memory and Language, 33, 715-736 Ferstl, E. C., Garnham, A., & Manouilidou, C. (2011). Implicit causality bias in English: A corpus of 300 verbs. Behavior Research Methods, 43(1), 124-135. Filiaci, F., Sorace, A., & Carreiras, M. (2013). Anaphoric biases of null and overt subjects in Italian and Spanish: a cross-linguistic comparison. Language, Cognition, and Neuroscience, 29, 825–843. Fillmore. C. J. (1970). The grammar of hitting and breaking. In R. Jacobs & P. Rosenbaum (Eds.), Readings in English transformational grammar (pp. 120-133). Waltham. MA: Ginn. Finegan, E. (1995). Subjectivity and subjectivisation: An introduction. In D. Stein & S. Wright (Eds.), Subjectivity and subjectivisation: Linguistic perspectives (pp. 1–15). Cambridge: Cambridge University Press Frana, I. (2008). The role of discourse prominence in the resolution of referential ambiguities: Evidence from co-reference in Italian. University of Massachusetts Occasional Papers in Linguistics 37: Semantics and Processing. Amherst: GLSA. Gelormini-Lezama, C., & Almor, A. (2011). Repeated names, overt pronouns, and null pronouns in Spanish. Language and cognitive processes, 26(3), 437-454. Gernsbacher, M. A., & Hargreaves, D. J. (1988). Accessing sentence participants: The advantage of first mention. Journal of memory and language, 27(6), 699-717. Gernsbacher, M. A., Hargreaves, D. J., & Beeman, M. (1989). Building and accessing clausal representations: The advantage of first mention versus the advantage of clause recency. Journal of memory and language, 28(6), 735-755. Gillund, G., & Shiffrin, R. M. (1984). A retrieval model for both recognition and recall. Psychological Review, 91(1), 1–67. Givón, T. (1983). Topic continuity in discourse: A quantitative cross-language study. Philadelphia: John Benjamins Publishing Company. Gordon, P. C., Grosz, B. J., & Gilliom, L. A. (1993). Pronouns, names, and the centering of 192 attention in discourse. Cognitive science, 17(3), 311-347. Gordon, P., & Scearce, K. (1995). Pronominalization and discourse coherence, discourse structure and pronoun interpretation. Memory and Cognition, 23, 313- 323. Gordon, P. C., Hendrick, R., & Levine, W. H. (2002). Memory-load interference in syntactic processing. Psychological Science, 13(5), 425–430. https://doi.org/10.1111/1467- 9280.00475 Gordon, P. C., Hendrick, R., Johnson, M., & Lee, Y . (2006). Similarity-based inference during language comprehension. Journal of Experimental Psychology: Learning, Memory, and Cognition, 32(6), 1304–1321. https://doi.org/10.1037/0278-7393.32.6.1304 Greene, S. B., & McKoon, G. (1995). Telling something we can’t know: Experimental approaches to verbs exhibiting implicit causality. Psychological Science, 6(5), 262-270. Grober, E. H., Beardsley, W., & Caramazza, A. (1978). Parallel Function Strategy in pronoun assignment. Cognition, 6(2), 117-133. Grosz, B. J., Joshi, A. K., & Weinstein, S. (1995). Centering:A framework for modeling the local coherence of discourse. Computational Linguistics, 21, 203–225. Gundel, J. K., Hedberg, N., & Zacharski, R. (1993). Cognitive status and the form of referring expressions in discourse. Language, 274-307. Hall, K. G. (2021). Relations and Structure: A View from Dependency Resolution (Doctoral dissertation, Northwestern University). Hammerly, C., Staub, A., & Dillon, B. (2019). The grammaticality asymmetry in agreement attraction reflects response bias: Experimental and modeling evidence. Cognitive psychology, 110, 70-104. Han, N. R. (2006). Korean zero pronouns: analysis and resolution. Doctoral dissertation, University of Pennsylvania. Hasegawa, N. (1985). On the so-called "zero pronouns" in Japanese, Linguistic Review 4, 289-341. Hartshorne, J. K., & Snedeker, J. (2013). Verb argument structure predicts implicit causality: The advantages of finer-grained semantics. Language and Cognitive Processes, 28(10), 1474- 1508. Hartshorne, J. K., O'Donnell, T. J., & Tenenbaum, J. B. (2015). The causes and consequences explicit in verbs. Language, cognition and neuroscience, 30(6), 716-734. Haviland, S. E., & Clark, H. H. (1974). What's new? Acquiring new information as a process in comprehension. Journal of verbal learning and verbal behavior, 13(5), 512-521. 193 Hendriks, P. (2004). Coherence relations, ellipsis and contrastive topics. Journal of Semantics, 21(2), 133-153. Hendriks, P., Banga, A., Van Rij, J., Cannizzaro, G., & Hoeks, J. (2011). Adults’ on-line comprehension of object pronouns in discourse. In Production-comprehension asymmetries in child language (pp. 193-216). De Gruyter Mouton. Hobbs, J. (1979). Coherence and coreference. Cognitive Science, 3, 67-90. Hobbs, J. (1990). Literature and cognition, lecture notes 21. Stanford, CA: CSLI. Hoji, H. (1998). Null object and sloppy identity in Japanese. Linguistic Inquiry, 29, 127-152. Hopper, P. J., & Thompson, S. A. (1980). Transitivity in grammar and discourse. language, 251- 299. Horst, J. S., & Hout, M. C. (2016). The Novel Object and Unusual Name (NOUN) Database: A collection of novel images for use in experimental research. Behavior research methods, 48, 1393-1409. Huang, C.-T. J. (1984). On the distribution and reference of empty pronouns, Linguistic Inquiry 15, 531–574. Huang, C.-T. J. (1989). Pro-Drop in Chinese. The Null subject parameter. Jaeggli, O. and K. Safir, eds. Kluwer Academic Publishers: Dordrecht. Huang, C.-T. J. (1991). Movement structures and bounding theory. Manuscript. University of California, Irvine. Hudson, S. B., Tanenhaus, M. K., & Dell, G. S. (1986, August). The effect of the discourse center on the local coherence of a discourse. In Proceedings of the 8th Annual Meeting of the Cognitive Science Society (pp. 96-101). Hwang, H., & Kaiser, E. (2014). The role of the verb in grammatical function assignment in English and Korean. Journal of Experimental Psychology: Learning, Memory, and Cognition, 40(5), 1363. Jackendoff, R. (1972). Semantic Interpretation in Generative Grammar. Cambridge, MA: The MIT Press Jackendoff, R. (1990). Semantic Structures. MIT Press, Cambridge, Massachusetts. Jäger, L. A., Engelmann, F., & Vasishth, S. (2017). Similarity-based interference in sentence comprehension: Literature review and Bayesian meta-analysis. Journal of Memory and Language, 94, 316-339. Kameyama, M. (1985). Zero anaphora: The case of Japanese. Unpublished Dissertation, Stanford 194 University. Kameyama, M. (1998). Intrasentential Centering: A case study. In M. Walker, A. Joshi & E. Prince (Eds.), Centering Theory in discourse (pp. 89-114). Oxford: Clarendon Press. Kaiser, E. M. K. (2003). The quest for a referent: A crosslinguistic look at reference resolution. Doctoral dissertation, University of Pennsylvania. Kaiser, E. (2009). Investigating effects of structural and information-structural factors on pronoun resolution. In M. Zimmermann and C. Féry (eds.), Information Structure from Different Perspectives. NY: Oxford University Press, 332-353. Kaiser, E. (2011). Focusing on pronouns: Consequences of subjecthood, pronominalisation, and contrastive focus. Language and Cognitive Processes, 26(10), 1625-1666. Kaiser, E., & Trueswell, J.C. (2008). Interpreting pronouns and demonstratives in Finnish: Evidence for a form-specific approach to reference resolution. Language and Cognitive Processes, 23(5), 709–748. Kehler, A. (2002). Coherence, reference, and the theory of grammar. Stanford: CSLI Publications. Kehler, A. (2008). Rethinking the SMASH approach to pronoun interpretation. Reference. Interdisciplinary Perspectives, 95-122. Kehler, A., Kertz, L., Rohde, H., & Elman, J. (2008). Coherence and Coreference Revisited. Journal of Semantics, 25, 1-44. Kehler, A., & Rohde, H. (2013). A probabilistic reconciliation of coherence-driven and centering- driven theories of pronoun interpretation. Theoretical Linguistics, 39(1-2), 1-37. Kehler, A., & Rohde, H. (2017). Evaluating an expectation-driven question-under-discussion model of discourse interpretation. Discourse Processes, 54(3), 219-238. doi:10.1080/0163853X.2016.1169069 Kim, H. (2018). Cross-Linguistic Activation in Korean L2 Learners' Processing of Remention Bias in English (Doctoral dissertation, University of Hawai'i at Manoa). Kim, H., & Grüter, T. (2019). Cross-linguistic activation of implicit causality biases in Korean learners of English. Bilingualism: Language and Cognition, 22(3), 441-455. Kim, H. Y . (2009). Factors in the choice of referential forms in Korean discourse: Salience, speaker perspective and thematic importance. Korean Language in America, 14(1), 1-24. Kim, M. (2003). Zero vs. Overt NPs in Korean Discourse: A Centering Analysis. Korean Journal of Linguistics, 28-1, 29-49. Kim, Y . J. (2000). Subject/object drop in the acquisition of Korean: A cross-linguistic comparison. 195 Journal of East Asian Linguistics, 9, 325_351. Koornneef, A.W. & van Berkum, J.J.A. (2006). On the use of verb-based implicit causality in sentence comprehension: Evidence from self-paced reading and eye tracking. Journal of memory and language, 54, 445-465. Knuth, D. E. (1965). On the translation of languages from left to right. Information and Control, 8(6), 607–639. Kuroda, S.-Y . (1965) Generative Grammatical Studies in the Japanese Language, Doctoral dissertation, MIT. Kuznetsova, A., Brockhoff, P. B., & Christensen, R. H. B. (2017). lmerTest Package: Tests in Linear Mixed Effects Models. Journal of Statistical Software, 82, 1–26. doi:10.18637/jss.v082.i13 Kweon, S. O. (2011). Processing null and overt pronoun subject in ambiguous sentences in Korean. International Journal of Linguistics, 3(1), 1-13. Kwon, N., & Sturt, P. (2013). Null pronominal (pro) resolution in Korean, a discourse-oriented language. Language and Cognitive Processes, 28(3), 377-387. Lee, E. (2019). Korean syntax and semantics. Cambridge University Press. Lee, E.-J. (2005). The status of the null object in the VP ellipsis context. Studies in Generative Grammar 15:609–629. Lee, E., Park, M., and Madigan, S. (2015). An Introduction to Korean Linguistics. NewYork: Routledge Lee, H. B. (1987). On empty categories in Korean. Korean Journal of Linguistics, 12(2), 261-294 Lee, I., Lee, S., & Chae, W. (1997). Han-kwuk-uy en-e ( 한국의 언어). Seoul:sin-kwu-mwun- hwa-sa( 신구문화사). Lee, S. H. (2022). Processing the dynamicity of events in language. Doctoral dissertation, University of Southern California. Levin, B. (1993). English verb classes and alternations: A preliminary investigation. Chicago: University of Chicago Press. Lewis, R. L., & Vasishth, S. (2005). An activation-based model of sentence processing as skilled memory retrieval. Cognitive science, 29(3), 375-419 231 Lewis, R. L., Vasishth, S., & Van Dyke, J. A. (2006). Computational principles of working memory in sentence comprehension. Trends in Cognitive Sciences, 10(10), 447-454. doi: 196 10.1016/j.tics.2006.08.007 Li, C. and Thompson, S. (1976). Subject and topic: a new typology of language, in C. Li, ed., Subject and Topic, Academic Press, New York Lim, H. & Chang, S. (1995). kwuk-e-mwun-pep-lon ( 국어문법론). Seoul: Korea National Open University. Liversedge, S. P., & Van Gompel, R. P. G. ms. Resolving anaphoric and cataphoric pronouns McElree, B. (2000). Sentence comprehension is mediated by content-addressable memory structures. Journal of psycholinguistic research, 29(2), 111-123 McElree, B., & Dosher, B. A. (1993). Serial retrieval processes in the recovery of order information. Journal of Experimental Psychology: General, 122(3), 291–315. McElree, B., Foraker, S., & Dyer, L. (2003). Memory structures that subserve sentence comprehension. Journal of Memory and Language, 48(1), 67–91. McElree, B. (2006). Accessing recent events. Psychology of learning and motivation, 46, 155-200 McKee, C. (1992). A comparison of pronouns and anaphors in Italian and English acquisition. Language Acquisition 2:21–54. McKoon, G., Greene, S. B., & Ratcliff, R. (1993). Discourse models, pronoun resolution, and the implicit causality of verbs. Journal of Experimental Psychology: Learning, Memory, and Cognition, 19(5), 1040. Miltsakaki, E. (2002). Toward an aposynthesis of topic continuity and intrasentential anaphora. Computational Linguistics, 28(3), 319-355. Moon, G-S. (2010). Null arguments redux. The Linguistic Association of Korea Journal 18: 67-92. Oh, S. Y . (2010). Invoking categories through co-present person reference: The case of Korean conversation. Journal of Pragmatics, 42(5), 1219-1242. Oh, S. Y . (2007). The interactional meanings of quasi-pronouns in Korean conversation. In: Enfield NJ, Stivers T (eds) Person Reference in Interaction: Linguistic, Cultural and Social Perspectives. Cambridge: Cambridge University Press, pp. 203–225. Otani, K., & Whitman, J. (1991). V-raising and VP-ellipsis. Linguistic Inquiry, 345-358. Pander Maat, H. L. W., & Sanders, T. J. M. (2001). Subjectivity in causal connectives: An empirical study of language in use. Cognitive Linguistics, 12(3), 247–273. Papoutsaki, A., Sangkloy, P., Laskey, J., Daskalova, N., Huang, J., & James Hays. (2016). WebGazer: Scalable webcam eye tracking using user interactions. Proceedings of the 25th 197 International Joint Conference on Artificial Intelligence (IJCAI) (pp. 3839-3845). Park, B.-S., & Bae, S. (2012). Identifying null arguments: sometimes pro, sometimes ellipsis. Korean Journal of Linguistics, 37, 845-866. Park, B.-S., & Oh, S.-R. (2013). Identifying null arguments with ellipsis. Studies in Generative Grammar, 23, 155-169. Park, M.-K. (1994). A morpho-syntactic study of Korean verbal inflection. Doctoral dissertation, University of Connecticut, Storrs. Park, M.-K. (2013). Debunking null objects in Korean: From pro to DP/NP ellipsis. Studies in Modern Grammar 72, 41–66. Park, K. (2009). Effects of Implicit Causality: Evidence from Korean Interpersonal Verbs, The Korean Journal of Biological Psuchology.21(3), 191-213. Parker D., Phillips C. (2014). Selective priority for structure in memory retrieval, in 27th Annual CUNY Conference on Human Sentence Processing (Columbus, OH: Ohio State University; ), 100 Parker, D., & Phillips, C. (2017). Reflexive attraction in comprehension is selective. Journal of Memory and Language, 94, 272-290. doi: https://doi.org/10.1016/j.jml.2017.01.002 Parker, D., Shvartsman, M., & Van Dyke, J. A. (2017). The cue-based retrieval theory of sentence comprehension: New findings and new challenges. In L. Escobar, V . Torrens, & T. Parodi (Eds.), Language Processing and Disorders (pp. 121–144). Newcastle: Cambridge Scholars Publishing. Patil, U., Vasishth, S., and Lewis, R. L. 2016b. Retrieval interference in syntactic processing: The case of reflexive binding in English. Plank, F. (1979). Ergativity, syntactic typology and universal grammar: Some past and present viewpoints. Ergativity: Towards a Theory of Grammatical Relations, Academic Press, New York, 385-404. Poesio, M. (2000, May). Annotating a Corpus to Develop and Evaluate Discourse Entity Realization Algorithms: Issues and Preliminary Results. In LREC. Poesio, M., Stevenson, R., Eugenio, B. D., & Hitzeman, J. (2004). Centering: A parametric theory and its instantiations. Computational linguistics, 30(3), 309-363. Patil, U., Vasishth, S., & Lewis, R. L. (2016). Retrieval Interference in Syntactic Processing: The Case of Reflexive Binding in English. Frontiers in Psychology, 7, 329. Qualtrics (2020). Qualtrics [Computer software]. Retrieved from https://www.qualtrics.com. 198 Quirk, R., Greenbaum, S., Leech, G., & Svartvik, J. (1985). A comprehensive grammar of the English language. London: Longman. R Core Team. (2017). R: A language and environment for statistical computing. R Foundation for Statistical Computing, Vienna, Austria. Rohde, H. (2008). Coherence-Driven Effects in Sentence and Discourse Processing. Ph.D. thesis. UC San Diego Rohde, H., & Kehler, A. (2008). The bidirectional influence between coherence establishment and pronoun interpretation [Poster presentation]. The 21st Annual CUNY Conference on Human Sentence Processing. Rudolph, U., & Forsterling, F. (1997). The psychological causality implicit in verbs: A review. Psychological Bulletin, 121, 192–218. Runner, J. T., & Ibarra, A. (2016). Information structure effects on null and overt subject comprehension in Spanish. Empirical perspectives on anaphora resolution, 87-112. Ryu, B. R. 2001. Centering and zero anaphora in the Korean discourse. MS thesis, Seoul National University. Salverda, Anne Pier, Dave Kleinschmidt, and Michael K. Tanenhaus. "Immediate effects of anticipatory coarticulation in spoken-word recognition." Journal of memory and language 71, no. 1 (2014): 145-163. Sanders, T. J. M., Sanders, J., & Sweetser, E. E. (2009). Causality, cognition and communication: A mental space analysis of subjectivity in causal connectives. In T. Sanders & E. Sweetser (Eds.), Causal categories in discourse and cognition (pp. 19–60). Berlin: Mouton de Gruyter. Sanders, T. J. M., & Spooren, W. P. M. S. (2015). Causality and subjectivity in discourse: The meaning and use of causal connectives in spontaneous conversation, chat interactions and written text. Linguistics, 53(1), 53–92. Schumacher, P. B., Backhaus, J., & Dangl, M. (2015). Backward-and forward-looking potential of anaphors. Frontiers in psychology, 6, 1746. Schumacher, P. B., Dangl, M., & Uzun, E. (2016). Thematic role as prominence cue during pronoun resolution in German. In Empirical perspectives on anaphora resolution (pp. 213- 240). de Gruyter. Schumacher, P. B., Roberts, L., & Järvikivi, J. (2017). Agentivity drives real-time pronoun resolution: Evidence from German er and der. Lingua, 185, 25-41. Sheldon, A. (1974). The role of parallel function in the acquisition of relative clauses in English. Journal of Verbal Learning and Verbal Behavior, 13, 278-281. 199 Slim, M. S., & Hartsuiker, R. (2021). Moving visual world experiments online? A web-based replication of Dijkgraaf, Hartsuiker, and Duyck (2017) using PCIbex and WebGazer.js. doi:10.31234/osf.io/5adgf Smyth, R. (1994). Grammatical Determinants of Ambiguous Pronoun Resolution. Journal of Psycholinguistic Research, 23, 197-229. Storbeck, J. (2022). Discourse-level processing of nominal possessive constructions. Doctoral Dissertation, University of Southern California. Streb, J., Hennighausen, E., & Rosler, F. (2004). Different anaphoric expressions are investigated by event-related brain potentials. Journal of Psycholinguistic Research, 33(3), 175-201 Strube, M., & Hahn, U. (1996). Functional Centering. Proceedings of the 34 th Annual Meeting of the Association for Computational Linguistics (ACL), 270-277. Song, J., & Kaiser, E. (2020). Effects of discourse factors on the interpretation of Korean null pronouns in subject and object position. Proceedings of the Linguistic Society of America, 5(1), 655-669. Song, J., & Kaiser, E. (2020). Forward-looking effects in subject pronoun interpretation: What comes next matters. Proceedings of the 42nd Annual Meeting of the Cognitive Science Society, 702-708. Streb, J., Hennighausen, E., & Rosler, F. (2004). Different anaphoric expressions are investigated by event-related brain potentials. Journal of Psycholinguistic Research, 33(3), 175-201 Stevenson, R. J., Nelson, A. W. R., & Stenning, K. (1993). Strategies in pronoun comprehension. Proceedings of the Fifteenth Annual Conference of the Cognitive Science Society (pp. 976–981). Hillsdale, NJ: Erlbaum. Stevenson, R. J., Crawley, R. A., & Kleinman, D. (1994). Thematic roles, focus and the representation of events. Language and cognitive processes, 9(4), 519-548. Stevenson, R., Nelson, A., and Stenning, K. (1995). The role of parallelism in strategies of pronoun comprehension. Language and Speech, 38, 393-418. Suri, L. Z., & McCoy, K. F. (1994). RAFT/RAPR and Centering: A comparison and discussion of problems related to processing complex sentences. Computational Linguistics, 20(2), 301- 317. Taboada, M., & Hadic Zabala, L. (2008). Deciding on units of analysis within Centering Theory. Corpus Linguistics and Linguistic Theory, 4(1), 63-108. Taboada, M., & Wiesemann, L. (2010). Subjects and topics in conversation. Journal of Pragmatics, 42(7), 1816-1828. 200 Tanenhaus, M. K., Spivey-Knowlton, M. J., Eberhard, K. M., & Sedivy, J. C. (1995). Integration of visual and linguistic information in spoken language comprehension. Science, 268(5217), 1632-1634. doi:10.1126/science.7777863 Traxler, M. J., Sanford, A. J., Aked, J. P., & Moxey, L. M. (1997). Processing causal and diagnostic statements in discourse. Journal of Experimental Psychology: Learning, Memory, and Cognition, 23(1), 88. Traxler, M. J., Bybee, M. D., & Pickering, M. J. (1997). Influence of connectives on language comprehension: eye tracking evidence for incremental interpretation. The Quarterly Journal of Experimental Psychology Section A, 50(3), 481-497. Traxler, M. J., Pickering, M. J., & Clifton Jr, C. (1998). Adjunct attachment is not a form of lexical ambiguity resolution. Journal of Memory and Language, 39(4), 558-592 Ueno, M., & Kehler, A. (2016). Grammatical and pragmatic factors in the interpretation of Japanese null and overt pronouns. Linguistics, 54(6), 1165-1221. Um, H.-J. (2011). The nature of the null arguments in Korean. Studies in Modern Grammar 63, 73–93. Van Dyke, J. A., & Lewis, R. L. (2003). Distinguishing effects of structure and decay on attachment and repair: A cue-based parsing account of recovery from misanalyzed ambiguities. Journal of Memory and Language, 49(3), 285-316 Van Dyke, J. A., & Lewis, R. L. (2003). Distinguishing effects of structure and decay on attachment and repair: A cue-based parsing account of recovery from misanalyzed ambiguities. Journal of Memory and Language, 49(3), 285-316 Van Dyke, J. A., & Johns, C. L. (2012). Memory Interference as a Determinant of Language Comprehension. Language and Linguistics Compass, 6(4), 193–211. Van Dyke, J. A., & McElree, B. (2011). Cue-dependent interference in comprehension. Journal of Memory and Language, 65(3), 247–263. Van Valin, R.D., Jr. & R. LaPolla (1997): Syntax: structure, meaning, and function. Cambridge: Cambridge University Press. van Rij, J., Hollebrandse, B., & Hendriks, P. (2016). Children’s eye gaze reveals their use of discourse context in object pronoun resolution. In Empirical perspectives on anaphora resolution (pp. 267-294). De Gruyter. Vasishth, S., Brüssow, S., Lewis, R. L., & Drenhaus, H. (2008). Processing polarity: How the ungrammatical intrudes on the grammatical. Cognitive Science, 32(4), 685– 712. https://doi.org/10.1080/03640210802066865 Verhoeven, E. (2009). Subjects, agents, experiencers, and animates in competition: Modern Greek argument order. Linguistische Berichte, 2009(219), 355-376. 201 Verhoeven, E. (2014). Thematic prominence and animacy asymmetries. Evidence from a cross- linguistic production study. Lingua, 143, 129-161. V os, M., Minor, S., & Ramchand, G. C. (2022). Comparing infrared and webcam eye tracking in the Visual World Paradigm. Glossa Psycholinguistics, 1(1). Walker, M., Iida, M., & Cote, S. (1994). Japanese discourse and the process of Centering. Computational Linguistics, 20(2), 1-37. Walker, M. A., Joshi, A. K., & Prince, E. F. (1998). Centering in naturally occurring discourse: An overview. Centering theory in discourse, 128. Wagers, M. W., & Phillips, C. (2014). Going the distance: Memory and control processes in active dependency construction. Quarterly Journal of Experimental Psychology, 67(7), 1274- 1304 Wei, Y ., Mak, W. M., Evers-Vermeul, J., & Sanders, T. J. (2019). Causal connectives as indicators of source information: Evidence from the visual world paradigm. Acta psychologica, 198, 102866. Winograd, T. (1972). Understanding natural language. Cognitive psychology, 3(1), 1-191. Wolf, F., Gibson, E., & Desmet, T. (2004). Discourse coherence and pronoun resolution. Language and Cognitive Processes, 19(6), 665-675. Xiang, M., Dillon, B., & Phillips, C. (2009). Illusory licensing effects across dependency types: ERP evidence. Brain and Language, 108(1), 40–55. Yadav, H., Paape, D., Smith, G., Dillon, B. W., & Vasishth, S. (2022). Individual differences in cue weighting in sentence comprehension: An evaluation using Approximate Bayesian Computation. Open Mind, 6, 1-24. Zehr, J., & Schwarz, F. (2018). PennController for Internet Based Experiments (IBEX). doi:10.17605/OSF.IO/MD832 202 APPENDIX A. I. Target sentences for Experiment 1 1. Tom {intrigued/ missed} Chris because he dated { Ø /him}. 2. Britney {terrified/ recognized} Anna because she zoobed { Ø /her}. 3. Jake {harassed/ disdained} Michael because he frobbed { Ø /him}. 4. Jenny {dissatisfied/ feared} Mary because she plooned { Ø /her}. 5. Henry {surprised/ observed} Kevin because he yarted { Ø /him}. 6. Sue {enlightened/ hated} Kate because she shapped { Ø /her}. 7. Eric {annoyed/ tolerated} John because he soarmed { Ø /him}. 8. Amanda {wounded/ trusted} Lily because she voored { Ø /her}. 9. Peter {frustrated/ resented} George because he rubbeted { Ø /him}. 10. Sarah {teased/ toasted} Emma because she blolled { Ø /her}. 11. Joe {preoccupied/ respected} Robert because he glauched { Ø /him}. 12. Rosie {aggravated/ mourned} Isabella because she yolled { Ø /her}. 13. Nathan {captivated/ esteemed} Paul because he farged { Ø /him}. 14. Wendy {puzzled/ exalted} Jill because she blibbed { Ø /her}. 15. Ethan {disheartened/ spotted} Patrick because he droozed { Ø /him}. 16. Samantha {upset/ abhorred} Veronica because she palped { Ø /her}. 17. Daniel {astonished/ dreamed} Vincent because he perfed { Ø /him}. 18. Miranda {bothered/ noticed} Emily because she goafed { Ø /her}. 19. Oliver {bugged/ supported} Steven because he blothed { Ø /him}. 20. Jenna {mortified/ worshipped} Amber because she verped {Jasmine/her}. 21. Josh {distracted/ cherished} Adam because he speaned { Ø /him}. 22. Stella {stimulated/ appreciated} Jessica because she crozed { Ø /her}. 23. Aaron {tormented/ distrusted} Jason because he spraged { Ø /him}. 24. Stacy {discouraged/ believed} Hanna because she kreeged { Ø /her}. 203 II. Target sentences for Experiment 2 1. Tom {intrigued/ missed} Chris because he dated {Dean/him}. 2. Britney {terrified/ recognized} Anna because she zoobed {Lexie/her}. 3. Jake {harassed/ disdained} Michael because he frobbed {Preston/him}. 4. Jenny {dissatisfied/ feared} Mary because she plooned {Paula/her}. 5. Henry {surprised/ observed} Kevin because he yarted {Marcus/him}. 6. Sue {enlightened/ hated} Kate because she shapped {Martha/her}. 7. Eric {annoyed/ tolerated} John because he soarmed {Brian/him}. 8. Amanda {wounded/ trusted} Lily because she voored {Natasha/her}. 9. Peter {frustrated/ resented} George because he rubbeted {Derek/him}. 10. Sarah {teased/ toasted} Emma because she blolled {Naomi/her}. 11. Joe {preoccupied/ respected} Robert because he glauched {Felix/him}. 12. Rosie {aggravated/ mourned} Isabella because she yolled {Elisa/her}. 13. Nathan {captivated/ esteemed} Paul because he farged {Walter/him}. 14. Wendy {puzzled/ exalted} Jill because she blibbed {Bethany/her}. 15. Ethan {disheartened/ spotted} Patrick because he droozed {Tanner/him}. 16. Samantha {upset/ abhorred} Veronica because she palped {Edith/her}. 17. Daniel {astonished/ dreamed} Vincent because he perfed {Spencer/him}. 18. Miranda {bothered/ noticed} Emily because she goafed {Clara/her}. 19. Oliver {bugged/ supported} Steven because he blothed {Raymond/him}. 20. Jenna {mortified/ worshipped} Amber because she verped {Jasmine/her}. 21. Josh {distracted/ cherished} Adam because he speaned {Mattew/him}. 22. Stella {stimulated/ appreciated} Jessica because she crozed {Leah/her}. 23. Aaron {tormented/ distrusted} Jason because he spraged {Noah/him}. 24. Stacy {discouraged/ believed} Hanna because she kreeged {Jocelyn/her}. 204 III. Target sentences for Experiment 3 1. Tom {advised/chastened} Chris because he dated {Dean/him}. 2. Britney {floored/denounced} Anna because she zoobed {Lexie/her}. 3. Jake {dominated/blamed} Michael because he frobbed {Preston/him}. 4. Jenny {defied/mocked} Mary because she plooned {Paula/her}. 5. Henry {caressed/pardoned} Kevin because he yarted {Marcus/him}. 6. Sue {played/chided} Kate because she shapped {Martha/her}. 7. Eric {courted/lauded} John because he soarmed {Brian/him}. 8. Amanda {grazed/prosecuted} Lily because she voored {Natasha/her}. 9. Peter {deserted/castigated} George because he rubbeted {Derek/him}. 10. Sarah {tracked/scorned} Emma because she blolled {Naomi/her}. 11. Joe {corrupted/chastized} Robert because he glauched {Felix/him}. 12. Rosie {harmed/criticized} Isabella because she yolled {Elisa/her}. 13. Nathan {tailed/saluted} Paul because he farged {Walter/him}. 14. Wendy {ordered/complicated} Jill because she blibbed {Bethany/her}. 15. Ethan {disobeyed/excused} Patrick because he droozed {Tanner/him}. 16. Samantha {deceived/honored} Veronica because she palped {Edith/her}. 17. Daniel {echoed/reprimanded} Vincent because he perfed {Spencer/him}. 18. Miranda {cheated/condemned} Emily because she goafed {Clara/her}. 19. Oliver {repaid/ridiculed} Steven because he blothed {Raymond/him}. 20. Jenna {shadowed/celebrated} Amber because she verped {Jasmine/her}. 21. Josh {trailed/applauded} Adam because he speaned {Mattew/him}. 22. Stella {lied to/commended} Jessica because she crozed {Leah/her}. 23. Aaron {betrayed/supported} Jason because he spraged {Noah/him}. 24. Stacy {followed/welcomed} Hanna because she kreeged {Jocelyn/her}. 205 IV . Target sentences for Experiment 4 1. What about {Tom/Chris}? {He/Tom} chastened {Chris/him} because he daxed {Dean/him}. 2. What about {Britney/Anna}? {She/Britney} denounced {Anna/her} because she zoobed {Lexie/her}. 3. What about {Jake/ Michael}? {He/Jake} blamed {Michael/him} because he frobbed {Preston/him}. 4. What about {Jenny/ Mary}? She mocked {Mary/her} because she plooned {Paula/her}. 5. What about {Henry/ Kevin}? He pardoned {Kevin/him} because he yarted {Marcus/him}. 6. What about {Sue/ Kate}? She chided {Kate/her} because she shapped {Martha/her}. 7. What about {Eric/ John}? He lauded {John/him} because he soarmed {Brian/him}. 8. What about {Amanda/ Lily}? She prosecuted {Lily/her} because she voored {Natasha/her}. 9. What about {Peter/ George}? He castigated {George/him} because he rubbeted {Derek/him}. 10. What about {Sarah/ Emma}? She scorned {Emma/her} because she blolled {Naomi/her}. 11. What about {Joe/ Robert}? He chastized {Robert/him} because he glauched {Felix/him}. 12. What about {Rosie/ Isabella}? She criticized {Isabella/her} because she yolled {Elisa/her}. 13. What about {Nathan/ Paul}? He saluted {Paul/him} because he farged {Walter/him}. 14. What about {Wendy/ Jill}? She complicated {Jill/her} because she blibbed {Bethany/her}. 15. What about {Ethan/ Patrick}? He excused {Patrick/him} because he droozed {Tanner/him}. 16. What about {Samantha/ Veronica}? She honored {Veronica/her} because she palped {Edith/her}. 17. What about {Daniel/ Vincent}? He reprimanded {Vincent/him} because he perfed {Spencer/him}. 18. What about {Miranda/ Emily}? She condemned {Emily/her} because she goafed {Clara/her}. 19. What about {Oliver/ Steven}? He ridiculed {Steven/him} because he blothed {Raymond/him}. 20. What about {Jenna/ Amber}? She celebrated {Amber/her} because she verped {Jasmine/her}. 206 21. What about {Josh/ Adam}? He applauded {Adam/him} because he speaned {Matthew/him}. 22. What about {Stella/ Jessica}? She commended {Jessica/her} because she crozed {Leah/her}. 23. What about {Aaron/ Jason}? He supported {Jason/him} because he spraged {Noah/him}. 24. What about {Stacy/ Hanna}? She welcomed {Hanna/her} because she kreeged {Jocelyn/her}. V . Target sentences for Experiment 5 1. { 혜진이에/ 서영이에} 대해서 말하자면, {Ø/ 서영이를} 그대로 따라했는데 { 민경이를/ Ø} 빈호했기 때문이야. 2. { 지연이에/ 은지에} 대해서 말하자면, {Ø/ 은지를} 뒤쫓았는데 {혜정이를/ Ø} 곤파했기 때문이야. 3. { 은주에/ 영희에} 대해서 말하자면, {Ø/ 영희를} 속였는데 {지은이를/ Ø} 주외했기 때문이야. 4. { 승준이에/ 영민이에} 대해서 말하자면, {Ø/ 영민이를} 미행했는데 {동욱이를/ Ø} 유단했기 때문이야. 5. { 형식이에/ 창현이에} 대해서 말하자면, {Ø/ 창현이를} 불복종했는데 {정환이를/ Ø} 희발했기 때문이야. 6. { 상철이에/ 석진이에} 대해서 말하자면, {Ø/ 석진이를} 해쳤는데 { 정수를/ Ø} 노종했기 때문이야. 7. { 채연이에/ 희진이에} 대해서 말하자면, {Ø/ 희진이를} 배신했는데 {민지를/ Ø} 국절했기 때문이야. 8. { 서희에/ 영지에} 대해서 말하자면, {Ø/ 영지를} 어루만졌는데 {수진이를/ Ø} 의건했기 때문이야. 207 9. { 혜림이에/ 아영이에} 대해서 말하자면, {Ø/ 아영이를} 불렀는데 { 유리를/ Ø} 민파했기 때문이야. 10. { 재원이에/ 승현이에} 대해서 말하자면, {Ø/ 승현이를} 버렸는데 { 민국이를/ Ø} 촉시했기 때문이야. 11. { 병재에/ 재호에} 대해서 말하자면, {Ø/ 재호를} 추적했는데 {영수를/ Ø} 옹정했기 때문이야. 12. { 도준이에/ 준우에} 대해서 말하자면, {Ø/ 준우를} 타락시켰는데 { 재환이를/ Ø} 찬경했기 때문이야. 13. { 우성이에/ 정환이에} 대해서 말하자면, {Ø/ 정환이를} 지배했는데 {진규를/ Ø} 은대했기 때문이야. 14. { 예은이에/ 지현이에} 대해서 말하자면, {Ø/ 지현이를} 따라갔는데 {연정이를/ Ø} 하난했기 때문이야. 15. { 희은 이에/ 미영이에} 대해서 말하자면, {Ø/ 미영이를} 지도했는데 {한나를/ Ø} 준란했기 때문이야. 16. { 현주에/ 수정이에} 대해서 말하자면, {Ø/ 수정이를} 따돌렸는데 { 예나를/ Ø} 다별했기 때문이야. 17. { 영찬이에/ 준상이에} 대해서 말하자면, {Ø/ 준상이를} 접촉했는데 {우찬이를/ Ø} 수린했기 때문이야. 18. { 연우에/ 동윤이에} 대해서 말하자면, {Ø/ 동윤이를} 훼방놓았는데 {찬영이를 / Ø} 모둔했기 때문이야. 19. { 철수에/ 은호에} 대해서 말하자면, {Ø/ 은호를} 납치했는데 {동현이를/ Ø} 간타했기 때문이야. 20. { 태훈이에/ 민혁이에} 대해서 말하자면, {Ø/ 민혁이를} 유린했는데 {인호를/ Ø} 자준했기 때문이야. 21. { 영재에/ 동운이에} 대해서 말하자면, {Ø/ 동운이를} 매수했는데 { 원우를/ Ø} 모행했기 때문이야. 208 22. { 소영이에/ 지예에} 대해서 말하자면, {Ø/ 지예를} 이겼는데 {주은이를/ Ø} 건해했기 때문이야. 23. { 은하에/ 수빈이에} 대해서 말하자면, {Ø/ 수빈이를} 초대했는데 { 혜란이를/ Ø} 두회했기 때문이야. 24. { 세희에/ 혜은이에} 대해서 말하자면, {Ø/ 혜은이를} 의심했는데 { 예리를/ Ø} 주개했기 때문이야. VI. Target sentences for Experiment 6 1. Frank avoided Dave towards the end of May because {Amy/he} had frimoused the jerv to {him/Amy}. 2. Stacy calmed June in the front office because {Ryan/she} had reandered the croab to {her/ Ryan}. 3. Tommy encouraged Max after the winter vacation because {Hanna/he} had wengered the neend to {him/ Hanna}. 4. Betty interrupted Susan during the lunch break because {Josh/she} had chambled the crail to {her/ Josh}. 5. Joe confided in Ethan behind the high school because {Molly/he} had bistorked the plorf to {him/ Molly}. 6. Julie embraced Bella on the sidewalk because {Will/she} had pladdened the mert to {her/ Will}. 7. Eric forgave Nick at the airport terminal because {Nicole/he} had maggled the broit to {him/ Nicole}. 8. Maggie consulted Ava at the coffee shop because {Owen/she} had histed the yolch to {her/ Owen}. 9. Ben approached Aaron on the front porch because {Cindy/he} had grastled the zike to {him/ Cindy. 209 10. Abby advised Mia at the bus stop because {Phil/she} had tharsted the chise to {her/ Phil}. 11. Bill bothered Mark during the train ride because {Lizzie/he} had parooved the harst to {him/ Lizzie}. 12. Alice questioned Mary at the fitness center because {Tony/she} had yoppled the dirb to {her/ Tony}. 13. Alan insulted James in the living room because {Lisa/he} had skalled the proamer to {him/ Lisa}. 14. Mindy comforted Jane at the courthouse because {Noah/she} had falped the tanger to {her/ Noah}. 15. Paul surprised Jack at the soccer game because {Nora/he} had gwayned the jaloom to {him/ Nora}. 16. Sue called Jill from the high-rise hotel because {Henry/she} had spaved the ninder to {her/ Henry}. 17. Sean reassured Tom during the phone call because {Anne/he} had plooned the ligot to {him/ Anne}. 18. Jenna humiliated Kate on new year's eve because {Joseph/she} had kleezed the wirin to {her/ Joseph}. 19. Kyle threatened Adam on the street because {Emma/he} had voored the pargle to {him/ Emma}. 20. Beth shamed Lily in the back yard because {Liam/she} had fromped the mipper to {her/ Liam}. 21. Leo consoled Ian at the birthday party because {Kelly/he} had bloored the drossler to {him/ Kelly}. 22. Lucy tickled Ivy on the park bench because {John/she} had yolled the lyfander to {her/ John}. 23. Luke provoked Thomas at the movie theater because {Liz/he} had clorved the meynit to {him/ Liz}. 24. Ruth hypnotized Sarah in the late evening because {Jake/she} had spraged the gebron to {her/ Jake}. 210 APPENDIX B. 1. Statistical models for Experiment 1: • Final glmer model: glmer(he_obj ~ refst_type * ic_type + (1 + refst_type | participant) + (1 | item), data=data1, family=binomial) • glmer models used for planned comparisons:  SE-only conditions: glmer(he_obj ~ refst_type + (1 | participant) + (1 | item), data=data1_1, family=binomial)  ES-only condition: glmer(he_obj ~ refst_type + (1 | participant) + (1 | item), data=data1_2, family=binomial) 2. Statistical models for Experiment 2 • Final glmer model: glmer(he_obj ~ refst_type * ic_type + (1 + refst_type | participant) + (1 | item), data=data2, family=binomial) • glmer models used for planned comparisons:  SE-only condition: glmer(he_obj ~ refst_type + (1 + refst_type | participant) + (1 | item), data=data2_1, family=binomial)  ES-only condition: glmer(he_obj ~ refst_type + (1 + refst_type | participant) + (1 + refst_type | item), data=data2_2, family=binomial) 3. Statistical models for Experiment 3 • Final glmer model: glmer(he_obj ~ refst_type * ic_type + (1 + refst_type | participant) + (1 | item) , data=data3, family=binomial) • Experiment1 vs. Experiment 3: glmer(he_obj ~ refst_type * exp_3 + (1 | participant) + (1 | item) , data=data4, family=binomial) • Experiment2 vs. Experiment 3: glmer(he_obj ~ refst_type * exp_3 + (1 | participant) + (1 | item) , data=data5, family=binomial) 211 4. Statistical models for Experiment 4 • Final glmer model: glmer(he_obj ~ refst_type * top_type + (1 + refst_type | participant) + (1 | item), data= exp4_stats_all_1, family=binomial) • glmer models used for planned comparisons:  Sbj-topic condition: glmer(he_obj ~ refst_type + (1 | participant) + (1 | item), data=exp4_sbjtop, family=binomial)  Obj-topic condition: glmer(he_obj ~ refst_type + (1 + refst_type | participant) + (1 | item), data=exp4_objtop, family=binomial) 5. Statistical models for Experiment 5 • Final glmer model: glmer(he_sbj ~ refst_type * top_type + (1 + refst_type | participant) + (1 | item), data=data3, family=binomial) • glmer models used for planned comparisons:  Sbj-topic condition: glmer(he_sbj ~ refst_type + (1 + refst_type | participant) + (1 + refst_type | item), data=data3, family=binomial)  Obj-topic condition: glmer(he_sbj ~ refst_type + (1 + refst_type | participant) + (1 | item), data=data3, family=binomial) • Experiment 4 vs. Experiment 5: glmer(he_obj ~ top_type * exp_type + (1 + top_type | participant) + (1 | item), data=exp45, family=binomial) 6. Statistical models for Experiment 6 • glmer formula for the model predicting selections of the last-mentioned referent: ▪ One-Object-Pronoun vs. Two-Pronoun condition: sbj_choice ~ Type + (1 + Type | prolificID) + (1 | itemID) ▪ One-Subject-Pronoun vs. Object-Pronoun condition: per3_choice ~ Type + (1 | prolificID) + (1 | itemID) 212 ▪ One-Subject-Pronoun vs. Two-Pronoun condition: per3_choice ~ Type + (1 | prolificID) • lmer formula for the model predicting the subject advantage scores to the offset of the subject pronoun: ▪ -500 - 0ms: lmer_diffScore ~ Type + (1 | prolificID) ▪ 0 – 500ms: lmer_diffScore ~ Type + (1 | prolificID) + (1 | itemID) ▪ 500 -1000ms: lmer_diffScore ~ Type + (1 | prolificID) ▪ 1000-1500ms: lmer_diffScore ~ Type + (1 | prolificID) + (1 | itemID) • lmer formula for the model predicting the subject advantage scores to the onset of the indirect object pronoun: ▪ -400 - 0ms: lmer_diffScore ~ Type + (1 | prolificID) + (1 | itemID) ▪ 0 – 500ms: lmer_diffScore ~ Type + (1 | prolificID) + (1 | itemID) ▪ 500 -1000ms: lmer_diffScore ~ Type + (1 | prolificID) + (1 | itemID) ▪ 1000-1500ms: lmer_diffScore ~ Type + (1 | prolificID) + (1 | itemID) ▪ 1500-2000ms: lmer_diffScore ~ Type + (1 | prolificID) ▪ 2000-2500ms: lmer_diffScore ~ refType + (1 | prolificID) ▪ 2500-3000ms: lmer_diffScore ~ Type + (1 + Type | prolificID) + (1 | itemID) ▪ 3000-3500ms: lmer_diffScore ~ Type + (1 + Type | prolificID) + (1 | itemID) ▪ 3500-4000ms: lmer_diffScore ~ Type + (1 | prolificID) ▪ 1000-3000ms: lmer_diffScore ~ Type + (1 | prolificID) + (1 | itemID)

Abstract (if available)

Abstract This dissertation explores the mechanisms underlying multiple pronoun resolution in comparison to single pronoun resolution, focusing on differences in referential structure – whether all or only one of the referents in the preceding clause are mentioned with a pronoun. This dissertation aims to answer four questions: (i) whether and how the process of multiple pronoun resolution is different from that of single pronoun resolution (ii) what factors contribute to effects of referential structure (iii) whether and how referential structure effects are influenced by differences in anaphoric forms, and (iv) how multiple pronouns are interpreted during real-time processing.
Experiments 1, 2 and 3 investigate the first question and test two hypotheses about the potential effects of referential structure in pronoun resolution. The independence view assumes that each pronoun’s anaphoric dependencies are resolved fully independently while the dependence views assume that there are interactions between multiple referential dependencies. Accordingly, pronoun interpretation is predicted to be sensitive to differences in referential structure type only in the dependence views, but not in the independence view. The results from three offline picture-writing tasks support the dependence view, showing that pronoun interpretation is indeed sensitive to referential structure type,.
Experiment 4 addresses the second question and further examines what factors at different levels of representation contribute to effects of referential structure by comparing three types of the dependence view (syntax-level, discourse-level, and syntax-discourse dependence view), based on the influence of syntactic and/or discourse factors. The results from an offline picture-writing task support the syntax-discourse dependence view, suggesting that both syntax (i.e., cue-based parallelism) and discourse factors (i.e., Centering Theory) contribute to the interactions between anaphoric dependencies of the two different pronouns, resulting in referential structure effects.
Experiment 5 addresses the third question and broadens the scope of referential structure effects to another type of anaphoric form, null pronouns (pro) in Korean. This study tests whether the discourse-level coherence factor has a greater impact on referential structure effects than the syntax-level factor with Korean pro, which have a reduced anaphoric form compared to overt pronouns in English. The results from an offline picture-writing task show that the discourse factor indeed exerts a greater influence on pro resolution in Korean than overt pronoun resolution in English.
Experiment 6 investigates the last question by examining referential structure effects during real-time processing using a novel webcam-based visual-world eye-tracking method. Two competing hypotheses are tested regarding the timing of referential structure effects: the immediate activation hypothesis assumes that when the subsequent pronoun is encountered, the preceding pronoun’s referential dependencies are retrieved immediately while the delayed activation hypothesis assumes that the retrieval is delayed. The results indicate early emergence of referential structure effects during resolution of the subsequent pronoun (i.e., the immediate activation hypothesis), suggesting that the preceding pronoun’s dependency information becomes available as soon as a comprehender encounters the subsequent pronoun.
In conclusion, this dissertation provides novel evidence that multiple pronoun resolution is different from single pronoun resolution and that these differences stem from differences in referential structure – resolving referential dependency of one pronoun can influence on resolving dependency of the other pronoun. Thus, the findings of this dissertation highlight the need to consider referential structure in understanding pronoun interpretation.

Linked assets

University of Southern California Dissertations and Theses

Conceptually similar

PDF

Interpretation of pronominal forms across languages

PDF

Syntactic and non-syntactic factors in reflexive pronoun resolution in Mandarin Chinese

PDF

Vietnamese pronouns in discourse

PDF

Processing the dynamicity of events in language

PDF

Discourse-level processing of nominal possessive constructions

PDF

Discourse level processing and pronoun interpretation

PDF

How to think before you speak: getting from abstract thoughts to sentences

PDF

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

PDF

What 'you' and 'I' can say about reference resolution and non-structural constraints

PDF

Towards a correlational law of language: three factors constraining judgement variation

PDF

Investigating coordination of lexical and structural information cross-linguistically

PDF

Interaction between prosody and information structure: experimental evidence from Hindi and Bangla

PDF

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

PDF

Considering the effects of disfluent speech on children’s sentence processing capabilities and language development

PDF

Comparative iIlusions at the syntax-semantics interface

PDF

Prosody and informativity: a cross-linguistic investigation

PDF

Evaluating and improving the commonsense reasoning ability of language models

PDF

Syntax-aware natural language processing techniques and their applications

PDF

Perspective in Turkish complementation

PDF

Linking eyes to mouth: a schema-based computational model for describing visual scenes

Asset Metadata

Creator Song, Jina (author)

Core Title Dynamics of multiple pronoun resolution

School College of Letters, Arts and Sciences

Degree Doctor of Philosophy

Degree Program Linguistics

Degree Conferral Date 2023-05

Publication Date 03/29/2023

Defense Date 02/27/2023

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

Tag centering theory,cue-based retrieval,OAI-PMH Harvest,pronoun resolution,referential structure,sentence processing,webcam-based eyetracking

Format theses (aat)

Language English

Contributor Electronically uploaded by the author (provenance)

Advisor Kaiser, Elsi (committee chair), Mintz, Toby (committee member), Simpson, Andrew (committee member)

Creator Email jinas@usc.edu,joenjina@gmail.com

Permanent Link (DOI) https://doi.org/10.25549/usctheses-oUC112850242

Unique identifier UC112850242

Identifier etd-SongJina-11529.pdf (filename)

Legacy Identifier etd-SongJina-11529

Document Type Dissertation

Format theses (aat)

Rights Song, Jina

Internet Media Type application/pdf

Type texts

Source 20230329-usctheses-batch-1012 (batch), 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 author, as the original true and official version of the work, but does not grant the reader permission to use the work if the desired use is covered by copyright. It is the author, as rights holder, who must provide use permission if such use is covered by copyright. The original signature page accompanying the original submission of the work to the USC Libraries is retained by the USC Libraries and a copy of it may be obtained by authorized requesters contacting the repository e-mail address given.

Repository Name University of Southern California Digital Library

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

Repository Email cisadmin@lib.usc.edu