University of Southern California Dissertations and Theses

Flexibility in language production

(USC Thesis Other)

Flexibility in language production

PDF

Download a page range

Download transcript

Copy asset link

Request this asset

Transcript (if available)

Content FLEXIBILITY IN LANGUAGE PRODUCTION by Yu-Chi Huang A Dissertation Presented to the FACULTY OF THE USC GRADUA TE SCHOOL UNIVERSITY OF SOUTHERN CALIFORNIA In Partial Fulfillment of the Requirements for the Degree DOCTOR OF PHILOSOPHY (LINGUISTICS) August 2012 Copyright 2012 Yu-Chi Huang ii Acknowledgements This dissertation would not exist without the guidance, assistance and encouragement of a great many people. In particular, I am extremely indebted to Dr. Elsi Kaiser, my advisor and mentor. Elsi is truly a blessing throughout my graduate years at USC. As my academic advisor, she has provided me with invaluable instructions in order to develop my path to academic success. With her incredible patience and energy, she guided me smoothly through the process of developing budding ideas into a full-fledged project. I have benefited immeasurably from her constructive suggestions and detailed comments on my work. As my life mentor, Elsi has played the role of a reliable and wise friend, who always listens to my problems and reacts with an optimistic attitude and positive thinking. An encouraging conversation with her worked miraculously each time in dispelling my frustration, stress and worries. I can never thank Elsi enough for her support and faith in me. She is truly an honorable scholar and individual, who stands for a genuine role model of her profession. I am also greatly indebted to the other members of my dissertation committee: Dr. Audrey Li, Dr. Andrew Simpson, and Dr. Bosco Tjan. They spent long hours reading the early drafts of my dissertation, kindly offering useful comments on revisions and insightful suggestions for future research. I am really grateful to them for showing interest in my work and for generously sharing their expertise with me. iii In addition, I would like to thank the faculty, staff and fellow students of the Linguistics department at USC for giving me an unforgettable and fruitful journey filled with happy memories. Last but not least, special thanks go to my parents and parents-in-law, my husband and my three wonderful kids. Because of their unconditional support and love over these years, I finally reached the end of the tunnel! iv Table of Contents Acknowledgements……………………………………………………………………….ii List of Tables……………………………….……………………………………………vii List of Figures…………………………………………………………………………….ix Abstract ………………………………………………………………………………….xii Chapter 1: General Introduction 1.1 Language Production Models...............................................................................1 1.2 Flexibility in Sentence Production........................................................................6 1.2.1 The Competitive Theory..............................................................................7 1.2.2 The Incremental Theory...............................................................................8 1.3 Accessibility Effects in Sentence Production ........................................................9 1.4 The Scope of Advance Linguistic Planning ........................................................12 1.4.1 Lexical Incrementality...............................................................................13 1.4.2 Structural Incrementality ...........................................................................14 1.5 Overview of This Dissertation............................................................................15 1.5.1 Research Questions ...................................................................................15 1.5.2 Methodology.............................................................................................19 1.5.3 Structure of This Dissertation ....................................................................22 Chapter 2: Mandarin Sentence Production Task Using Agent-Theme Verbs 2.1 Introduction…………………………………………………..…………………..26 2.1.1 V. Ferreira (1996) on Syntactic Flexibility.....................................................29 2.1.2 My Work on Semantic Flexibility..............................................................31 2.1.3 F. Ferreira (1994) on Thematic Prominence and Animacy..........................32 2.1.4 Mandarin BA and BEI constructions..........................................................35 2.2 Experiment 1: Agent-theme Verbs .....................................................................40 2.2.1 Methodology.............................................................................................40 2.2.2 Hypotheses and Predictions .......................................................................54 2.2.3 Results ......................................................................................................58 2.3 Discussion..........................................................................................................72 Chapter 3: Mandarin Sentence Production Task Using Theme-Experiencer Verbs 3.1 Motivation .........................................................................................................74 3.2 Experiment 2......................................................................................................76 3.2.1 Methodology.............................................................................................76 3.2.2 Hypotheses and predictions ......................................................................79 3.2.3 Results ......................................................................................................83 v 3.3 Discussion..........................................................................................................95 3.4 Error Analysis of Experiments 1 and 2 .............................................................101 3.5 General Discussion of Experiments 1 and 2......................................................107 Chapter 4: English Sentence Production Task Using Three Verb Types 4.1 Introduction .....................................................................................................116 4.2 Experiment 3...................................................................................................118 4.2.1 Methodology...........................................................................................118 4.2.2 Hypotheses and Predictions .....................................................................125 4.2.3 Results ....................................................................................................130 4.3 Error Analysis of Experiment 3 ........................................................................139 4.4 Discussion........................................................................................................144 Chapter 5 : Analyzing Perfect Responses 5.1 Introduction ....................................................................................................149 5.2 Perfect Responses of Experiments 1 and 2 (Chinese)........................................154 5.2.1 Experiment 1...........................................................................................155 5.2.2 Experiment 2...........................................................................................166 5.2.3 Comparing Correct Responses and Perfect Responses in Experiments 1 and 2………………………………...................................................................177 5.3 Perfect Responses of Experiment 3 (English) ...................................................184 5.3.1 Results ....................................................................................................186 5.3.2 Comparing Correct Responses and Perfect Responses in Experiment 3 ...193 5.4 Theoretical Implications...................................................................................195 Chapter 6: English Production Tasks Investigating Strategic Control 6.1 Introduction .....................................................................................................201 6.2 Experiment 4: Accuracy -oriented Setting ........................................................207 6.2.1 Methodology...........................................................................................207 6.2.2 Hypotheses and Predictions .....................................................................208 6.2.3 Results ....................................................................................................211 6.3 Experiment 5: Speed-oriented Setting .............................................................216 6.3.1 Methodology..........................................................................................216 6.3.2 Hypotheses and Predictions ....................................................................217 6.3.3 Results ...................................................................................................221 6.4 Across-Experiment Analysis between Experiments 4 and 5 ..............................225 6.4.1 Production Latency Analysis ...................................................................225 6.4.2 Error Analysis .........................................................................................226 6.5 Discussion........................................................................................................232 vi Chapter 7: Conclusions 7.1 Summary of the Current Work…………………………………………………239 7.2 Conclusing Remarks on Future Work ..............................................................246 References……………………………………………………………………………....250 Appendices Appendix A: Target Items for Experiment 1...........................................................264 Appendix B: Target Items for Experiment 2...........................................................266 Appendix C: Target Items for Experiment 3, 4 and 5..............................................268 vii List of Tables Table 2.1: The distributions of thematic roles in the BA and BEI constructions (shaded areas indicate the preferred word order predicted by thematic prominence)………….....................................................................................39 Table 2.2: The experimental conditions of Experiment 1……………………………….42 Table 2.3: Number and percentage of sentence types in each condition of Experiment 1…………………………………………….………………………………...62 Table 3.1: The experimental conditions of Experiment 2…………………….……........77 Table 3.2: Number and percentage of sentence types in each condition of Experiment 2……………………….……………………………………………………...84 Table 3.3: Count of each error type in each condition of Experiment 1………….........102 Table 3.4: Count of each error type in each condition of Experiment 2………….........105 Table 4.1: The experimental conditions of Experiment 3………………………….......121 Table 4.2: The predicted difficulty level of each verb type in Experiment 3……………………………………………………………….…………….126 Table 4.3: Number and percentage of sentence types in each condition of Experiment 3…………………………………………………………………….……….131 Table 4.4: Count of different error types depending on verb type in Experiment 3…..................................................................................................................140 Table 4.5: Count of the subtypes of correct responses depending on verb type in Experiment 3……………………………………………………….……….142 Table 4.6: A comparison of the mean production latency of the Mandarin SA-ba / SA-bei conditions and the English active/passive conditions (shaded)…………….147 Table 5.1: Count of each subtype of correct responses in Experiment 1…………........154 viii Table 5.2: Count of each subtype of correct responses in Experiment 2…………..…...155 Table 5.3: Comparison between correct responses and perfect responses of Experiment 1.………………..…………………………………………………………...178 Table 5.4: Comparison between correct responses and perfect responses of Experiment 2……………………………………………………………………………..178 Table 5.5: Mean production latencies of correct responses and perfect responses in the SA no-prep. and DA no-prep. conditions in Experiment 1…………………181 Table 5.6: Mean production latencies of correct responses and perfect responses in the DA-ba and DA-bei conditions in Experiment 2………………….…...........183 Table 5.7: Count of each subtype of correct responses depending on verb type in Experiment 3………………………………………….…………………….185 Table 5.8: Comparison between correct responses and perfect responses of Experiment 3................................................................……………………………..……194 Table 6.1: The predicted difficulty level of each verb type in an accuracy-oriented setting…………………………………………………………………….....211 Table 6.2: The predicted difficulty level of each verb type in a speed-oriented setting…….....................................................................................................220 Table 6.3: Count of each error type in Experiments 4 and 5…………………………..229 Table 6.4: Count of the subtypes of correct responses depending on verb type in Experiment 4……………………………………………………………......231 Table 6.5: Count of the subtypes of correct responses depending on verb type in Experiment 5……………………………………………………………......231 ix List of Figures Figure 1.1: A consensus language production model (reprinted from Bock & Levelt, 1994) ……………………………………………………………………….....3 Figure 1.2: A parallel processing model (adapted from Dell & O'Seaghdha, 1994)……...5 Figure 2.1: Layout of one experimental trial in Experiment 1…………………………..45 Figure 2.2: Average rating of acceptability and plausibility for each condition in the norming study ( 1 = ‘totally unacceptable and incomprehensible’ ; 4 = ‘totally acceptable and comprehensible’)…………………………………………….50 Figure 2.3: Average rating as a function of animacy and preposition in the norming study………………………………………………………………………….62 Figure 2.4: Mean error rate as a function of animacy and preposition constraint in Experiment 1…………………………………………………………………64 Figure 2.5: Mean production latency as a function of animacy and preposition constraint in Experiment 1………………………………………………………………67 Figure 2.6: Mean production latency of ba and bei sentences in the SA-no prep.condition vs. the DA-no prep. condition in Experiment 1……………………………...69 Figure 2.7:Mean production latency of the SA-ba vs. DA-ba conditions and the SA-bei vs. DA-bei conditions in Experiment 1………………………………………….70 Figure 3.1: Mean error rate as a function of animacy and preposition constraint in Experiment 2…………………………………………………………………87 Figure 3.2: Mean production latency as a function of animacy and preposition constraint in Experiment 2………………………………………………………………90 Figure 3.3: Mean production latency of ba and bei sentences in the SA-no prep. condition vs. the DA-no prep. condition in Experiment 2……………………………...92 Figure 3.4:Mean production latency in the SA-ba vs. DA-ba conditions and the SA-bei vs. DA-bei conditions in Experiment 2………………………………………….93 x Figure 3.5: Count of role reversal in DA-ba vs. DA-bei conditions with two verb types...............................................................................................................106 Figure 4.1: Layout of one experimental trial in Experiment 3…………………………122 Figure 4.2: Mean error rate as a function of position and voice in Experiment 3……...132 Figure 4.3: Mean error rate depending on verb type in the active and passive conditions in Experiment 3………………………………………………………………..133 Figure 4.4: Mean production latency as a function of position and voice in Experiment 3……………………………………………………………………………..136 Figure 4.5: Mean production latency depending on verb type in the active and passive conditions in Experiment 3…………………………………………………137 Figure 5.1: Mean perfect-response rate as a function of animacy and preposition constraint in Experiment 1………………………………………………….157 Figure 5.2: Mean production latency of perfect responses as a function of animacy and preposition constraint in Experiment 1……………………………………..161 Figure 5.3: Mean production latency of ba and bei sentences categorized as perfect responses in the SA-no prep. condition vs. the DA-no prep. condition in Experiment 1………………………………………………………………..163 Figure 5.4: Mean production latency of perfect responses in the SA-ba vs. DA-ba conditions and the SA-bei vs. DA-bei conditions in Experiment 1………...164 Figure 5.5: Mean perfect-response rate as a function of animacy and preposition constraint in Experiment 2………………………………………………..168 Figure 5.6: Mean production latency of perfect responses as a function of animacy and preposition constraint in Experiment 2……………………………………..172 Figure 5.7: Mean production latency of ba and bei sentences categorized as perfect responses in the SA-no prep. vs. the DA-no prep. conditions in Experiment 2……………………………………………………………………………..174 xi Figure 5.8: Mean production latency of perfect responses in the SA-ba vs. DA-ba conditions and the SA-bei vs. DA-bei conditions in Experiment 2………...175 Figure 5.9: Mean perfect-response rate depending on verb type in the active and passive conditions in Experiment 3…………………………………………………188 Figure 5.10: Mean production latency of perfect responses depending on verb type in the active and passive conditions in Experiment 3…………………….….........191 Figure 6.1: Mean production latency as a function of verb type and voice in Experiment 4……………………………………………………………………………..213 Figure 6.2: Mean production latency depending on verb type in the active and passive conditions in Experiment 4…………………………………………………214 Figure 6.3: Mean production latency as a function of verb type and voice in Experiment 5……………………………………………………………………………..222 Figure 6.4: Mean production latency depending on verb type in the active and passive conditions in Experiment 5…………………………………………………223 Figure 6.5: Mean production latency as a function of verb type and voice in Experiments 4 and 5………………………………………………………........................226 xii Abstract The study of sentence production investigates how speakers turn preverbal messages into utterances. Researchers generally characterize the language production system into three broad components: the message component (conceptualization), the grammatical component (formulation), and the phonological component (articulation) (Bock 1995; Bock & Levelt 1994; Garrett, 1975; Levelt, 1989). Specifically, my work aims to explore the nature and processes of the message component and grammatical component by investigating three primary questions. First, what would happen if there is freedom of choice for message formulation (termed semantic flexibility in my work)? Second, do accessibility effects of thematic prominence and animacy condition syntactic choice in Mandarin? Third, how does speakers’ strategic control in response to communicative goals influence the processes of message and grammatical encoding? In my dissertation, I report five sentence production tasks in Mandarin Chinese and English designed to address these questions. Experiments 1 and 2 explored the effects of semantic flexibility in the production of Mandarin BA and BEI constructions. The results suggest that message formulation processes are highly sensitive to the interaction between thematic role and animacy: Whether having choice of multiple propositions hinders or facilitates production is contingent upon verb thematic structure. Additionally, these findings extend effects of thematic prominence and animacy to Mandarin BA and BEI constructions, providing cross-linguistic evidence for the incrementality-based account of grammatical encoding. xiii Experiment 3 investigated both effects of thematic prominence and semantic flexibility in the production of English actives and passives. The results confirm effects of thematic prominence in English, but do not replicate the asymmetric pattern of semantic flexibility as found in Mandarin Chinese, suggesting that semantic properties of a syntactic construction may also play a role in message formulation. In sum, the findings of Experiments 1-3 indicate that the operation of the message component may be carried out competitively or incrementally, in contrast to the incremental nature of grammatical encoding. In a broad sense, linguistic information such as verb thematic structure and semantic properties of a syntactic construction may guide the process of sentence production at an early stage. Experiments 4 and 5 studied how extra-linguistic factors, in particular, speakers’ strategic control, influences sentence production in both the message and grammatical component. Existing work suggests that speakers’ strategy or task sensitivity has the potential to affect both referent processing in message planning and the scope of advance linguistic planning. Thus, I hypothesize that in prepared speech vs. fast speech, effects of semantic flexibility and thematic prominence would vary with the communicative goal aiming at accuracy or speed. The results of Experiments 4 and 5 show that consistent with the results of Experiment 3, effects of semantic flexibility are not found. Contrastively, effects of thematic prominence are found in prepared speech but not in fast speech. In conclusion, speakers’ strategic control can affect the processes of grammatical encoding, depending on whether the goal speakers aim to achieve is speed or accuracy. xiv As a whole, my findings suggest that message formulation is sensitive to the interaction of linguistic information such as verb thematic structure, conceptual accessibility such as noun animacy, and semantic properties pertaining to syntactic constructions. In contrast to grammatical encoding, message formulation may not always proceed incrementally. Moreover, the language production system is susceptible to extra-linguistic factors such as strategic control, in particular the grammatical component. By exploring the nature and processes of the individual components as well as the interactions between them, my research helps shed light on how people produce speech. 1 Chapter 1: General Introduction 1.1 Language Production Models Speech seems to be easy. On a daily basis, people engage in language production to express their ideas and communicate with others. Despite the seeming effortlessness, the processes of turning intended messages into linguistic expressions involve incredibly complicated maneuvers. A theory of sentence production must specify what knowledge speakers have and how the information processing system puts that knowledge into use. For the past several decades, psycholinguists have focused on characterizing the production system and the processes that allow speech to proceed in an effortless and efficient way. Early work explored language production by studying speech errors and disfluencies (e.g., Blackmer & Mitton, 1991; Brédart, 1991; Butterworth, 1980; Christenfeld, 1994; Clark, 2002; Clark & Clark, 1977; Clark & Wasow, 1998; Dell, 1990; Fromkin, 1971, 1973; Garrett, 1975; Goldman-Eisler, 1968; Henderson, 1974; Holmes, 1988; Kemper, 1992; Kowall, O’Connell & Sabin, 1975; Levelt, 1983, 1989; MacKay, 1970, 1972, 1992a, 1992b; Postma, 2000; Postma & Kolk, 1993; Postma, Kolk & Povel, 1990; Schlenk, Huber, & Willmes, 1987, among others). More recent research applied experimental methods to explore productive processes in progress (e.g., Bock 1982, 1986a, 1986b, 1987a, 1995; Bock & Levelt, 1994; Bock, Loebell & Morey, 1992; Bock & Warren, 1985; Brysbaert, Fias & Noel, 1998; De Smedt, 1990, 1994; Dell, 1986; Dell & O'Seaghdha,1992; F. Ferreira, 1991, 1994; F. Ferreira & Swets, 2002, 2005; V . Ferreira, 1996, 2003; Griffin, 1998, 2001, 2003; Griffin & Bock, 2000; Konopka, 2012; Levelt, 1989; Meyer, 1991, 1996, 1997; Prat-Sala & Branigan, 2000; Smith & Wheeldon, 2 1999, 2004; Stallings, MacDonald & O’Seaghdha, 1998; Wheeldon & Lahiri, 1997, among others). Over these decades, a number of researchers have proposed language production models that characterize the processes which encode “preverbal messages” (Levelt, 1989) into linguistic expressions (e.g., Bock, 1995; Bock & Levelt, 1994; Dell, 1985, 1986; Fromkin, 1971; Garrett, 1975, 1980, 1988; Levelt, 1989; also see V. Ferreira & Slevc, 2007, for a detailed overview). Although the process of language production is not yet fully understood, existing accounts share some common views, in particular, the view that the production system consists of three components or levels: the message component, the grammatical component, and the phonological component. Figure 1.1 illustrates a widely-accepted language production model proposed by Bock & Levelt (1994). 3 Figure 1.1: A consensus language production model (reprinted from Bock & Levelt, 1994) The main function of each component is sketched below. First, in the message component, a preverbal message is encoded to express a proposition (i.e., conceptualization). Once at least part of a preverbal message is encoded, it is then sent to the grammatical component for grammatical encoding (i.e., linguistic formulation). In his two-stage model, Garret 4 (1975) proposed that the grammatical component is subdivided into two stages: In the functional stage (i.e., content processing), lemmas are retrieved from the mental lexicon and grammatical roles (i.e., subject or object) are assigned. In the positional stage (i.e., structural processing), words are assembled linearly into a sentence. After at least part of the syntactic structure is built, it is then sent to the phonological component for phonological encoding, and then spelled out in its verbal form (i.e., articulation). Although researchers generally agree that there is a hierarchy of levels or components which comprise language production models, there is no unanimous view about how the process of sentence production is executed. Some argue for a serial processing model (e.g., Fromkin, 1971; Garret, 1975; Bock & Levelt, 1994), which claims that speech is produced serially. The process is composed of a series of sequential stages or modules, with earlier stages comprising of the large units (i.e., sentences and phrases), and later stage comprising of their smaller unit constituents (i.e., distinct features like voicing, phonemes, morphemes, syllables). Early versions of these models (Fromkin, 1971; Garret, 1975) claim that the stages are independent of one another and that flow of information is unidirectional. Later versions (Bock & Levelt, 1994) incorporate a bidirectional flow of information, but still involve discrete serial processing. In contrast to the serial models, some researchers argue for parallel processing models (e.g., Dell, 1985,1986), which claim that the levels of representation can interact with each other and are processed in parallel, as depicted by Figure 1.2. Also referred to as the Connectionist Model, the model in Figure 1.2 posits that speech is produced by a number of connected nodes 5 representing distinct units of speech that interact with one another in any direction by spreading activation across different levels. An important feature of this model is that nodes of the same level spread inhibitory activations and compete with one another, while nodes of different levels have excitatory connections. Figure 1.2: A parallel processing model (adapted from Dell & O'Seaghdha, 1994) Crucially, the serial and parallel models make distinct predictions regarding effects of syntactic flexibility, a notion which will be introduced in the next section. The current work is mainly situated within the theoretical framework of a serial processing model. However, throughout this document, I will involve theoretical accounts from both models for developing arguments and elaborating discussions. 6 1.2 Flexibility in Sentence Production To encode a preverbal message into a well-formed sentence, the speaker must respond to a series of consecutive calls for decision-making at the planning stage. To name just a few, at the message planning stage, the speaker needs to determine the speech act to be carried out (e.g., question vs. declarative), the event and entities to be mentioned, and the predicate type to be adopted (e.g., sell/buy alternative). Additionally, in the linguistic planning stage, he or she needs to select proper lexical items from semantically-related cohorts (e.g., cat or tabby), and to decide on the linear word order and the eventual syntactic structure. It is truly a wonder how a task that involves coordination of different encoding components is almost always executed with accuracy and efficiency. In reality, every language provides its speakers with freedom in some aspects. For example, in some languages such as Finnish and Turkish, speakers can determine the linear word order of a sentence (e.g., SOV or OSV). Even in English, a language which has a relatively fixed word order, it is still possible for speakers to express an event of a cat chasing a mouse in multiple syntactic structures: the dog chased the cat, the cat was chased by the dog, or even it is the dog that chased the cat. An interesting question related to decision-making is: Given limitations on working-memory capacity (e.g., Baddeley & Hitch, 1974; Caplan & Waters, 1999; D’Esposito, Postle, Ballard & Lease, 1999; Gordon, Hendrick & Johnson, 2001; Gordon, Hendrick & Levine, 2002; Hartsuiker & Barkhuysen, 2006; Palmer & Schendel, 2002), is it malignant to have options of alternative syntactic structures when formulating a sentence? In response to this question, 7 V. Ferreira (1996) discussed two competing views which lead to opposing predictions: the competitive theory vs. the incremental theory. He illustrated the two models with an example involving English double object construction (e.g., “I gave the boy the toy”) and dative construction (e.g., “I gave the toy to the boy.”) More detailed discussions regarding these two competing view and their predictions are given below. 1.2.1 The Competitive Theory The first view is a competitive model of grammatical encoding based on the interactive activation architecture (McClelland & Rumelhart 1981; Dell, 1986; Dell, Burger & Svec, 1997). A competitive model claims that alternative syntactic plans actively compete with each other. In the interactive activation framework, words and syntactic structures are represented as connected nodes in different levels. Syntactic structures are represented by the ‘structure plan’ nodes, and excitatory activation spreads from the word/lemma nodes to the syntactic structures that contain those words. For instance, activation of the verb give will spread to the ‘structure plan’ nodes of the dative construction and the double object construction. Crucially, there are inhibitory connections between the structure plan nodes. Activation in one node suppresses activation in other nodes. Thus, if two nodes representing alternative structure plans such as the double object construction and the dative construction are both activated at the same time, the two nodes would suppress each other, leading to lower overall activation levels and resulting in more difficulty in choosing a structure. Due to this competitive nature, the competitive model predicts that 8 having alternative structures leads to greater difficulty, as reflected by longer production latencies and higher error rates. That is, having syntactic options hinders production. 1.2.2 The Incremental Theory The second view is an incremental model of grammatical encoding (Bock, 1992; de Smedt, 1990; F. Ferreira, 2000; F. Ferreira & Engelhardt, 2006; Kempen & Hoenkamp, 1987; Levelt, 1989; Smith & Wheeldon, 1999; Wheeldon, Meyer & Smith, 2002). In contrast to the competitive model, an incremental model claims that grammatical encoding proceeds in a serial fashion. In this framework, syntactic structures are composed of ‘slots’ to be filled by activated lexical items. Based on the ‘first-retrieved, first-mentioned’ principle under the assumption that processing capacity is limited, a lexical item that is conceptually more accessible (i.e. easier to retrieve from memory) is usually mentioned earlier in a sentence, buying time for retrieving less accessible items. Consider a situation where the speaker attempts to express the idea that Peter gave some money to the church. After the structure ‘Peter gave…’ has been built, there are two possible items for the immediate postverbal ‘slot’: money or church. If money is inserted first, grammatical encoding can proceed by constructing a dative construction: Peter gave the money to the church. Instead, if church is inserted first, grammatical encoding can proceed by constructing a double object construction: Peter gave the church the money. In contrast, if the speaker says ‘Peter donated…’, only the word money can be inserted postverbally: Peter donated money to the church. If church is more accessible 9 than money for some reason, grammatical encoding will be delayed until money is retrieved. According to this account, syntactic flexibility facilitates production since it accommodates a greater variety of lexical activation states. Importantly, the two competing theories predict the effects of syntactic flexibility in opposite directions: If grammatical encoding proceeds in parallel, then having syntactic options would hinder production. Contrastively, if grammatical encoding proceeds in a serial fashion, then having syntactic options would facilitate production. In Chapter 2, I will provide a further discussion on the theoretical implications generated by the findings in V. Ferreira (1996). His work is of great relevance to the current work because a primary question of my research is to investigate the effects of another type of flexibility, namely, semantic flexibility, which involves the message component. Research questions centering on this issue are presented in 1.5. 1.3 Accessibility Effects in Sentence Production Another issue closely related to the grammatical component of a serial processing model is how speakers decide the linear word order in a sentence. A generally agreed-upon view is the most ‘accessible’ entity tends to claim the most prominent syntactic function (i.e., the subject). These kinds of patterns are generally termed accessibility effects. One explanation for accessibility effects is they serve the purpose of facilitating grammatical encoding. This account is based on two assumptions. First, the grammatical encoding 10 system must create structures piece by piece (i.e., a serial processing) instead of having to wait for all parts of the sentence to become accessible. If grammatical encoding proceeds in an incremental fashion, then ‘dispatching’ the accessible content sooner can buy time to access the less accessible content. Second, the grammatical encoding system can only operate unidirectionally. That is, an earlier syntactic position must be claimed before a later position. Otherwise, it would still be possible to assign a more accessible content word to a later position even if it is retrieved from memory sooner. Existing research has identified a number of factors that seem to correlate with accessibility, including syntactic priming (Bock, 1986a; V. Ferreira & Bock, 2006), semantic priming (Bock, 1986b), phonological interference (Bock, 1987), imageability (James, Thompson & Baldwin, 1973; Bock & Warren, 1985), animacy (H. H. Clark, 1965; Harris, 1978; Bock, Loebell, & Morey, 1992), given information (Carroll, 1958; Tomlin, 1983) and thematic salience or prominence (F. Ferreira, 1994; Prat-Sala and Branigan, 2000). What concerns us here is the factor of thematic prominence, first explored in F. Ferreira (1994). Thematic roles are generally hypothesized to possess different degrees of conceptual prominence and to comprise a hierarchy which is ordered by their prominence (e.g., Grimshaw, 1990; Jackendoff, 1987, 1990). As illustrated in (1), in this hypothetical ordering, AGENT is ranker highest; EXPEREINCER is immediately below, followed by LOCA TION, SOURCE, and GOAL. At the bottom of the hierarchy are THEME and 11 PA TIENT. Presumably, a noun with greater thematic prominence is conceptually more accessible, and thus is more easily retrieved from memory. (1) Agent > Experiencer> Location, Source, Goal > Theme, Patient To investigate how thematic prominence influences grammatical encoding, F. Ferreira studied language production using two groups of verbs, namely (i) normal verbs, including agent-theme verbs (e.g., criticize) and experiencer-theme verbs (e.g., adore) and (ii) theme-experiencer verbs (e.g., surprise). In her experiment, participants were provided with verbs and two nouns and asked to produce sentences. She manipulated both the verb type (normal verbs vs. theme-experience verbs) and the nouns (either two animate nouns or an animate-inanimate noun pair). She grouped agent-theme verbs and experiencer-theme verbs together as opposed to theme-experiencer verbs because in the former group of verbs, a more prominent thematic role (i.e., AGENT and EXPERIENCER) is mentioned first in active voice, wheareas in the latter group, a less prominent thematic role (i.e., THEME) is mentioned first in active voice. Her results showed (i) overall, participants produce more active than passive sentences, but that (ii) passives occur more frequently with theme-experiencer verbs (e.g., surprise) than with normal verbs. For example, speakers were more likely to say “John was surprised by Peter” than “John was hit by Peter.” Based on this finding, F. Ferreira concluded that speakers attempt to place the more prominent thematic role in subject position (e.g., 12 EXPERIENCER rather than THEME; AGENT rather than THEME). Moreover, the thematic prominence effect was larger when the two nouns differed in animacy, suggesting that an animacy contrast also helps to distinguish the prominence of the two entities. Based on F. Ferreira’s findings, another key component of my work is to explore whether thematic prominence and animacy effects can be extended to Mandarin Chinese. I will elaborate this issue in 1.5. 1.4 The Scope of Advance Linguistic Planning Sentence formulation requires the speaker to integrate information about an event with suitable sentence frames. Naturally, words need to be retrieved and inserted into the right slots of a structural frame, and the frame must have the right number of slots to accommodate these words. To identify the time course of lexical-structural integration and the process to bring content and structure together has been a notoriously difficult problem, dubbed the “coordination problem” (Bock, 1987b). Findings about the scope of linguistic planning (i.e., the amount of linguistic information that speakers prepare in advance before they initiate speaking) are relevant for debates about the coordination problem, as they are suggestive of how lexical-structural integration occurs. Although a fair amount of experimental results agree that sentence production proceeds in an incremental way, these results have made incongruent suggestions as to how incremental this process is. In some studies, radical incrementality is found, suggesting that utterances can be initiated as soon as the smallest bit of linguistic structure (i.e., a phonological 13 word) is available (e.g., Brysbaert, Fias & Noel, 1998; Wheeldon & Lahiri, 1997). In contrast, the results of other studies suggest that speakers are more cautious and do not speak until they have planned a reasonably large chunk of an upcoming utterance (e.g., Lindsley, 1975; Meyer, 1996; Stallings, MacDonald & O’Seaghdha, 1998). Due to the variability in the scope of advance planning, Bock et al. (2003, 2004) propose that advance linguistic planning can be either lexically or structurally incremental, under the assumption that either the structural process or the lexical process can take precedence and guide early sentence planning. The two proposals are briefly reviewed here. 1.4.1 Lexical Incrementality In lexically incremental processes, words may play a role in selecting the necessary structural frame. A speaker might retrieve the word corresponding to one message element at the outset of sentence formulation and continue building the sentence word by word from that point (Gleitman, January, Nappa & Trueswell, 2007). In this scenario, since the increments that can be added to build the structure at one time are limited to a sub-phrasal unit, the ease of lexical retrieval should play the most important role in early sentence planning, while the ease of building a phrasal structure only plays a secondary role. This account is compatible with the functional accounts which give lexical items a leading role in the acquisition and generation of sentence structure (e.g., Bates & MacWhinney, 1982; Bock, 1982; Tomasello, 2000; also see Garnsey, Pearlmutter, Myers, & Lotocky, 1997). 14 1.4.2 Structural Incrementality In structurally incremental processes, structural information may control the timing of lexical retrieval, with words becoming active when a structural frame calls for them. In this view, sentence formulation begins with the generation of a sentence plan that captures the relationships between various message elements (Bock et al., 2004; Kuchinsky & Bock, 2010). In this scenario, sentence formulation need not begin with the retrieval of a word; rather, a frame may be in place before lexical retrieval and may specify the order in which words are to be retrieved. This account is compatible with the abstract structural accounts that emphasize the relative self-sufficiency of structural processes in sentence production, on the assumption that structure-building procedures do not hinge upon lexical retrieval (e.g., Bock, 1990; Bock et al., 2003, 2004; Christianson & F. Ferreira, 2005; Dell, 1986; V. Ferreira & Dell, 2000; Fisher, 2002; Konopka & Bock, 2009). Since the production system is flexible, lexical and structural processing may shift as speakers try to meet different communicative goals or to adapt to the pressure in different contexts. These shifts may contribute to the variability in the planning scope (e.g., Ferreira & Swets, 2002; Oomen & Postma, 2001; Wagner, Jescheniak & Schriefers, 2010). In other words, as the extent of advance planning may be conditioned by speakers’ strategy, the grammatical encoding operations may fall somewhere along the continuum of lexical and structural incrementality. Consequently, existing findings may be more compatible with either end at different times or under different circumstances. 15 Under the assumption that the language production mechanism is flexible enough to adjust between lexical and structural processing, and is susceptible to speakers’ planning strategies, the third key component of my work is to fiuther investigate how speakers’ strategic control may affect (i) the time course of lexical-structural integration in the grammatical component, and (ii) the process of assigning thematic roles to the referents in the message component. This topic is introduced in more depth in 1.5. 1.5 Overview of This Dissertation 1.5.1 Research Questions My dissertation aims to explore the language production mechanism by studying its individual components as well as the interactions between them. The current work is composed of three key components. The first main issue concerns the effects of semantic flexibility. As already mentioned, V . Ferreira (1996) asked whether having freedom of choice in grammatical encoding (termed syntactic flexibility in his study) hinders or facilitates production. Focusing on a different domain, I investigated whether having freedom of choice in message formulation (termed semantic flexibility in this study) hinders or facilitates production. The second main issue concerns the effects of thematic prominence and animacy, in the hope of extending current findings of accessibility effects to Mandarin Chinese. The third main issue concerns the effects of speakers’ strategic control, in attempt to find out how speakers’ planning strategies in different scenarios may influence the process of sentence production. Research questions related to the three main issues are introduced respectively in the following. 16 Effects of Semantic Flexibility. The first research question asks whether having options of multiple propositions in the message component facilitates or complicates the process of sentence production. Consider a scenario: When someone is asked to describe a photograph depicting a beach scene to someone, he could talk about the child building a sand castle, the surfers in the waves, or the girl watching the child building the sand castle. Assuming forming a message is a resource-demanding process and cognitive load tend to increase with conceptual complexity, it is natural to ask what happens when more than one possible message could be conveyed? Does the decision-making about which message to convey slow down the process of message information? My research was built on existing work on syntactic flexibility in grammatical encoding. V. Ferreira (1996) investigated whether the freedom in syntactic choice hinders or facilitates the process of language production. In his experiment, speakers had options between alternative structures in some conditions but not in others. Intuitively, having a choice appears detrimental for decision-making, since competition for selection may potentially cause delays. However, contrary to intuition, Ferreira’s findings suggest that syntactic flexibility facilitates production, as indicated by lower error rates and shorter speech onset latencies. He attributed this result to the highly incremental nature of the grammatical encoding process. Contrastively, little is know about how an intended message is formed in the message component, which serves as the input for grammatical encoding. This is mostly due to the fact that concepts are abstract representations, and our thoughts are organized extemporaneously without conscious awareness. Therefore, 17 investigating how freedom of choice influences message formulation may pave way to our understanding of how ideas are put together in an abstract domain. Effects of Thematic Prominence and Animacy. The second question concerns whether the findings of accessibility effects can be extended to a different language. F. Ferreira (1994) concluded that English speakers tend to produce more passives with theme- experience verbs (e.g., “John was scared by Peter”) than with normal verbs (e.g., “John was criticized by Peter”) due to a preference for placing a more prominent thematic role (i.e., AGENT or EXPERIENCER) in an earlier sentence position. If this tendency serves the purpose of facilitating production, similar patterns should be found in a different language. My study investigated Mandarin speakers’ online production of the BA and BEI constructions. To the best of my knowledge, this is the first work to compare these two constructions from a production perspective. Moreover, BA and BEI constructions differ from English actives and passives in two ways. First, although English active voice is used far more frequently and considered more natural than passive voice in general, BA and BEI constructions are more balanced in overall frequency. Second, although English active and passive voices can be used interchangeably in most cases without affecting semantic meaning, BA and BEI sentences sometimes need to be used exclusively depending on event structure and plausibility (see Huang, Li & Li, 2009 for a comprehensive overview). For example, Zhang (2001) pointed out the asymmetric semantic relations between BA and BEI associated with the ‘affectedness’ and ‘causality’ 18 features. Therefore, it is also worth exploring how semantic constraints imposed by the structures to be produced may interact with congnitive factors such conceptual accessibility. Effects of Strategic Control. The third question asks to what extent and in what way speakers’ strategic control may influence the process of sentence production. Experimental results from existing work suggest that strategic control factors may affect speakers’ behaviors of sentence planning and production (e.g., F. Ferreira & Swets, 2002, 2005; Oomen & Postma, 2001; van Nice & Dietrich, 2003; Wagner, Jescheniak & Schriefers, 2010). Importantly, the present study aims to explore the process of message formulation as well as grammatical encoding. Consider a situation where a speaker is under time pressure and aims at speed vs. another situation where there is less time pressure and accuracy is crucial. Will grammatical encoding proceed more incrementally in the former situation given less speech preparation time? In addition, in the latter situation, will the process of message formulation be adaptable to strategic control such that speakers will cram in more information for conceptualization given more planning time? This line of research can offer new insights into the operation of the production system and advance our understanding of which part of the system is automatic or controllable. 19 To summarize, my dissertation is based on the serial language production model proposed by Bock & Levelt (1994), and focuses on the nature and processes of the message and grammatical components. This line of research can contribute to inquiries about how people produce speech when they are faced with choices between multiple propositions and alternative structures, and to what extent the language production mechanism is flexible and under control. 1.5.2 Methodology Method A series of five experiments were reported in this dissertation. They were designed in the form of sentence production task. In general, speakers saw a few meaning-carrying words, including nouns and verbs, on a computer screen and were asked to create a complete sentence with those words. Participants were also told to glance over the words and to plan out their sentence before speaking. In order to complete this task, speakers had to come up with an event based on the given entities, and to arrange words in a proper order. This approach engaged participants in activities pertaining to both the message component and the grammatical component. In addition, a production deadline was imposed on speakers’ articulation time. Participants were instructed to start speaking before the deadline. The purpose of the deadline is two-fold. First, in natural speech striving for efficiency, speakers normally do 20 not spend a long time contemplating on what to say before they initiate an utterance. Compared with an experiment setting where no deadline is required, a setting with time pressure would potentially motivate extemporaneous speech. Second, because error rate is an important indicator of production difficulty when interpreting experiment results, the incidence of observed errors could be increased under time pressure. Data preparation The difficulty level of production allows us to examine the effects in interest and to test predictions made by competing theories. For example, to examine whether flexibility hinders or facilitates production, one needs to compare the difficulty level of conditions with and without flexibility. For statistical analyses, I measured both the error rate and the production latency (i.e., the time needed for speakers to plan their utterances prior to articulation). Longer production latency indicates that certain complexity gets involved in the process of formulation. For example, it is found that as syntactic complexity increases, so does production latency (F. Ferreira, 1991; Hartsuiker & Barkhuysen, 2006). Both the production lantency and error rate data were analyzed because the principle of speed-accuracy tradeoffs suggests that the accuracy of a movement tends to decrease when its speed increases (e.g., Plamondon & Alimi, 1997; Schmidt, Zelaznik, Hawkins, Frank & Quinn, 1979; Wickelgren, 1977; W oodworth, 1899; but see Beilock, Bertenthal, McCoy & Carr, 2004 for a different observation). When it comes to speech production, 21 speakers may generate a sentence rapidly at the expense of a higher probability of speech errors (Palmer & Pfordresher, 2003), such as slips of the tongue in speech (Bock, 1995; Dell, 1986; Garrett, 1980). In contrast, higher accuracy can be achieved by delayed articulation. Thus, when interpreting the experimental results, it is necessary to examine both the accuracy and speed aspects to avoid misinterpretation. In addition to production latency and error rate, I also analyzed speech errors and speech disfluencies observed in the experiments. Although speech is generally conducted in a way aiming at both speed and accuracy, speech errors and disfluencies often occur during sentence production. This is probably because in natural speech, speakers often initiate speech before message planning or linguistic planning is completed. On one hand, speech disfluencies include behaviors such as stuttering, self-repair, repetition, hesitation, etc. There can be different sources for speech disfluencies. For example, behaviors of stuttering can suggest certain articulatory difficulties during the phonological encoding. Filled pauses and repetitions are sometimes regarded as by-products of covert repairs, resulting from errors that have been detected and corrected through inner loop monitoring at such an early point that the error does not become overt (e.g., Blackmer & Mitton, 1991; Levelt, 1983, 1989; Schlenk, Huber & Willmes, 1987; Brédart, 1991; Postma & Kolk, 1993). However, a number of different accounts for filled pauses and repetitions have been proposed, which suggest that they reflect planning problems involving conceptualizing, syntactic frame construction, and lexical entrance (e.g., Goldman-Eisler, 1968; Kowall, O’Connell & Sabin, 1975; Kemper, 1992; Christenfeld, 1994; Clark & 22 Wasow, 1998). Additionally, hesitation may be consequences of uncertainty in the message, rather than disruptions in production mechanisms themselves (e.g., Goldman- Eisler, 1968; Schachter et al., 1991). On the other hand, speech errors include errors involving different levels of representation, including lexical, semantic, phonological, morphological and syntactic errors (e.g., Fromkin, 1971, 1973; MacKay, 1970, 1972, 1992a, 1992b). To sum up, speech disfluencies and speech errors can provide valuable information for the attempt to identify different sources of difficulty in sentence production, which may not be directly observed from the results of error rate or production latency. In a nutshell, in my dissertation, experimental results were analyzed and reported both quantitatively (i.e., production latency and error rate) and qualitatively (i.e., speech disfluencies and errors) in the hope of revealing intriguing facts within. 1.5.3 Structure of This Dissertation This document is organized as follows. In Chapter 1, I first review the literature centering on sentence production, starting from the outlines of widely-accepted language production models and the characterization of individual levels or components. Next, in order to establish the theoretical background for my research interests, I first introduce existing work which investigates effects of syntactic flexibility (i.e. having freedom of choice in grammatical encoding) and introduce two competing theoretical views that make opposite predictions on the effects of syntactic flexibility. Secondly, I review the 23 literature on accessibility effects on grammatical encoding. Specifically, I introduce existing work on how accessibility effects concerning thematic prominence and animacy affect speakers’ syntactic choice in English, and argue why it is worthwhile to extend the findings of English to Mandarin Chinese. Thirdly, I present two proposals accounting for the lexical-structural integration problem in the grammatical component, and discuss how these two views could correlate with the variability in advance planning scope. Meanwhile, I propose to manipulate the factor of speakers’ strategic control to test the flexibility of different components in the language production system. Finally, a general introduction of the research questions and methodology of the experiments are provided. In Chapter 2, I report the results of a sentence production task in Mandarin Chinese (i.e., Experiment 1) designed to investigate effects of semantic flexibility and accessibility in the meantime. Experiment 1 involved producing Mandarin BA and BEI constructions containing an agent-theme verb. The experimental conditions allow us to examine effects of semantic flexibility in the absence or presence of syntactic flexibility. Moreover, the condition with syntactic flexibility allows us to examine whether speakers’ syntactic option hinges upon verb thematic structure. For example, with an agent-theme verb ‘hit’, do speakers have the tendency to say “the director criticized the movie” rather than “the movie was criticized by the director”? In Chapter 3, I report the results of Experiment 2, which was built on Experiment 1 but tested a different verb type: theme-experiencer verbs. The distinctive thematic structure 24 associated with these two verb types leads to opposite predictions about the effects of semantic flexibility and accessibility. Therefore, Experiment 2 can complement the findings of Experiment 1. In addition, I present the error analyses of these two experiments. A comparison of the findings is made, followed by a general discussion. Chapter 4 presents an English sentence production task designed to examine the effects of semantic flexibility and thematic prominence on three verb types: agent-theme verbs, experiencer-theme verbs and theme-experiencer verbs. In this design, speakers were required to produce only active or passive sentences in each condition to avoid confounds from effects of syntactic flexibility. The effects of semantic flexibility were examined by comparing agent-theme and experiencer-theme verbs, while effects of thematic prominence were examined by comparing experiencer-theme and theme-experiencer verbs. The primary goal of Experiment 3 is to further investigate whether in English, effects of semantic flexibility are also conditioned by the interaction of thematic role and animacy, or it is just a language-specific phenomenon in Mandarin Chinese. In Chapter 5, I present comparisons between the correct responses and perfect responses in Experiments 1 to 3. The purpose is to find systematic discrepancies between these two sets of data, under the assumption that their production might involve processes that vary in degrees of incrementality. The hypothesis is the scope of advance linguistic lanning may be narrower in fast speech than in prepared response. Experiments 4 and 5 were designed as the follow-ups to test this hypothesis. 25 In Chapter 6, I present Experiments 4 and 5, which were based on the design of Experiment 3 and used strategic control as the between-participants variable, under the assumption that speakers’ strategic control can potentially affect the scope of message and linguistic planning. The settings of Experiments 4 and 5 were intended to motivate speakers to aim at speed and accuracy respectively. The goal is to find out whether the processes of message formulation and grammatical encoding are subject to speakers’ control, such that effects of semantic flexibility and thematic prominence will vary in different settings. Finally, in Chapter 7, I attempt to pull the results of these experiments together and summarize the findings in this work. With guidance from the literature, I will discuss the theoretical implications of my findings on the nature of individual components as well as the interactions between different components in the hope of providing a clearer picture and better understanding of the language production system. Future research directions are proposed. 26 Chapter 2: Mandarin Sentence Production Task Using Agent-Theme Verbs 2.1 Introduction In this chapter, I present an experiment designed to investigate how the process of message formulation is influenced by semantic flexibility, i.e. having freedom of choice about what propositional meaning to convey. Previous work looked into effects of syntactic flexibility, that is, having freedom of choice about what structure to use with a fixed meaning. For instance, V. Ferreira (1996) proposed that presence of syntactic flexibility facilitates the process of grammatical encoding. He attributes the facilitative effect to the incremental nature of grammatical encoding. However, little is known about the effects of semantic flexibility and, more broadly, how message formulation is carried out during real-time language production. Unlike the process of grammatical encoding, which relies on serial processing to a large extent such as arranging words in a linear order, concepts are abstractly represented and not necessarily temporally organized. The extensively accepted ‘Parallel Distributed Processing’ (PDP) or ‘Connectionist’ models describe conceptual representations or features as a large numbers of densely interconnected units in a network with independent but parallel dimensions (e.g., Feldman & Ballard, 1982; Hinton, 1981; Hinton, McClelland & Rumelhart, 1986; McClelland, & Rumelhart, 1985, 1986; among others). It remains unclear whether availability of multiple propositions in the process of message formulation would facilitate or hinder production. For example, when describing an event of a couple kissing each other, the speaker could say “the man is kissing the woman”, “the woman is kissing 27 a man” or even “the man and the woman are kissing” (for discussion about symmetric predication, see Boadi, 1975; Dimitriadis, 2008; Levin,1993). Existing studies have shown that a number of factors can influence the speaker’s choice, including perspective taking or discourse orientation such as old or new information, figure and ground, etc. (e.g., Polinsky, 1996; Talmy, 2000). In addition to investigating what factors may contribute to the decision-making process, a fundamental issue is: given conceptual complexity can be defined in terms of the number of independent dimensions that needs to be represented and there are limitations on processing capacity or working memory capacity (e.g., Baddeley & Hitch, 1974; Cowan, 1995, 2005; Miller, 1956), do speakers welcome multiple propositional options when they are forming a message? To investigate this question, I conducted a sentence production task which manipulated the availability of semantic flexibility in its design. Examining the effects of semantic flexibility helps shed light on the process of message formulation because, similar to effects of syntactic flexibility, these effects potentially reflect whether preverbal messages are encoded in an incremental or competitive fashion. In addition to the effects of semantic flexibility, the other primary goal of Experiment 1 is to investigate how the factors of thematic prominence and animacy can influence the sentence production of Mandarin Chinese speakers. Although accessibility effects on sentence production have been extensively studied in the literature, this issue is rarely studied in Mandarin Chinese. For example, Hsu (2008) investigated the syntactic priming 28 effect on Mandarin-speaking preschoolers and concluded that such an effect was stronger in the production-to-production condition than in the comprehension-to-production condition, suggesting that a learning process is involved when preschoolers pick up the accessibility effects in production. To the best of my knowledge, the current work is the first to investigate accessibility effects concerning the interaction of thematic prominence and animacy in Mandarin Chinese. This line of research leads to a broader question: how do speakers choose from among the syntactic options available for conveying the same message? The Mandarin sentence production tasks in the current work target on the production of BA and BEI constructions. The issue of whether BA and BEI constructions are alternative structures is often addressed by Chinese linguists. It is generally observed that in some cases they can be used alternatively to express the same proposition, paralleling English actives and passives, but in other cases only one of the two constructions is legitimate. The Mandarin experiments were designed to explore this issue from a psycholinguistic perspective to see if the findings of accessibility effects can be extended to Mandarin, in the hope of demonstrating how the online syntactic options between the BA and BEI constructions are conditioned by factors such as thematic structure and animacy of nouns. 29 The chapter is structured as follows. First, I discuss existing research by V. Ferreira (1996) and F. Ferreira (1994), upon which the present study is partially built. Then, I turn to the research questions and explain the rationale underpinning my experimental design, followed by the proposed hypotheses and predictions. Finally, I report the results and discuss the theoretical implications for the research questions. 2.1.1 V. Ferreira (1996) on Syntactic Flexibility V. Ferreira (1996) investigated the effects of having syntactic options on sentence production. He compared situations where speakers were faced with a choice between two structures against situations where a speaker had no syntactic options. One of his experiments explored the production of the dative construction vs. the double object construction (e.g., “I gave the toy to the children” vs. “I gave the children the toys.”) On each trial, participants first saw the first-person pronoun and a verb (e.g., ‘I gave…’). After the words disappeared, participants then saw two or three more words, two of which were nouns, and the third, if it appeared, was a preposition (e.g., ‘toy/ /children’ or ‘toy/to/children’ ). These words appeared in a random order. Participants were instructed to create a sentence which began with “I”, continued with the verb, and then finished using all the remaining words they saw. Ferreira manipulated VERB TYPE (i.e., give- type vs. donate-type) and PREPOSITION CONTRAINT (i.e., the presence or absence of ‘to’), such that only when speakers received a give-type verb and the preposition was absent could either a dative sentence (e.g., “I gave the toys to the children”) or double 30 object sentence (e.g., “I gave the children the toys”) be uttered. When speakers received a donate-type verb and the preposition was absent, they needed to add a function word ‘to’ for sentence completion. In contrast to conditions without ‘to’, when the preposition was present, speakers could only produce a dative sentence regardless of verb type. As a whole, syntactic flexibility was only available in one of the four conditions, which he termed ‘flexible condition’ as opposed to ‘inflexible condition’. The mean error rate and production latency were compared between the inflexible condition and the flexible conditions as indicators of the difficulty level of production. In Ferreira’s design, the task is limited to the positional stage of grammatical encoding. As soon as the subject and the verb appear, participants realize the sentence to be completed is a three-place predicate, anticipating that the direct object and the indirect object would come up next to fill in the remaining argument positions. The need to add a preposition is also hinted when the verb appears (i.e. optional with give-type verbs; obligatory with donate-type verb). In other words, by the time they initiate an utterance, either one or two syntactic structures have been activated. Notably, the main task to be performed is uniform across all conditions, that is, to put two nouns in a serial order. Intuitively, having a choice is not felicitous for decision-making, since competition for selection may potentially delay this process. However, contrary to intuition, the results in Ferreira’s experiment show that speakers were less error-prone with shorter production latencies when two syntactic structure were available (e.g. with give), compared to 31 conditions where only syntactic structure was possible (e.g. with donate). Based on these results, he concluded that syntactic flexibility facilitates production, providing evidence for the incrementality-based model of grammatical encoding. Based on the ‘first- retrieved, first-mentioned’ principle, syntactic flexibility facilitates production because a faster-retrieved noun can be inserted into an earlier syntactic position when the condition allows both the dative and double object constructions. The results do not support a competition-based theory which predicts that selection time (which is directly related to the production latency) would be longer when alternative structures compete. 2.1.2 My Work on Semantic Flexibility In addition to having freedom of choice in the grammatical component, it is also possible to have freedom of choice in the message component. In my study, semantic flexibility is made available by carefully controlling the factors of plausibility and noun animacy such that it is possible to have multiple thematic-role assignments. For instance, speakers were instructed to make a sentence using a few words given. With two animate nouns director and actress and the verb criticize, semantic flexibility is present because the speaker can either say “the director criticized the actress” or “the actress criticized the director.” However, if the given nouns differ in animacy such as director and movie, the speaker can only assign AGENT role to the director due to semantic restriction. In this case, semantic flexibility is not available. Under the assumption of the incremental vs. competitive views on grammatical encoding, facilitatory effects of syntactic flexibility 32 suggest an incremental process while inhibitory effects of syntactic flexibility suggest a competitive process. Following this rationale, the findings concerning semantic flexibility effects can help gain insight into the process within the message component by suggesting how message formulation may be carried out. More detail of the experiments will be given in 2.2. 2.1.3 F. Ferreira (1994) on Thematic Prominence and Animacy According to a hypothetical hierarchy ordered by the prominence of thematic role, AGENT is ranked highest, immediately followed by EXPEREINCER, LOCA TION/SOURCE/ GOAL, with THEME and PA TIENT at the bottom of the hierarchy (Grimshaw, 1990; Jackendoff, 1987, 1990). Under the assumptions that a noun with greater thematic prominence is more accessible conceptually, and that a more accessible noun tends to be mentioned earlier in a sentence, F. Ferreira (1994) investigated the sentence production of English speakers when they were required to produce actives or passives with two verb types. One type of verbs was termed normal verbs, including agent-theme verbs (e.g., criticize) and experiencer-theme verbs (e.g., adore); the other type was termed theme-experiencer verbs (e.g., surprise). The fundamental difference of these two types of verbs lies in the reverse mappings between grammatical roles and thematic roles. In an active sentence, the more prominent thematic role (AGENT or EXPERIENCER) corresponds to the subject role for normal verbs (e.g., AGENT criticized THEME). Instead, the less prominent thematic role (THEME) 33 corresponds to the subject role for theme-experiencer verbs (e.g., THEME surprised EXPERIENCER). The mappings are reverse in a passive sentence. Accessibility effects predict that speakers would prefer to produce active sentences with normal verbs and passive sentences with theme-experiencer verbs. The results of Ferreira’s experiment showed that despite the fact that actives are dominant in numbers, passives occurred more frequently with theme-experiencer verbs (28%) than with normal verbs (6%). For example, it was more likely for speakers to say “John was surprised by Peter” than for them to say “John was hit by Peter”. This is because in the former sentence, passive voice allows the EXPERIENCER to occur earlier than THEME, so the word order conforms to the proposed thematic hierarchy. Additionally, it is generally agreed that an animate noun is conceptually more accessible than an inanimate noun (Bates & MacWhinney, 1982; Bock, 1982; Bock, Loebell & Morey, 1992; MacWhinney, 1977). Taking animacy into consideration, F. Ferreira included two types of noun pair in her experiments. In one type, the two nouns were both animate. In the other type, the noun pair consisted of an animate noun and an inanimate noun. The results demonstrated that passives occurred more frequently when the two nouns differed in animacy rather than both being animate, indicating that an animacy contrast also helps to distinguish the prominence of the two entities. Based on her findings, F. Ferreira concluded that speakers attempt to place a more prominent thematic role in subject position, especially when it is animate, thus resulting 34 in a higher proportion of passives with theme-experiencer verbs. As Bock (1986) argued, once the activated concept grabs the earliest syntactic position in the sentence, the language production system then adjusts the structure of the sentence to accommodate that placement. Similar views are held among other researchers (e.g., de Smedt, 1996; Kempen and Hoenkamp, 1987; Levelt, 1989; van Nice & Dietrich, 2003). Crucially, the language production model assumes that role assignment occurs in the massage component (see V. Ferreira & Slevc, 2007 for an introduction). Generally, the conceptual level that feeds linguistic processing is assumed to involve a kind of mental model. It is worth mentioning that in a mental-models approach, roles are assigned to referents in the model, not to linguistic expressions (e.g., Garnham 1997, 2001). Animacy is a conceptual feature of a referent, and thus it can affect the accessibility of the referent in the mental model. Moreover, thematic-role assignment involves assignment of conceptual-level roles to the same referent. Consequently, animacy and thematic roles have potential to interact at the model level. However, thematic roles are linguistic notions related to verb thematic structure. A large body of literature has demonstrated that argument structure information stored with the verb is made immediately available during sentence processing (e.g., Ahrens, 2003; Boland, Tanenhaus & Garnsey, 1990; Boland, Tanenhaus, Garnsey & Carlson, 1995; Shapiro, Brookins, Gordon & Nagel, 1991). For example, Shapiro et al. (1991) used cross-modal lexical decision (CMLD) studies and found that normal listeners show exhaustive activation of a verb’s arguments upon hearing the verb. In F. Ferreira’s experiments, 35 participants first ‘read’ the words (as opposed to picture-naming tasks) and then planned a sentence. Thus, it is not surprising that linguistic information such as noun animacy and verb argument structure is available at the conceptual level. This suggests that speakers are sensitive to the semantic properties of lemmas at a very early stage of production. 2.1.4 Mandarin BA and BEI Constructions The experiment in this chapter elicited the production of BA and BEI sentences in Mandarin. Although the BA and BEI constructions are two of the most studied constructions in Mandarin Chinese, the present study is the first to investigate the two constructions side by side with an online production method. Prior studies mostly focus on describing the semantic or syntactic representations of the two constructions, while my work concerns the two constructions in the production domain. For discussions on the BA construction, one can refer to Chao (1968), Cheng (1986), Frei (1956), Goodall (1987, 1990), Hashimoto (1971), Hsueh (1987), Huang (1974), Koopman (1984), Liu (1997), Li & Thompson (1981), A. Li (1990, 2005), Mei (1978), among others. For the BEI construction, one can refer to Chiu (1995), Feng (1995), Hashimoto (1969, 1987), A. Li (1985, 1990), Ting (1995, 1996), Travis (1984), Wang (1970), Wei (1994), among others. Additionally, Huang, Li & Li (2009) provides a comprehensive review for the semantic and syntactic properties of the two constructions. 36 What is relevant to the current work is when carefully controlling plausibility, the two constructions can convey the same proposition, as shown in (1a) and (1b). (1) a. Zhangsan BA Lisi chi-ku le. Zhangsan THEME BA Lisi EXP . anger-cry ASP ‘Zhangsan angered Lisi such that Lisi burst into tears.’ b. Lisi BEI Zhangsan chi-ku le. Lisi EXP . BEI Zhangsan THEME anger-cry ASP ‘Lisi was angered by Zhangsan and burst into tears.’ In (1), the verb phrase is composed of a theme-experiencer verb chi ‘to anger’, followed by a resultative complement ku ‘cry’, which indicates the resulting state of the EXPERIENCER, Lisi, after he got angered. The subject Zhangsan in (1a) is the THEME who caused Lisi to become angry. In comparison, (1b) expresses the same proposition with a different syntactic structure. In (1b), though the thematic role mappings remain the same, the grammatical roles for Lisi and Zhangsan are switched. Note that when the THEME is animate, the action may or may not be performed volitionally. Therefore, in (1), Zhangsan made Lisi cry either deliberately or accidentally. Thus, an inanimate noun which lacks volition can also be the THEME, as shown in (2a) and (2b). 37 (2) a. Yaoyan BA Lisi chi-ku le. rumor THEME BA Lisi EXP . anger-cry ASP ‘Rumor angered Lisi such that Lisi burst into tears.’ b. Lisi BEI yaoyan chi-ku le. Lisi EXP . BEI rumor THEME anger-cry ASP ‘Lisi was angered by rumor and he burst into tears.’ In addition to theme-experiencer verbs, the two constructions can also contain an agent- theme verb, as shown in (3a) and (3b). (3) a. Zhangsan BA Lisi piping-le-yidun Zhangsan AGENT BA Lisi THEME criticize-ASP-quite a bit ‘Zhangsan criticized Lisi quite a bit.’ b. Lisi BEI Zhangsan piping-le-yidun Lisi THEME BEI Zhangsan AGENT criticize-ASP-quite a bit ‘Lisi was criticized by Zhangsan quite a bit.’ In (3), the verb phrase is composed of an agent-theme verb piping ‘to criticize’, followed by a degree adverbial phrase yidun “quite a bit”. The AGENT subject Zhangsan in (3a) criticized the THEME object Lisi. In (3b), to express the same proposition with bei, the 38 two thematic roles switch syntactic positions. Again, the THEME can also be inanimate, as shown in (4a) and (4b). (4) a. Zhangsan BA dianying piping-le-yidun Zhangsan AGENT BA movie THEME criticize-ASP-quite a bit ‘Zhangsan criticized the movie quite a bit.’ b. Dianying BEI Zhangsan piping-le-yidun movie THEME BEI Zhangsan AGENT criticize-ASP-quite a bit ‘The movie was criticized by Zhangsan quite a bit.’ To summarize, examples (1)-(4) demonstrate the fact the BA and BEI constructions can be used alternatively to convey the same relational meaning of an event, paralleling the pattern of active voice and passive voice in English. Moreover, these examples demonstrate that either agent-theme verbs or theme-experiencer verbs are compatible with the two constructions, when the two nouns are both animate or differ in animacy. A summary of the distributions of thematic roles in the BA and BEI constructions is given in Table 2. 1. 39 Table 2.1: The distributions of thematic roles in the BA and BEI constructions (shaded areas indicate the preferred word order predicted by thematic prominence) Crucially, in the two constructions, the thematic roles have reverse syntactic positions. This property provides us with the opportunity to look into the effects of thematic prominence (F. Ferreira, 1994). Specifically, the tendency to mention more prominent thematic roles earlier in the sentence predicts that depending on verb thematic structure, one construction (shaded in the table) would be favored over the other construction. Example 1(a) & 1(b) Theme - Experiencer verbs grammatical role Subject BA Object V thematic role THEME EXPERIENCER grammatical role Subject BEI Object THEME V thematic rol e EXPERIENCER Example 3(a) & 3 (b) word order Agent-Theme verbs grammatical role Subject BA Object V thematic role AGENT THEME grammatical role Subject BEI Object V thematic role THEME AGENT word order 40 2.2 Experiment 1: Agent-theme Verbs 2.2.1 Methodology Method Participants saw words on a computer screen and were instructed to create a complete sentence using those words (see F. Ferreira 1994; V. Ferreira, 1996 for work using a similar task). On critical trials, the screen showed (i) a verb (agent-theme verb in Experiment 1), (ii) two nouns and (iii) sometimes a preposition, depending on the condition. Participants were instructed to plan their sentence before speaking. Producing a sentence requires speakers to come up with an event based on the given entities, and to arrange the words in a proper order. Thus, this task engages both the message component and the grammatical component. Participants Thirty students from National Chi-nan University in Taiwan participated. All of them are native speakers of Mandarin Chinese spoken in Taiwan. They were paid NT$100 for participating in this experiment. Design The experiment had a 2x3 factorial design. I manipulated (i) the animacy of the nouns on the screen and the (ii) presence/absence of a preposition on the screen. In the case of 41 animacy, we manipulated whether the two nouns on the screen (i) were both animate (SA=same animacy, e.g. judge vs. lawyer), or (ii) differed in animacy (DA=different animacy, e.g. judge vs. evidence). Manipulating animacy allows us to condition the availability of semantic flexibility, which only exists when both nouns are animate (i.e. the SA conditions). The second factor is the presence/absence of prepositions, referred to as the preposition constraint. 1 Thus, the screen included (i) ba, (ii) bei, or (iii) no preposition (no-prep.). As a result, two-thirds of the target trials contained four words (verb, noun1, noun2, ba/bei) and one-third contained only three words (verb, noun1, noun2). Manipulating the presence of prepositions allows us to condition syntactic flexibility: When ba or bei is given, speakers have no choice about which structure to produce: the structure to be produced is already specified. In contrast, when no preposition is given, speakers have freedom in choosing between BA and BEI constructions. In a nutshell, this 2 x3 design allows us to investigate the effect of semantic flexibility in the presence/absence of syntactic flexibility. The conditions are summarized in Table 2.2 and an explanation is given for each condition below. 1 Our use of the term ‘preposition constraint’ is simply for convenience’s sake. We do not mean to suggest that ba and bei in Mandarin Chinese are prepositions. Analyses of the syntactic category of ba and bei are quite controversial and beyond the scope of the present study (c.f. Huang, Li & Li, 2009). 42 Table 2.2: The experimental conditions of Experiment 1 Preposition constraint Animacy No preposition BA BEI Different animacy (DA) Words given judge, evidence, ignored judge, evidence, ignored, BA judge, evidence, ignored,BEI No. of syntactic choices at least 2 1 1 Preposition constraining? No Yes Yes Semantically flexible? No [-SemFlex] No [-SemFlex] No [-SemFlex] Syntactically flexible? Yes [+SynFlex] No [-SynFlex] No [-SynFlex] Same animacy (SA) Words given judge, lawyer,ignored judge, lawyer, ignored, BA judge, lawyer, ignored, BEI No. of syntactic choices at least 4 2 2 Preposition constaining? No Yes Yes Semantically flexible? Yes [+SemFlex] Yes [+SemFlex] Yes [+SemFlex] Syntactically flexible? Yes [+SynFlex] No [-SynFlex] No [-SynFlex] To begin with, the different-animacy (DA) conditions--DA-ba and DA-bei--are inflexible because only one sentence is possible (e.g. “judge BA/BEI evidence ignored.”) All the other conditions have some amount of flexibility. As mentioned above, there are two sources for freedom of choice: semantic flexibility and syntactic flexibility. In the same- animacy (SA) conditions -- SA-ba and SA-bei, where both arguments are animate, there are two ways of doing thematic-role assignment. The speaker can make either lawyer or judge the AGENT of the sentence (e.g., “ lawyer BA judge ignored” or “ judge BA lawyer ignored”). However, since ba or bei is given, syntactic flexibility is not available because the structure has been specified. Contrastively, in the DA condition where no preposition is given--DA-no prep, only one proposition is possible (i.e., judge but not evidence can 43 be AGENT). However, the speaker can choose between alternative structures (e.g., ba or bei or even other prepositions such as jiang or gei). Finally, in the SA-no prep. condition, where the two nouns are both animate and no preposition is given, freedom of choice results from both semantic and syntactic flexibility. Participants have options about what proposition to convey, as well as what structure to use. One primary goal of this experiment is to investigate the effects of semantic flexibility. It is worth mentioning that in a sentence production task, there is no way to look into the message component exclusively since production always involves the grammatical component and the phonological component. Based on this experimental paradigm, the effects of semantic flexibility can be indirectly observed in two ways. First, one can compare the SA-no prep. condition ([+SemFlex,+SynFlex]) with the DA-no prep. condition ([-SemFlex, +SynFlex]), where syntactic flexibility is present in both conditions but only the SA-no prep. condition has semantic flexibility. Second, one can compare the SA-ba/bei condition ([+SemFlex, -SynFlex]) with the DA-ba/bei condition ([-SemFlex, -SynFlex]), where syntactic flexibility is absent in both conditions but only the SA-ba/bei condition has semantic flexibility. 44 Materials Targets. 30 verb phrases were created, consisting of an agent-theme verb such as piping “to criticize” followed by a complement including a locative phrase, a resultative phrase, or a degree adverbial. Since Chinese has canonical SVO order, using a complex verb phrase instead of a bare verb enhances the naturalness of ba and bei sentences. This is because Chinese is subject to a postverbal constraint, which generally allows only one postverbal constituent. Therefore, when the verb is already followed by a complement, the object tends to be placed preverbally following ba or bei (Chao,1968; C.-T. Huang, 1982; A. Li, 1990). Verb phrases were 4 to 6 characters long. Each verb phrase was assigned two sets of noun pairs. The animate-animate pair had two animate common nouns which were matched carefully to avoid bias for thematic-role assignment in an event (e.g., painter/ writer/ frighten). That is, both nouns can either be a plausible agent or patient for the verb they occur with. In contrast, the inanimate noun of the animate-inanimate pair was semantically connected to the verb (e.g., painter/accident/frighten) to serve as a plausible theme. In each pair, the nouns were of the same length (2-3 characters). Each target item resulted in six frames based on the 2x3 design. The resulting 180 frames were assigned into six lists and rotated through the six conditions based on a 6x6 Latin square design. Participants saw each verb once in one of the six conditions, but encountered all six conditions across the 30 targets, amounting to 5 different targets per condition. A total of 45 150 trials across participants were collected for each condition. See Appendix I for a list of the target items used in Experiment 1. The verb phrase was placed at the top, the two nouns were in the middle of the screen, on the left and right side, and the preposition (if present) was at the bottom, as shown in Figure 2.1. Left/right position of the nouns was counterbalanced to avoid reading bias. The fillers had the same layout as the targets. 女明星女明星女明星女明星 actress 名導演名導演名導演名導演 director 批評了一頓批評了一頓批評了一頓批評了一頓 criticize quite a bit 把把把把 BA Figure 2.1: Layout of one experimental trial in Experiment 1 Fillers. Experiment 1 contained 50 filler items. To add variability to the sentence types participants need to create, five different types of fillers were constructed. The first type included 8 fillers, consisting of an adverb, an animate or inanimate noun, and an 46 intransitive verb. Participants did not need to add any word to this type of sentences. The second type included 8 fillers, consisting of an adjective, an animate or inanimate noun, and an intransitive verb. For this type, participants were expected to add a progressive marker zai for a complete sentence. The third type consisted of an animate-animate noun pair an animate-inanimate noun pair, and a verb phrase containing a ditransitive verb and a direct object. There were 16 fillers of this type. Half of them contained the preposition gei ‘to’, so participants did not need to add any word. In contrast, the other half contained no preposition, so participants needed to add gei in order to complete the sentence. The fourth type included 10 fillers, consisting of two animate nouns and an intransitive verb. In half of this type, a conjunction word he or yu ‘and’ was given, so no words needed to be added. In the other half, no conjunction word was given, so participants needed to add a conjunction word themselves. Finally, there were 8 items in the fifth type. Half of them consisted of an animate noun, an intransitive verb and a locative phrase. Participants needed to add the preposition zai ‘in’ or ‘at’ for sentence completion. In the other half of the fifth type, the preposition zai was given. Norming Study for Targets. To test the validity of the target items used in the sentence production experiment, a norming study was carried out before Experiment 1. The purpose is to ensure that each target item is equally acceptable and plausible in both ba and bei sentences. Thirty-two native Mandarin speakers in Taiwan participated in an internet-based survey, conducted with Qualtrics software. They were randomly assigned 47 to one of four lists in the form of a questionnaire. None of the participants participated in Experiment 1. The norming study had a 2x2 factorial design with two factors: (i) ANIMACY: whether the two nouns are both animate (SA) or differed in animacy (DA), and (ii) PREPOSITON (i.e., ba or bei). A total of 28 items were randomly selected from the 30 targets used for Experiment 1 to fit in a 4x4 Latin square design. In addition to the 28 target items, 35 fillers were constructed to add variability to the questionnaire. The fillers varied in degree of plausibility and grammaticality, including 12 syntactically ill- formed sentences, 12 well-formed sentences with semantic anomaly, and 11 well-formed sentences without semantic anomaly. Participants were asked to rate the acceptability and plausibility of each sentence based on a 1-4 scale, where 1 means ‘totally unacceptable and incomprehensible’ and 4 means ‘totally acceptable and comprehensible’. An example for each condition is shown below: (i) The SA-ba condition Mingdaoyan BA numingxin piping-le yidun. famous director BA actress criticize-ASP quite a bit “The famous director criticized the actress quite a bit.” (ii) The SA-bei condition Mingdaoyan BEI numingxin piping-le yidun. famous director BEI actress criticize-ASP quite a bit “The famous director was criticized by the actress quite a bit.” 48 (iii) The DA-ba condition Mingdaoyan BA xindianying piping-le yidun. famous director BA new movie criticize-ASP quite a bit “The famous director criticized the new movie quite a bit.” (iv) The DA-bei condition Xindianying BEI mingdaoyang piping-le yidun. new movie BEI famous director criticize-ASP quite a bit “The new movie was criticized by the famous director quite a bit.” Although it is possible to alternate the argument position of the nouns when both nouns are animate, I did not manipulate this factor in the norming study. The assumption is that when both nouns are animate, switching argument positions does not affect plausibility. For instance, “the director criticized the actress” and “ the actress criticized the director” are both plausible. The subject noun in each target of the norm study was selected randomly from the two animate nouns in a pair. In contrast, when one noun is animate and the other is inanimate, switching argument positions always leads to semantic anomaly (e.g., # “The wine bottle took the sailor to the door.”) The average rating for each condition is shown in Figure 2.2. Paired sample t-tests revealed that no significant differences existed among the four conditions both by 49 subjects and by items (all ps>.05). The average rating for the DA-ba condition (M=3.725, SE=.046) was not significantly higher than that for the DA-bei condition (M=3.598, SE=.072, t1(31)=1.446, p=.158; t2(27)=1.656, p=.109), for the SA-ba condition (M=3.597, SE=.065, t1(31)=1.869, p=.071; t2(27)=1.690, p=.103), or for the SA-bei condition (M=3.607, SE=.056, t1(31)=1.712, p=.097; t2(27)=1.161, p=.256). In addition, no significant differences were found between the DA-bei condition and the SA-ba condition (t1(31)=.013, p=.990; t2(27)=.003, p=.998), or the SA-bei condition (t1(31)= -.127, p=.900; t2(17)=-.071, p=.944). Finally, the SA-ba condition and the SA-bei condition did not differ significantly either (t1(31)=-.139, p=.890, t2(27)=-.086, p=.932). 50 3.725 3.598 3.597 3.607 1 2 3 4 The DA-ba condition The DA-bei condition The SA-ba condition The SA-bei condition rating Figure 2.2: Average rating of acceptability and plausibility for each condition in the norming study ( 1 = ‘totally unacceptable and incomprehensible’ ; 4 = ‘totally acceptable and comprehensible’) Additionally, the average rating as a function of animacy and preposition in the norming study is shown in Figure 2.3. ANOVAs revealed that there was no main effect of ANIMACY (F1(1,31)=1.272, Mse=.106, p=.268; F2(1,27)=.612, Mse=.102, p=.441). In addition, there was no main effect of PREPOSITION (F1(1,31)=1.056, Mse=.110, p=.312; F2(1,27)=.936, Mse=.103, p=.342). The two factors did not interact either (F1(1,31)=2.135, Mse=.141, p=.154; F2(1,27)=1.771, Mse=.129, p=.194). 51 1 2 3 4 ba bei Preposition Average rating Different animacy Same animacy Figure 2.3: Average rating as a function of animacy and preposition in the norming study To conclude, the results of this norming study suggest that the target items constructed for Experiment 1 are equally acceptable and plausible in either a BA or BEI sentence. Therefore, any differences found between conditions in Experiment 1 cannot be attributed to the acceptability and plausibility factors. Post-hoc Norming study. To find out whether the paired animates in each target item potentially caused a bias toward a certain thematic relation, a post-hoc norming study was conducted in the form of questionnaire. The purpose of the norming study was to see if for any of the target items, speakers would favor one proposition over the other when semantic flexibility is available. For example, would they prefer to say “the boy held the 52 girl tightly” rather than “the girl held the boy tightly”, or “the boy was held by the girl tightly” rather than “the girl was held by the boy tightly?” If such a preference were to exist, it could possible create confounds for the dependent variable (i.e. production latency), since more plausible sentences would presumably be easier to formulate conceptually than less plausible sentences. The target items of Experiment 1 were grouped into four lists based on a 2 x 2 design, contingent upon the position of a noun (i.e. left or right) and sentence type (i.e. ba or bei sentences). Thirty-five filler items with varied degrees of plausibility and various sentences types were included in addition to the target items. A total of 40 native speakers of Mandarin participated in this norming study (10 people per list). They were instructed to rate each sentence on a 4-point scale (1=implausible, 4=plausible). The study was conducted over the internet using Qualtrics software. To analyze participants’ responses, independent samples t-tests were conducted on each target item within sentence type to see if the plausibility rating differed when the two nouns exchanged thematic roles. Of the 30 target items used in Experiment 1, marginally significant differences were found on only 3 items with ba (.05<ps<.1) when the two nouns exchanged positions. No significant differences were found on the remaining 27 items (all ps >.1) for switching noun positions in either ba or bei sentence. The results indicate that in most cases, it was equally likely for the paired animate nouns to take either thematic role. In other words, the influence caused by the bias toward one of the two propositions when semantic flexibility was available was quite limited in Experiment 1. 53 Procedure Participants sat in front of an ASUS U6 laptop. Spoken responses were recorded with a head-mounted USB microphone. The experiment was run using Paradigm developed by [Perception Research Systems]. Each trial was initiated and ended by pressing the space bar. The words appeared on the screen simultaneously. The experiment began with instructions and 12 practice trials, followed by 80 test trials. Participants were instructed to create well-formed sentences that contained every word shown on the screen. They were told that sometimes it might be necessary to add a function word. However, they were not told explicitly when to add function words or which function words to add. In fact, it is necessary to add a function word on some filler trials and on all target trials of the no preposition condition. Participants were also instructed that they could freely arrange word order, but were not supposed to insert a noun into a verb phrase. Importantly, participants were reminded to produce their utterances as fast as they could without sacrificing accuracy. Additionally, to impose time pressure, after the words had appeared on the screen for 5 seconds, a 10000 Hz tone beeped for 350 ms. Participants were encouraged to complete their utterances before the onset of the tone. The entire experiment took approximately twenty minutes. The recordings were then transcribed and analyzed. The dependent variables are production latency and error rate of each condition as they serve as indicators of production difficulty level. Notably, in conditions where two nouns are both animate (i.e. the SA conditions), an error would rarely occur since whichever 54 noun is mentioned first does not affect the plausibility of that sentence. Therefore, production latency may be a better indicator than error rate for these conditions. In addition, the proportions of ba and bei sentences in the no-prep. Conditions are also studied as they reflect speakers’ preference for one construction over the other when faced with syntactic options. 2.2.2 Hypotheses and Predictions The first research question of Experiment 1 concerns the effects of semantic flexibility with agent-theme verbs. The hypothesis is that linguistic information such as thematic structure and noun animacy can interact and guide the process of sentence production very early on (i.e., in the message-formulation level). In other words, when deciding the relational meaning, the speaker would be sensitive to the thematic structure determined by verb type. The thematic structure of an agent-theme verb is associated with AGENT and THEME/PA TIENT roles. Animacy-role affinities have long been noted for certain thematic roles (e.g., Dowty, 1991; Foley & van Valin, 1984; Gruber, 1967, 1976; Jackendoff, 1972; Lakoff, 1968; Lyons, 1977). It is often claimed that the stereotypic agent is animate and the stereotypic patient inanimate. For instance, Jackendoff (1972:32) asserted that “only animate NPs can function as Agents.” Gruber (1976:158) also asserted that “if the subject is nonAnimate the possibility of Agent does not exist.” Empirically, in his Swedish corpus study, Dahl (1997) found that, with transitive sentences, 92% of the agent subjects were animate and only 9% of the (nonreflexive) patient/theme direct objects were animate. A similar result held for English sample from the London–Lund 55 corpus (Svartvik and Quirk 1980), where 88% of the agent subjects were animate versus only 11% of the patient/theme objects were animate. This animacy distribution is role related, rather than grammatical function related. For copular (stative) verbs, only 42% of the Swedish subjects and 45% of the English subjects were animate. In other words, there is a strong inclination for the AGENT role to be realized by an animate noun. Under the assumption that two animate nouns are equally accessible, they would compete for the AGENT role. According to the competitive theory, competition for selection leads to production difficulty and delays production. In contrast, competition for AGENT will not take place between an animate noun and an inanimate noun. Therefore, the prediction is that semantic flexibility hinders production with agent-theme verbs as the two animate nouns compete for the AGENT role. Recall that in our experimental paradigm, semantic flexibility is conditioned by animacy and only available when two nouns are both animate (i.e. the SA conditions), but not available when two nouns differ in animacy (i.e. the DA conditions). Specifically, if competition exists in the SA consitions, we expect to find higher error rates and longer production latencies in the SA conditions than in the DA conditions. The second goal is to investigate the effects of thematic prominence and animacy by means of comparing the production of ba and bei sentences. The hypothesis is that due to the tendency to place a more prominent thematic role in an earlier sentence position, with agent-theme verbs, ba-sentences (i..e. AGENT is mentioned first) would be easier to produce than bei-sentences (i.e. THEME is mentioned first). Specifically, we expect to 56 find lower error rates and shorter production latencies in the ba conditions than in the bei conditions. Moreover, animate nouns are more accessible than inanimate ones and tend to be mentioned first due to animacy effects. Therefore, the difference in production latency between the ba and bei conditions should be greater when the nouns differ in animacy (i.e., the DA condition) than when the nouns are both animate (i.e., the SA conditions), since it would be harder to say a bei sentence starting with an inanimate subject than an animate subject (e.g., “the movie BEI the director criticize” vs. “the actress BEI the director criticize.”) Moreover, when speakers have freedom to choose between alternative structures (i.e., the no-prep. conditions), I predict to see more ba sentences than bei sentences with agent-theme verbs, because in a ba sentence, AGENT is mentioned earlier than THEME. Taken these two hypotheses together, the predictions are summarized as follows. First, based on the hypothesis concerning effects of semantic flexibility, we expect to see: (i) The SA-no prep. condition ([+SemFlex], [+SynFlex]) would have higher error rates and longer production latencies the DA-no prep. condition ([-SemFlex], [+SynFlex]). (ii) The SA-ba/bei condition ([+SemFlex], [-SynFlex]) would have higher error rates and longer production latencies than the DA-ba/bei condition ([-SemFlex], [-SynFlex]). 57 Second, based on the hypothesis concerning effects of thematic prominence and animacy, we expect to see: (i) Due to thematic prominence effects, the SA-ba and DA-ba conditions would have lower error rates and shorter production latencies than the SA-bei and DA-bei conditions, respectively. (ii) Due to animacy effects, the difference in production latency and error rate between the DA-ba and DA-bei conditions would be greater than those between the SA-ba and SA-bei conditions. (iii) In the SA-no prep. and DA-no prep. conditions, where syntactic options are available, the proportion of ba sentences would be higher than that of bei sentences. 58 2.2.3 Results 2.2.3.1 Data Preparation Exclusions and Replacements Production latencies more than 2.5 standard deviations from the mean in each condition were removed from subsequent analyses (3.1% of correct responses). One item was excluded in the item analysis because of missing data due to programming error. In addition, as with many production tasks, the high error rate left four cells in the item analysis without any correct observations. The means for these empty cells were estimated using the row and column means 2 as in Winer (1971). Coding the Responses Different response types were coded according to the content of the actual utterance to evaluate the difficulty participants experienced in their production. Participants’ utterances were grouped into two main categories: correct responses and errors (See V. Ferreira (1996) for similar but not identical categorizations in his analysis). A correct response includes the following subcategories: (1) Perfect responses: the sentence contained all given words, was well-formed, fluently produced and semantically plausible. (2) Correct but imperfect responses: the sentence contained all the words and 2 In the item analysis, the rows are arranged by items and the columns are arranged by conditions. An empty cell results from missing data in any condition of a certain target item. For example, it is possible that all the five responses collected for the SA no-prep condition of target no. 1 were errors and thus no data of production latencies were available for this cell. 59 was well-formed and semantically plausible, but accompanied by speech disfluency such as hesitation, stuttering, repetition, or self-correction (e.g. uh…the gard...gardener was hurt by the sailor). (3) Correct but unexpected responses: in the no prep. conditions, a sentence could still be well-formed by adding a preposition such as jiang or gei instead of ba or bei to achieve the same meaning. It was also possible to add a conjunction word to some items in the no-prep. conditions, especially when both given nouns were animate (e.g. The gardener and the sailor were hurt). An error is defined as one of the following criteria: (1) Element missing: the participant failed to use one or more of the given words. (2) Wrong preposition: the participant failed to use the given preposition. (3) Ungrammaticality: in the no-prep. conditions, the resulting sentence with SVO order was marginal or ungrammatical. (4) Object incorporation: Participants were instructed not to incorporate the direct object into a verb phrase (e.g. in The writer sent the painter to school, the painter is incorporated into the verb phrase sent to school). (5) Role reversal: the resulting sentence had reversed thematic role assignment (e.g. #The movie criticized the director instead of The director criticized the movie). (6) No response: nothing was said by the participant. Calculating Production Latencies Production latencies were calculated manually in Praat. Due to frequent speech disfluencies, two sets of production latency were collected for analysis. For the first set, 60 the production latency was measured from the onset of the presentation of a trial (i.e. when the words appeared on the screen) to the onset of the total recorded utterance (e.g. the onset of the first sailor in Sailor…the gardener hit the sailor.) For the second set, the production latency was measured from the onset of the presentation of a trial to the onset of the first word of the correct sentence (e.g. the onset of the gardener in Sailor…the gardener hit the sailor.) In my analysis, mean production latency for correct responses was calculated based on the second set of data, including the time of hesitation, repetition, self-correction or stuttering. The duration of speech difficulty is considered as part of the formulation time of a sentence because it may indicate ongoing speech planning. Under time pressure, participants tend to initiate an utterance before linguistic planning is complete. For example, F. Ferreira and Swets (2002) suggested that speakers produced sentences more incrementally under a production deadline. Thus, the onset of the first word of the correct sentence may serve as a better ending point for calculating production latencies. 61 2.2.3.2 Results Three sets of results are reported. The first set concerns the proportions of different sentence types in each condition. Following V. S. Ferreira (1996), these results can be used to empirically evaluate whether participants exploited the availability of syntactic flexibility available to them. The second set concerns the error rate for each condition. The third set concerns the mean production latency for correct responses in each condition. Proportions of Sentence Types Table 2.3 summarizes the proportions of different sentence types participants produced, including both correct responses and errors in each condition. As can be seen, the no-prep. conditions exhibited a smaller difference in the proportions of ba and bei sentences and a greater number of sentence types than the ba and bei conditions, validating the assumption that a variety of syntactic options exists in the no-prep. conditions, while syntactic options are not available when ba or bei is present ([-SynFlex]). Moreover, the prediction that ba sentences would occur more frequently than bei sentences with agent-theme verbs is confirmed. In the DA-no prep. condition, ba sentences occurred more frequently (45.3%) than bei sentences (27.3%) (X 2 =6.688, p<.01). In the SA -no prep. condition, ba sentences also occurred more frequently (27.3%) than bei sentences (17.3%) (X 2 =3.358, p=.06). The results indicated that the effect of 62 thematic prominence indeed played a role. In general, participants preferred to place AGENT instead of THEME in the subject position. This explains why the proportions of ba and bei sentences are not close to even although syntactic options are available. Note that the percentage of using conjunction in the SA-no prep. condition was quite high (34.7%). This tendency could be due to an effect of syntactic persistence (Bock, 1986; Bock, Loebell & Morey, 1992) as 10 filler items were of the conjunction type. It could also be attributed to speakers’ tendency to assign both animate nouns for the AGENT role and subject position assuming the two nouns are equally accessible conceptually. Table 2.3: Number and percentage of sentence types in each condition of Experiment 1 Preposition constraint Animacy No preposition BA BEI Different animacy (DA) BA-construction 68 (45.3) 149 (99.3) 1 (0.7) Bei-construction 41 (27.3) 1 (0.7) 145 (96.7) Conjunction 1 (0.7) 0 0 O ther preposition 19 (12.7) 0 0 SVO 16 (10.7) 0 2 (1.3) N/A 5 (3.3) 0 2 (1.3) Total 150 (100) 150(100) 150 (100) Same animacy (SA) BA-construction 41 (27.3) 143 (95.3) 0 Bei-construction 26 (17.3) 0 142 (94.7) Conjunction 52 (34.7) 0 0 O ther preposition 10 (6.7) 0 0 SVO 15 (10) 1 (0.7) 0 N/A 6 (4) 6 (4) 8 (5.3) Total 150 (100) 150 (100) 150 (100) Note. Percentage in parentheses. Total possible number of observation in each cell is 150. N /A refers to incomplete utterances or no responses. 63 Error rate The error rate for each condition is shown in Figure 2.4. The figure shows that the two factors ANIMACY and PREPOSITION CONSTRAINT were interacting. The SA condition resulted in a much higher error rate than the DA condition when no preposition was shown on the screen. In contrast, when the screen display included either ba or bei, the difference became much smaller: the SA-ba condition had slightly more errors than the DA-ba condition, while the DA-bei condition had slightly more errors than the SA-bei condition. The analyses of variance (ANOVAs) demonstrated that the main effect of ANIMACY was highly reliable (F1(1,29)=8.74, MSE=.217, p<.05; F2(1, 28)=6.535, MSE=.207, p<.05). The main effect of PREPOSITION CONSTRAINT was also highly reliable (F1(2,58)=87.269, MSE=2.340, p<.001; F2(2,56)=44.068, MSE=2.562, p<.001). The interaction between the two factors was highly reliable (F1(2, 58)=19.057, MSE=.439, p<.001; F2(2,56)=23.602, MSE=.557, p<.001). 64 0 0.1 0.2 0.3 0.4 0.5 no preposition ba bei Preposition constraint Production error rate Different animacy Same animacy Figure 2.4: Mean error rate as a function of animacy and preposition constraint in Experiment 1 Additionally, to gain a better understanding of what differences are driving the interaction between ANIMACY and PREPOSITION CONSTRAINT, I took a closer look at the different conditions by means of planned comparisons. Semantic Flexibility. First, to test the effects of semantic flexibility, a paired samples t- test was conducted between the SA-no prep. condition and the DA-no prep. condition. The prediction is that semantic flexibility hinders production with agent-theme verbs as the two animates tend to compete for AGENT, so the SA-no prep. condition would have a 65 higher error rate than the DA-no prep. condition. This prediction was verified. Specifically, the SA-no prep. condition had a much higher error rate (45.3%) than the DA-no prep. condition (20%) with a significant difference (t1(29)=4.347, p<.001; t2(28)=4.416, p<.001). The results revealed that when semantic flexibility was available, speakers were indeed more error-prone. However, the patterns were not as clear in the ba and bei conditions. The prediction is that, the SA-ba condition would have a higher error rate than the DA-ba condition, and the SA-bei condition would also have a higher error rate than the DA-bei condition. Paired samples t-tests demonstrated that the SA-ba condition had a significantly higher error rate (5.3%) than the DA-ba condition (1.3%) by subjects (t1(29)=2.262, p<.05; t2(28)=1.722, p=.096). However, contrary to the prediction, the SA-bei condition had a significantly lower error rate (2.1%) than the DA- bei condition (10.7%) (t1(29)=-2.664, p<.05; t2(28)=-2.703, p<.05). Thematic Prominence. Second, to test the effects of thematic prominence, paired samples t-tests were conducted on the error rates between (i) the SA-ba condition and the SA-bei condition, and (ii) the DA-ba condition and the DA-bei condition. Based on the prediction that ba sentences would be easier to produce than bei sentences with agent- theme verbs, participants would be less error-prone in the ba conditions than in the bei conditions. The results were mixed: the DA-ba condition (M=0.013, SE=.009) indeed had a significantly lower error rate than the DA-bei condition (M=.106, SE=.026) (t1(29)= -3.294, p<.005; t2(28)=-3.520, p<.005). However, participants were less error-prone in 66 the SA-bei condition (M=0.021, SE=0.12) than in the SA-ba condition (M=0.053, SE=0.164), the difference being significant by subjects (t1(29)=2.249, p<.05; t2(28)=1.440, p=.161). Animacy. Finally, an effect of animacy was found. The difference in the error rate between the DA-ba condition and the DA-bei condition is greater than that between the SA-ba and SA-bei conditions (9.3% vs. 3.2%). The result suggests that effects of animacy and thematic prominence are independent of each other. At the first sight, the mixed patterns of results seem perplexing. However, if we take a closer look at what an error means in the SA-ba and SA-bei conditions, it is obvious that an error rarely occurs when both nouns are animate, which means participants’ performances in these two conditions can easily reach a ceiling. As mentioned above, production latency instead of error rate may be a better indicator for the difficulty level of these conditions. Now, let’s turn to the analysis of the production latency data. Mean Production Latency The mean production latency for correct responses as a function of animacy and preposition constraint is shown in Figure 2.5. The figure shows that the two factors ANIMACY and PREPOSITION CONSTRAINT were not interacting. In general, the SA 67 conditions had longer latencies than the DA conditions across the three PREPOSITION CONSTRAINT levels. Moreover, the patterns found in the SA conditions and the DA conditions looked similar. Across the board, sentences were produced fastest in the ba conditions and slowest in the bei conditions. ANOVAs confirmed the above observations. Reliable main effects were found for both ANIMACY (F1(1,29)=22.787, MSE=4.662, p<.01; F2(1,28)=8.382, MSE=3.340, p<.01) and PREPOSITION CONSTRAINT (F1(2,58)=9.736, MSE=2.476, p<.005; F2(2,56)=7.669, MSE=1.963, p<.01). The two factors did not interact (F1(2, 58)=2.083, MSE=.248, p=.134; F2(2,56)=1.355, MSE=.526, p=.265). 2000 2200 2400 2600 2800 3000 no preposition ba bei Preposition constraint Mean production latency (ms) Different animacy Same animacy Figure 2.5: Mean production latency as a function of animacy and preposition constraint in Experiment 1 68 Additionally, to test the hypotheses concerning effects of semantic flexibility, thematic prominence and animacy, independent and paired samples t-tests were performed on the no-prep. conditions and ba/bei conditions, respectively. Semantic Flexibility. First, to explore the effect of semantic flexibility, independent sample t-tests were conducted on the SA-no prep. and DA-no prep. conditions. 3 The prediction is that semantic flexibility hinders production with agent-theme verbs, so participants would spend longer production latencies producing both ba and bei sentences in the SA-no prep. condition than in the DA- no prep. condition. Participants’ correct responses were grouped by sentence type and then the mean production latencies of ba and bei sentences were compared respectively between these two conditions. The mean production latencies of each sentence type in the SA-no prep. condition vs. the DA-no prep. condition are shown in Figure 2.6. 3 Independent sample t-tests instead of paired t-tests were performed here due to unequal numbers of observations of ba and bei sentences in the SA-no prep. and DA-no prep. conditions. 69 2 4 1 8 . 6 2 4 1 8 . 6 2 4 1 8 . 6 2 4 1 8 . 6 2 8 4 7 . 4 2 8 4 7 . 4 2 8 4 7 . 4 2 8 4 7 . 4 2 8 8 1 . 6 2 8 8 1 . 6 2 8 8 1 . 6 2 8 8 1 . 6 2 4 9 3 . 6 2 4 9 3 . 6 2 4 9 3 . 6 2 4 9 3 . 6 2000 2200 2400 2600 2800 3000 ba bei sentence type production latency (ms) Different animacy Same animacy Figure 2.6: Mean production latency of ba and bei sentences in the SA-no prep.condition vs. the DA-no prep. condition in Experiment 1 The results revealed that for ba-sentences, the SA-no prep. condition had a significantly longer production latency (M=2.847, SE=.106) than the DA-no prep. condition (M=2.493, SE=.122, t(141)=2.11, p<.05). Additionally, for bei-sentences, the SA-no prep. condition had a marginally significantly longer production latency (M=2.881, SE=1.109) than the DA-no prep. condition (M=2.418, SE=.756, t(55)=1.87, p=.067). The results support the hypothesis that having semantic flexibility hinders production with agent-theme verbs. In addition, paired samples t-tests were conducted on (i) the SA-ba condition vs. the DA- ba condition, and (ii) the SA-bei condition vs. the DA-bei condition. Again, the prediction is that the SA-ba and SA-bei conditions would have longer production 70 latencies than the DA-ba and DA-bei conditions respectively. As shown in Figure 2.7, the SA-ba condition (M=2.753, SE=.112) indeed had a significantly longer production latency than the DA-ba condition (M=2.294, SE=.084) (t1(29)=4.935, p<.005; t2(28)= 3.585, p<.005). Likewise, the SA-bei condition (M=2.991, SE=.119) also had a significantly longer production latency than the DA-bei condition (M=2.786, SE=.085) by subjects (t1(29)=2.344, p<.05; t2(28)= 1.005, p=.324). The results again support our hypothesis. 2 2 9 4 . 5 2 2 9 4 . 5 2 2 9 4 . 5 2 2 9 4 . 5 2 7 8 6 . 8 2 7 8 6 . 8 2 7 8 6 . 8 2 7 8 6 . 8 2 7 5 3 . 8 2 7 5 3 . 8 2 7 5 3 . 8 2 7 5 3 . 8 2 9 9 1 . 2 2 9 9 1 . 2 2 9 9 1 . 2 2 9 9 1 . 2 2000 2200 2400 2600 2800 3000 ba bei sentence type production latency (ms) Different animacy Same animacy Figure 2.7: Mean production latency of the SA-ba vs. DA-ba conditions and the SA-bei vs. DA-bei conditions in Experiment 1 71 Thematic Prominence. Second, to test the effect of thematic prominence, planned comparisons were conducted on (i) the SA-ba vs. SA-bei conditions and (ii) the DA-ba vs. DA-bei conditions. As previously noted, production latency may serve as a better indicator than error rate because the SA-ba and SA-bei conditions may incur ceiling effects. Thematic prominence effects predict that participants would spend shorter production latency in the ba conditions than in the bei conditions. This prediction was confirmed: The SA-ba condition (M=2.7538, SE=.112) had a significantly shorter production latency than the SA-bei condition (M=2.9912, SE=.119) by subjects (t1(29)= -2.487, p<.05; t2(28)=-1.721, p=.096). Moreover, the DA-ba condition (M=2.2945, SE=.084) had a significantly shorter production latency than the DA-bei condition (M=2.7868, SE=.085) (t1(29)=-6.574, p<.001; t2(28)=-5.341, p<.001). Clearly, it is easier for speakers to place AGENT rather than THEME in the subject position. Animacy. Finally, an effect of animacy was found by production latency. The difference in the mean production latency between the DA-ba and the DA-bei conditions was greater than that between the SA-ba and SA-bei conditions (492 ms vs. 238 ms). This result corroborates the evidence found by error rate, indicating that as well as thematic prominence, animacy independently contributes to the saliency/accessibility of a noun. 72 2.3 Discussion The results of Experiment 1 are summarized as follows. First, the results showed that when semantic flexibility was available, participants were more prone to make errors and spent longer latencies on sentence production. The tendency was consistent whether syntactic flexibility was present or absent. In specific, the SA no-prep. condition ([+SemFlex],[+SynFlex]) had a higher error rate and a longer mean production latency than the DA no-prep. condition ([-SemFlex], [+SynFlex]). Furthermore, the SA-ba and SA-bei conditions ([+SemFlex], [-SynFlex]) also had longer production latencies than the DA-ba and DA-bei conditions ([-SemFlex], [-SynFlex]). The results provide evidence for the hypothesis that with agent-theme verbs, the two animate nouns would compete for the AGENT role and thus delays production. . Second, effects of thematic prominence and animacy were also confirmed. First of all, when participants had syntactic options (i.e. the no-prep. conditions), more ba sentences than bei sentences were produced, indicating that speakers tend to mention a more prominent thematic role earlier in a sentence. Moreover, despite the fact that error rate may not be valid due to the ceiling effects in the SA-ba and SA-bei conditions, such preference was confirmed by production latency: The SA-ba condition had a significantly longer production latency than the SA-bei condition. In addition, the DA-ba condition has a significantly higher error rate and a longer production latency than the DA-bei condition. 73 Third, the results demonstrated an effect of animacy. Both the differences of error rate and production latency between the DA-ba and DA-bei conditions were larger than those between the SA-ba and SA-bei conditions, indicating that it is harder to produce bei- sentences with an inanimate THEME subject. As a whole, thematic prominence and animacy effects are independent factors contributing to the conceptual accessibility of a noun. Experiment 1 successfully extended F. Ferreira’s (1994) findings to Mandarin Chinese online production. It is worth noting that across the board, each of the SA conditions (i.e., when both nouns are animate) had a significantly longer production latency (all ps<.05) than their DA counterparts (i.e. when the nouns differ in animacy), whether syntactic flexibility is available or not. Therefore, the source for production difficulty across the SA level is most likely to be semantic flexibility rather than syntactic flexibility. My hypothesis is based on the assumption that speakers are sensitive to the interaction of linguistic information such as thematic roles and animacy when they are doing role assignment in the conceptual level. More precisely, the interaction of thematic roles and noun animacy causes two animate nouns to compete for AGENT as there is a strong inclination to associate the AGENT role with animacy. Competition of this kind may not only delay sentence production, but result in an unstable state in the production system and thus increase the probability of making errors. 74 Chapter 3: Mandarin Sentence Production Task Using Theme-Experiencer Verbs 3.1 Motivation The results of Experiment 1 suggest that when speakers try to figuring out the relational meaning of two referents in the conceptual or message formulation level, having multiple propositions would delay this process due to competition from an interaction of thematic roles and animacy. I attribute this phenomenon to a strong association between AGENT role and animate nouns, as animacy is one of the prototypical features of AGENT roles. It is natural for one to suspect whether the effects of semantic flexibility are always detrimental to production, as the facilitative effects of syntactic flexibility appear consistent due to the incremental nature of grammatical encoding (F. Ferreira 1996). In order to further investigate this issue, a different verb type, namely, theme-experiencer verbs, potentially serves this purpose since they provide complementary results for Experiment 1. Crucially, a theme-experiencer verb assigns THEME and EXPERIENCER roles (e.g. surprise). Since THEME and EXPERIENCER are equally likely to be animate (see Belletti & Rizzi,1988; Grimshaw, 1990; Pesetsky, 1995 for discussion on Psych-verbs), the competition between two animates for AGENT may not take place with theme-experiencer verbs. Thus, the combined results of Experiments 1 and 2 provide us with the opportunity to explore whether verb thematic structure can condition effects of semantic flexibility, at least when operationalized with noun animacy. 75 Moreover, arguably, the results of Experiment 1 might not show true effects of animacy and thematic prominence, but are simply a consequence of the relative ease for producing ba sentences when compared with the production of bei sentences. As the experiments in F. Ferreira (1996) showed, passives generally took longer to formulate than actives probably due to syntactic complexity and occurrence frequency, which possibly make active sentences the ‘default’ structure in English. In this sense, the results of Experiment 1 may also reflect effects of syntactic complexity and/or occurrence frequency. This is because – mirroring the active/passive frequency asymmetry in English -- the BA construction occurs more frequently than the BEI construction in Mandarin Chinese 4 . Thus, in order to test whether the results of Experiment 1 demonstrate true effects of thematic prominence and animacy, I turned to test on theme-experiencer verbs in Experiment 2. In contrast to the predictions made in Experiment 1, effects of thematic prominence predict that the BEI construction would be the preferred sentence type with theme-experiencer verbs, because EXPERIENCER is mentioned earlier than THEME in bei sentences. Thus, the possible confounds of structural difference and occurrence frequency can be eliminated. 4 To calculate the occurrence frequency of ba and bei sentences, I consulted the Academia Sinica Balanced Corpus of Modern Chinese. I found that the BA construction occurred more often than the BEI construction (i.e., the long passives only) with 333 vs. 97 tokens in NEWSPAPER category, and 164 vs. 31 tokens in INTERVIEW category. 76 3.2 Experiment 2 3.2.1 Methodology Experiment 2 used the same sentence production task as Experiment 1 except that the verb shown on the screen was a theme-experiencer verb. Thirty new people from National Chi-nan University in Taiwan took part in this study. As the design of Experiment 1, Experiment 2 had a 2x3 factorial design. I manipulated two factors (i) the animacy of the nouns on the screen and the (ii) presence/absence of a preposition (i.e. ba or bei) on the screen. On one hand, semantic flexibility was only available in the SA (same animacy) conditions when both nouns were animate, but not in the DA (different animacy) conditions. On the other hand, syntactic flexibility was only possible in the no- prep. condition where no preposition was given, but not in the ba or bei condition. In a nutshell, this 2 x3 design allows us to investigate the effect of semantic flexibility in the presence/absence of syntactic flexibility. Table 2.2 summarizes the different conditions and is repeated here as Table 3.1 as a reminder. 77 Table 3.1: The experimental conditions of Experiment 2 Preposition constraint Animacy No preposition BA BEI Different animacy (DA) Words given judge, evidence, ignored judge, evidence, ignored, BA judge, evidence, ignored,BEI No. of syntactic choices at least 2 1 1 Preposition constraining? No Yes Yes Semantically flexible? No [-SemFlex] No [-SemFlex] No [-SemFlex] Syntactically flexible? Yes [+SynFlex] No [-SynFlex] No [-SynFlex] Same animacy (SA) Words given judge, lawyer,ignored judge, lawyer, ignored, BA judge, lawyer, ignored, BEI No. of syntactic choices at least 4 2 2 Preposition constaining? No Yes Yes Semantically flexible? Yes [+SemFlex] Yes [+SemFlex] Yes [+SemFlex] Syntactically flexible? Yes [+SynFlex] No [-SynFlex] No [-SynFlex] The materials were constructed in the same way as Experiment 1, except that in Experiment 2, 18 complex verb phrases were created instead of 30 because psychological verbs are less common than regular verbs. See Appendix II for a list of the target items used in Experiment 2. The resulting 108 frames were assigned to six lists, and each participant saw three different target items for each condition. A total of 90 trials across participants were collected for each condition. Norming Study for Targets. As in Experiment 1, in order to find out whether the paired animates in each target item potentially caused a bias toward a certain thematic relation, a post-hoc norming study was conducted in the form of a questionnaire. The target items of Experiment 2 were grouped into four lists based on a 2 x 2 design, contingent upon the 78 position of a noun (i.e. left or right) and sentence type (i.e. ba or bei sentences). Thirty- five filler items with varied degrees of plausibility and various sentences types were included in addition to the target items. Each list was assigned 10 participants, who were instructed to rate a sentence on a 4-point scale (1=implausible, 4=plausible). The study was conducted over the internet using Qualtrics software. To analyze participants’ responses, independent samples t-tests were conducted on each target item within sentence type to see if the rating of plausibility differed when the two nouns exchanged thematic roles. Of the 18 target items used in Experiment 2, when the two nouns switched positions, a significant difference was found on one item with ba (p<.05), and a marginally significant difference was found on another item with bei (p=.065). No significant differences were found on the remaining 16 targets. The results indicate that in most cases, it was equally likely for the paired animate nouns to take either thematic role. In other words, the influence caused by the bias toward one of the two propositions when semantic flexibility was available was quite limited in Experiment 2. Procedure. The procedure was the same as in Experiment 1. Participants initiated and ended a trial by pressing the space bar. They saw words on the screen and were told to make a sentence. The experiment began with instructions and 12 practice trials, followed by 68 test trials. The instructions used the same wording. To impose time pressure, a signal beeped for 350 ms after the words appeared on the screen for 5 seconds. The entire experiment took approximately twenty minutes. The recordings were then transcribed and 79 analyzed. As in Experiment 1, the dependent variables are mean production latency and error rate of each condition as indicators of production difficulty level. Notably, in conditions where two nouns are both animate and a preposition is present (i.e., the SA- ba/bei conditions), an error rarely occurs. Therefore, production latency may be a better indicator than error rate for these conditions. In addition, the proportions of ba and bei sentences are also studied as they reflect speakers’ preference for one construction over the other when faced with syntactic options. 3.2.2 Hypotheses and predictions The aim of Experiment 2 is two-fold. The first goal is to explore the effects of semantic flexibility with theme-experiencer verbs. Evaluating the results of Experiments 1 and 2 side by side allows us to get a complete picture of how effects of semantic flexibility are correlated with verb thematic structure. The second goal is to see whether we can extend the effects of thematic prominence and animacy found with agent-theme verbs to theme- experiencer verbs. Regarding the first goal, the hypothesis is that when the speaker is figuring out the relational meaning in the message component, s/he is sensitive to the thematic roles assigned by the verb. With a theme-experiencer verb, the referents involved in an event bear either a THEME or EXPERIENCER role. Just as in Experiment 1, semantic flexibility only exists when two nouns are both animate (e.g., “The director surprised the 80 actress” or “The actress surprised the director.”). Unlike in Experiment 1, since THEME and EXPERIENCER are equally likely to be animate, due to lack of competition, thematic role assignment may not be delayed. Moreover, if role-assignment also proceeds incrementally, then semantic flexibility may even facilitate sentence production because it allows multiple options for thematic-role mappings, in the same vein as the facilitative effects of syntactic flexibility on grammatical encoding (V. Ferreira, 1996). The interaction between thematic roles and animacy can be explained with a metaphor. Imagine that on a bus, two new passengers may bring up a fight if they both want to take a specific seat (i.e., AGENT in Experiment 1), but no fight will take place if two empty seats are available and these two passengers have no preferences (i.e., THEME and EXPERIENCER in Experiment 2). Importantly, the conditions where one noun is animate and the other is inanimate can serve as a baseline for comparison. In such cases, no competition will occur: The animate noun is assigned the EXPERIENCER role, and the inanimate noun is assigned the THEME role (e.g., “The director was surprised by the movie.”) In summary, for Experiment 2, I predict that semantic flexibility would not have a negative effect on production, and may even facilitate production. Consequently, equal or even shorter production latencies are expected in the conditions where both nouns are animate (i.e., the SA conditions) than in conditions where one noun is animate and the other is inanimate (i.e., the DA conditions). 81 Additionally, as for effects of thematic prominence, I predict that the BEI construction would be favored. In the BA construction, THEME takes the subject position, while in the BEI construction, EXPERIENCER takes the subject position. According to the hypothesized thematic hierarchy, EXPERIENCER is more prominent than THEME. As a result, in Experiment 2, if speakers tend to place a more prominent role in an earlier sentence position, they would produce more bei sentences than ba sentences when they have options between alternative structures (i.e. the no-prep. conditions). Moveover, when speakers are required to produce a specified structure, bei sentences should be easier to produce than ba sentences. Notably, the predicted pattern for Experiment 2 is the opposite of the observed pattern in Experiment 1. Thus, if the predicted pattern is verified, the production difficulty cannot be solely attributed to structural complexity or frequency effects. Moreover, as the results of Experiment 1 suggest, animacy effects exist independently. Therefore, I expect to see a larger difference between the ba and bei conditions in the DA condition than in the SA condition, because it would be more difficult to say a ba sentence with an inanimate THEME subject (e.g. “The movie BA the director surprise.”) To summarize, the predictions are summarized as follows. First, based on the hypothesis concerning effects of semantic flexibility, we expect to see: 82 (i) The SA-no prep. condition ([+SemFlex], [+SynFlex]) would have an equal or lower error rate and an equal or shorter mean production latency than the DA-no prep. condition ([-SemFlex], [+SynFlex]). (ii) The SA-ba/bei condition ([+SemFlex], [-SynFlex]) would have an equal or lower error rate and an equal or shorter mean production latency than the DA-ba/bei condition ([-SemFlex], [-SynFlex]). Second, based on the hypothesis concerning effects of thematic prominence and animacy, we expect to see: (iv) Due to thematic prominence effects, the bei conditions would have lower error rates and shorter production latencies than the ba conditions. (v) Due to animacy effects, the difference in production latency and error rate between the DA-ba and DA-bei conditions would be greater than those between the SA-ba and SA-bei conditions. (iii) In the SA-no prep. and DA-no prep. conditions, where syntactic options are available, the proportion of bei sentences would be higher than that of ba sentences. 83 3.2.3 Results 3.2.3.1 Data Preparation Exclusions and Replacements One item was excluded in the item analysis because of missing data due to programming error. In addition, the high error rate left two empty cells in the item analysis and one empty cell in the subject analysis. The means for these empty cells were estimated using the row and column means. 2.2% of the correct responses were excluded after trimming production latencies more than 2.5 standard deviations from the mean in each condition. Coding the Responses and Calculating Production Latencies I followed the methodology in Experiment 1 to code responses and calculate production latencies. Responses were categorized into correct response and error. Production latencies for correct responses were measured from the onset of the presentation of a trial to the onset of the first word of the correct sentence, including the time of hesitation, repetition, self-correction or stuttering. 3.2.3.2 Results As in Experiment 1, three sets of results are reported. The first set concerns the proportions of different sentence types in each condition. The second set concerns the 84 error rate for each condition. The third set concerns the mean production latency for correct responses in each condition. Proportions of Sentence Types Table 3.2 summarizes the proportions of different sentence types participants produced, including both correct responses and errors in each condition. As can be seen, the no-prep. conditions exhibited a smaller difference in the proportions of ba and bei sentences and a greater number of sentence types than the ba and bei conditions, verifying the assumption that a variety of syntactic options existed in the no-prep. conditions, while syntactic options were not available when ba or bei was present ([-SynFlex]). Table 3.2: Number and percentage of sentence types in each condition of Experiment 2 Preposition constraint Animacy No preposition BA BEI Different animacy (DA) BA-constructi on 13 (14.4) 87 (96.7) 0 Bei-construction 56 (62.2) 0 84 (93.3) Conjunction 4 (4.4) 0 0 Other preposition 2 (2.2) 0 0 SVO 6 (6.7) 0 1 (1.1) N/A 9 (10) 3 (3.3) 5 (5.6) Total 90 (100) 90 (100) 90 (100) Same animacy (SA) BA-constructi on 11 (12.2) 87 (96.7) 0 Bei-construction 45 (50) 0 90 (100) Conjunction 25 (27.8) 1 (1.1) 0 Other preposition 0 0 0 SVO 4 (4.4) 0 0 N/A 5 (5.6) 2 (2.2) 0 Total 90 (100) 90 (100) 90 (100) Note. Percentage in parentheses. Total possible number of observation in each cell is 90. N/A refers to incomplete utterances or no responses. 85 Moreover, the prediction that bei sentences would occur more frequently than ba sentences with theme-experiencer verbs was confirmed. In the DA-no prep. condition, bei sentences occurred more frequently (62.2%) than bei sentences (14.4%) (X 2 =26.797, p<.001). In the SA -no prep. condition, bei sentences also occurred more frequently (50%) than ba sentences (12.2 %) (X 2 =20.643, p<.001). The results indicated that effects of thematic prominence indeed played a role. In general, participants preferred to place EXPERIENCER instead of THEME in the subject position. This may explain why the proportions of ba and bei sentences were not close to even although syntactic options were available. Note that the percentage of using conjunction in the SA-no prep. condition was not as high here (27.8 %) as the percentage in Experiment 1 (34.7%) (e.g., “the director and the actress were surprised” vs. “the director and the actress were criticized”). This may be attributed to the fact that THEME and EXPERIENCER are equally likely to be animate, so speakers’ inclination toward placing both animate nouns in the subject position was not as strong as that in Experiment 1, where speakers tended to assign both animate nouns the AGENT role and place both nouns in the subject position. Importantly, the dominant proportion of bei-sentences in Experiment 2 demonstrated that the dominant proportion of ba-sentences in Experiment 1 does not just follow from frequency effects even if the BA construction occurs more frequently than the BEI construction as a whole. Instead, the result suggests that during sentence production, 86 speakers’ syntactic choice is to a large extent sensitive to the prominence of thematic roles. Error rate The error rate for each condition is shown in Figure 3.1. The figure shows that ANIMACY and PREPOSITION CONSTRAINT did not interact with each other (F1(2,58)=1.883, MSE=.065, p=.174; F2(2,32)=1.237, MSE=.024, p=.304). ANIMACY did not have a main effect (F1(1,29)=3.080, MSE=.089, p=.09; F2(1, 16)=1.501, MSE=.047, p=.238). Regardless of ANIMACY, the patterns in the SA conditions and the DA conditions are similar: the error rates were highest when no preposition was given. Contrastively, PREPOSITION CONSTRAINT had a main effect (F1(2,58)=22.176, MSe=.615, p<.001; F2(2,32)=6.204, MSe=.397, p<.05), indicating that participants are sensitive to the absence vs. presence of a preposition. Specifically, in both the DA and SA conditions, participants were more error-prone in the no-prep. conditions (20% and 21%) than in the ba condition (12.2% and 2%) or the bei condition (5.6% and 1.1%). 87 0 0.1 0.2 0.3 no preposition ba bei Preposition constraint Production error rate Different animacy Same animacy Figure 3.1: Mean error rate as a function of animacy and preposition constraint in Experiment 2 Semantic Flexibility. Additionally, paired samples t-tests were conducted on the SA and DA conditions to study the effects of semantic flexibility. The prediction is that semantic flexibility would not have a negative effect on production with theme-experiencer verbs. Indeed, no significant difference was found between the SA-no prep. condition (21%) and DA-no prep. condition (20%) (t1(29)=.183, p=.856; t2(16)=.174, p=.864). However, when a preposition was present, the error rates in the SA conditions were significantly lower than those in the DA conditions: the SA-ba condition had a significantly lower error rate than the DA-ba condition by subjects (t1(29)=-3.071, p<.05; t2(16)=-1.817, p=.088); the SA-bei condition also had a significantly lower error rate than the DA-bei 88 condition by subjects (t1(29)=-2.112, p<.05; t2(16)=-1.461, p=.163). The results are not surprising because just as in Experiment 1, the SA-ba and SA-bei conditions may have exhibited ceiling effects as errors rarely occur in these two conditions. Nevertheless, the prediction is verified by the no-prep. conditions such that having semantic flexibility does not result in any production difficulty with theme-experiencer verbs, in contrast to the findings in Experiment 1 with agent-theme verbs Thematic Prominence. Second, to test the effect of thematic prominence, paired samples t-tests were conducted on the SA-ba vs. SA-bei conditions and the DA-ba vs. DA-bei conditions. Based on the prediction that the bei conditions would be easier than the ba conditions with theme-experiencer verbs, participants should be less error-prone in the bei conditions than in the ba conditions regardless of animacy. Although the error rate for the SA-bei condition (M=.011, SE=.033) was lower than that for the SA-ba condition (M=.022, SE=.011), the difference was not significant (t1(29)=.571, p=.573; t2(16)=.566, p=.579). As mentioned before, the insignificance is very likely to have resulted from the ceiling effects existing in these two conditions. In contrast, the error rate was higher in the DA-ba condition (M=.122, SE=.033) than in the DA-bei condition (M=.056, SE=.023) with a marginal significance by subjects (t1(29)=1.989, p=.056; t2(16)=1.000, p=.579). It seems safe to say the effect of thematic prominence was partially verified in the DA conditions. 89 Animacy. Finally, an animacy effect was found. The difference in error rate between the DA-ba condition and the DA-bei condition was larger than that between the SA-ba and SA-bei conditions (6.6 % vs. 1.1%), confirming the prediction that animacy effects and thematic prominence effects are independent of each other. Mean Production Latency The mean production latency for correct responses as a function of animacy and preposition constraint is shown in Figure 3.2. The figure suggests that the two factors ANIMACY and PREPOSITION CONSTRAINT did not interact with each other (F1(2,58)=1.043, MSE=.382, p=.355; F2(2,32)=.970, MSE=.206, p=.384). In general, the DA conditions had longer production latencies than the SA conditions, indicating that with theme-experiencer verbs, it is easier to construct a sentence when both nouns were animate. However, the main effect of ANIMACY is only marginally reliable by subjects (F1(1,29)=3.870, MSE=1.975, p=.059; F2(1,16)=1.952, MSE=.896, p=.181). Somewhat surprisingly, the patterns in the SA and DA conditions go in opposite directions. In the SA conditions, the mean production latency was shortest in the no-prep. condition (M=2.545) and longest in the SA-bei condition (M=2.755). In contrast, in the DA conditions, the mean production latency was shortest in the DA-bei condition (M=2.794) and longest in the no prep. condition (M=2.882). As a results, there was no main effect of PREPOSITION CONSTRAINT (F1(2,58)=.113, Mse=.068, p=.863; F2(2,32)=1.141, Mse=.281, p=.332). 90 2400 2600 2800 3000 no preposition ba bei Preposition constraint Mean production latency (ms) Different animacy Same animacy Figure 3.2: Mean production latency as a function of animacy and preposition constraint in Experiment 2 To further examine the interaction between ANIMACY and PREPSITION CONSTRAINT, planned comparisons were conducted on different conditions. Semantic Flexibility. First, to explore the effect of semantic flexibility, I performed independent samples t-tests to look into the production latency difference between the SA-no prep. condition and the DA-no prep. condition. The correct responses were grouped by sentence type (i.e. ba or bei sentences), and then the mean production 91 latencies depending on sentence type were compared between these two conditions. Thus, the number of observations for each sentence type may not be equal. The prediction is that semantic flexibility has no negative effects or even facilitates production with theme-experiencer verbs, so the SA-no prep. condition would have equal or even shorter production latency than the DA- no prep. condition. Figure 3.3 shows that for ba-sentences, the DA-no prep. condition had a numerically longer production latency (M=3.157, SE=.353) than the SA-no prep. condition (M=2.549, SE=.396), but the difference was not statistically significant (t(22)=1.144, p=.265). Similarly, for bei- sentences, the DA-no prep. condition also had a numerically longer production latency (M=2.718, SE=.166) than the SA-no prep. condition (M=2.701, SE=.181), but the difference did not reach statistic significance, either (t(93)=.068, p=.946). 92 3 1 5 7 . 8 3 1 5 7 . 8 3 1 5 7 . 8 3 1 5 7 . 8 2 7 1 8 . 5 2 7 1 8 . 5 2 7 1 8 . 5 2 7 1 8 . 5 2 5 4 9 . 8 2 5 4 9 . 8 2 5 4 9 . 8 2 5 4 9 . 8 2 7 0 1 . 8 2 7 0 1 . 8 2 7 0 1 . 8 2 7 0 1 . 8 2000 2200 2400 2600 2800 3000 3200 ba bei sentence type production latency (ms) Different animacy Same animacy Figure 3.3: Mean production latency of ba and bei sentences in the SA-no prep. condition vs. the DA-no prep. condition in Experiment 2 The results support the prediction that having semantic flexibility does not have a negative effect on production with theme-experiencer verbs, as participants did not behave differently whether semantic flexibility was available or not. Importantly, the pattern contrasts with the pattern found with agent-theme verbs, where production was slowed down when semantic flexibility was present. 93 In addition, paired samples t-tests were conducted on (i) the SA-ba condition vs. the DA- ba condition, and (ii) the SA-bei condition vs. the DA-bei condition. Again, the prediction is that the SA-ba and SA-bei conditions would have equal or even shorter production latencies than the DA- ba and DA-bei conditions respectively. 2 8 6 1 . 9 2 8 6 1 . 9 2 8 6 1 . 9 2 8 6 1 . 9 2 7 9 4 . 3 2 7 9 4 . 3 2 7 9 4 . 3 2 7 9 4 . 3 2 6 0 9 2 6 0 9 2 6 0 9 2 6 0 9 2 7 5 5 . 4 2 7 5 5 . 4 2 7 5 5 . 4 2 7 5 5 . 4 2000 2200 2400 2600 2800 3000 ba bei sentence type production latency (ms) Different animacy Same animacy Figure 3.4: Mean production latency in the SA-ba vs. DA-ba conditions and the SA-bei vs. DA-bei conditions in Experiment 2 As shown in Figure 3.4, the DA-ba condition had a numerically longer production latency (M=2.861, SE=.144) than the SA-ba condition (M=2.609, SE=.126), but the difference was not statistically significant (t1(29)=1.636, p=.113; t2(16)=1.546, p=.141). Likewise, the DA-bei condition had a numerically longer production latency (M=2.794, 94 SE=.133) than the SA-bei condition (M=2.755, SE=.147), but the difference was not significant either (t1(29)=.226, p=.823; t2(16)=.166, p=.870). The results again confirm the prediction that semantic flexibility has no negative effects on production with theme- experiencer verbs. Thematic Prominence. Second, to test the effect of thematic prominence, planned comparisons were conducted on (i) the SA-ba vs. SA-bei conditions and (ii) the DA-ba vs. DA-bei conditions. The prediction is that the SA-bei and DA-bei conditions would have shorter production latencies than SA-ba and DA-ba conditions, respectively, because the BEI construction is preferred with theme-experiencer verbs (i.e. EXPERIENCER first). Paired samples t-tests demonstrated that the DA-bei condition had a numerically shorter production latency (M=2.794) than the DA-ba condition (M=2.861), but the difference was not significant (t1(29)=-.421, p=.677; t2(16)=-1.670, p=.114). Somewhat surprisingly, the SA-bei condition had a numerically longer production latency (M=2.755) than the SA-ba condition (M=2.609), but the difference was not significant, either (t1(29)=1.125, p=.270; t2(16)=.236, p=.816). As a whole, the results failed to show effects of thematic prominence because there were no differences in the production latencies of ba and bei sentences. 95 Animacy. Finally, an effect of animacy was not found due to the unexpected longer production latency in the SA-bei condition, where the subject was an animate EXPERIENCEER. Recall that in Experiment 1, the SA-bei condition, where the subject is an animate THEME, also resulted in the longest production latency among the three SA conditions, which is expected because the SA-bei condition in Experiment 1 suffers a double-penalty from effects of thematic prominence and semantic flexibility. However, neither thematic prominence nor semantic flexibility can account for the seeming difficulty in the SA-bei condition with theme-experiencer verbs. This issue will be discussed further in 3.3 3.3 Discussion The results of Experiment 2 are summarized as follows. First, the results suggest that when semantic flexibility was available, participants were not more error prone or slower in producing sentences. The tendency was consistent whether syntactic flexibility was present or not. Specifically, the SA no-prep. condition ([+SemFlex],[+SynFlex]) and the DA no-prep. condition ([-SemFlex], [+SynFlex]) did not differ significantly in error rate or mean production latency. Moreover, the SA-ba and SA-bei conditions ([+SemFlex],[- SynFlex]) were not significantly different from the DA-ba and DA-bei conditions ([- SemFlex],[-SynFlex]) in mean production latency. In conclusion, in Experiment 2, semantic flexibility did not have a negative effect on sentence production. This finding supports the hypothesis that with theme-experiencer verbs, no competition exists between 96 two animate nouns for a single thematic role. Moreover, though without statistical significance, the numerical tendency seems to suggest that with theme-experiencer verbs, semantic flexibility tends to facilitate sentence production due to the availability of multiple assignment options. Crucially, the results from Experiment 1 and 2 together provide evidence for the hypothesis that effects of semantic flexibility, when operationalized by noun animacy, can be conditioned by verb thematic structure, indicating that message formulation is sensitive to linguistic information. Second, effects of thematic prominence were found when speakers had freedom in deciding between the BA and BEI constructions. In both the SA-no prep. the DA-no prep. conditions, bei-sentences occur more often than ba-sentences with theme-experiencer verbs, supporting the hypothesis that speakers tend to place a more prominent thematic role in subject position (i.e., EXPERIENCER). However, effects of thematic prominence and animacy were not found when speakers had to produce a specified construction. Significant differences in production latency did not exist between the SA-ba and SA-bei conditions or between the DA-ba and DA-bei condition. The lack of effects of thematic prominence or animacy in Experiment 2 may be attributed to the following accounts. The first possible account is related to the actual thematic roles speakers had in mind when they figured out the relational meaning on a trial. When the THEME is animate, it can either be volitional or nonvolitional. A close examination of the verbs used in Experiment 2 shows that although all verbs are possibly to co-occur with an adverb such 97 as ‘intentionally’, the degrees of acceptability may vary depending on whether the THEME have control over or directly affects on the EXPERIENCER’s resultative state. Such semantic differences are subtle and can vary from person to person. According to my intuition, 4 of the 18 verb phrases are more problematic to co-occur with guyi ‘intentionally’ (e.g., frighten, move, amuse, elate), whereas the remaining 14 verb phrases are compatible with this adverb. It is worth noting that although the BA and BEI constructions are variations of the canonical SVO order, the two constructions differ in some important ways (cf. Huang, Li & Li, 2009 for a review). A point relevant for the present study is the requirement on the NP following ba being an affectee 5 . That is, only those that have been afflicted upon, or “dealt with” are acceptable as post-ba NPs. Zhang (2001) notes that the two main differences between ba and bei constructions are (i) direct or indirect affectedness and (ii) direct or indirect causation. With respect to the first difference: ba sentences require the post-ba NP to be directly affected by an action. In contrast, bei sentences may just express an indirect effect of an action. With respect to the second difference: the subject in ba sentences can be direct or indirect causer. In contrast, the object in bei sentence must be a direct causer. 5 As a result, the ba construction has been termed as the “disposal construction” (L. Wang 1954, Chao 1968, Li and Thompson 1981), “the executive construction” (Hashimoto,1971), the“accusative construction’ (Teng 1975), and “a highly transitive construction,” (Wang, 1987). 98 Keeping these differences in mind, let’s examine the SA-ba and SA-bei condition in Experiment 2 more closely. When a ba-sentence is intended for production (as hinted by the presence of ba), the semantic restriction on the BA construction requires the post-ba NP (EXPERIENCER) to be directly affected by an action. This requirement may lead speakers to interpret EXPERIENCER as PATIENT, which is conceptually closer to a role being directly affected. Meanwhile, the other NP (THEME) acts more like AGENT who intentionally imposes the action on the affectee. In contrast, when a bei-sentence is intended for production, the semantic restriction requires the post-bei NP (THEME) to be a direct causer, which is conceptually closer to AGENT, Meanwhile, the subject NP (EXPERIENCER) acts more like PATIENT. Crucially, thematic role assignment presumably takes place in the message component, which is prior to grammatical encoding. Thus, when grammatical encoding begins, according to the effect of thematic prominence, speakers would place AGENT before PATIENT. This means that SA-ba condition would be easier than SA-bei condition if the thematic roles involved are AGENT and PARTICIPANT. Indeed, the experimental results show that SA-ba condition had numerically shorter production latencies than the SA-bei condition. However, the account of thematic-role shift triggered by structural clues cannot explain why the DA-ba and DA-bei conditions did not differ in production latencies, either, as inanimate nouns are unlikely to have volition and thus would not undergo a THEME AGENT shift. I propose the lack of effects of thematic prominenc is due to a speed- accuracy tradeoff. Note that the error rates between DA-ba and DA-bei conditions 99 indeed showed effects of thematic prominence. It is possible that speaker were aiming at speed in these conditions and thus sacrificed accuracy, resulting in significant differences in error rates but not in production latencies. An obvious question to the role-shift account is: Suppose participants were interpreting theme-experiencer verbs rather like agent-theme verbs, how do we explain the fact that there were more bei constructions than ba constructions with theme-experiencer verbs in the SA no-prep. condition? A possible explanation concerns when the syntactic information becomes available during sentence production. As previously argued, the shifts in thematic roles (THEMEAGENT, EXPERIENCERPATIENT) are possibly triggered by the semantic properties pertaining to the BA and BEI constructions. In the ba and bei conditions, speakers identify the syntactic structure to be used as soon as the preposition appears on the screen. In other words, syntactic information is accessible to the speakers at the outset of sentence production. What follows naturally is the process of message formulation will be influenced by the semantic properties pertaining to that specific syntactic construction. In contrast, in the no-prep. conditions, speakers do not yet have access to the syntactic structure to be built in the conceptual level, so they simply rely on the verb thematic structure to assign thematic roles (i.e., EXPERIENCER and THEME). That is, speakers may not be aware of what sentence they will eventually utter until the subject of a sentence is produced. This lexically-driven account is shared by a number of researchers 100 (e.g., Bates & MacWhinney, 1982; Bock, 1982; De Smedt 1996; Garnsey, Pearlmutter, Myers & Lotocky, 1997; Kempen and Hoenkamp 1987; Levelt 1989). Since EXPERIENCER tends to grab the subject position due to its prominence, the proportion of bei sentences is higher. In other words, in the no-prep. condition, syntax is very likely to be lexically driven. To conclude, the timing of when syntactic information becomes available is very likely to have an impact on both the message and grammatical components. In other words, the processes taking place in the message component and the grammatical component may interact with each other, instead of proceeding in a staged manner where no backward feedback is available. The second possible account for the lack of a significant thematic prominence effects between the SA-ba and SA-bei conditions in Experiment 2 might be due to the variable distance between different thematic roles in the continuum of conceptual accessibility. Conceptual accessibility is defined as “the ease with which the mental representation of some potential referent can be activated in or retrieved from memory” (Bock & Warren, 1985:50). Bock & Warren (1985) proposed that there is a continuum of conceptual accessibility related to grammatical relations (i.e. subjects and direct/indirect objects), such that higher grammatical roles are typically occupied by noun phrases representing more accessible concepts. They also proposed that a continuum of conceptual accessibility underlies the hierarchy of thematic roles, such that a more prominent 101 thematic role is also more accessible. Taken together, a higher grammatical role is more likely to be occupied by a more prominent thematic role (i.e., thematic prominence effects). Compared with the results in Experiment 1, the smaller differences found in error rate and production latency between the ba and bei conditions in Experiment 2 may imply that in the conceptual accessibiity continuum, THEME and EXPERIENCER are closer to each other compared with AGENT and THEME. 3.4 Error Analysis of Experiments 1 and 2 In addition to analyzing quantitative data such as production latency and error rate, a qualitative analysis of errors can also provide useful information for identifying the source of difficulty during online production. In this section, I report the occurrence frequency of different error types in each condition, and discuss the possible source for these errors. A comparison is made between Experiments 1 and 2, along with theoretical accounts for how the observed patterns relate to the hypotheses of interest. It is worth keeping in mind that these comments are somewhat speculative due to a small number of observations. As a reminder, an error is assigned to one of the following types depending on its content: element missing, no response, role reversal, ungrammatical-SVO, and ungrammatical-conjunction. Table 3.3 shows the count of each error type in each condition in Experiment 1 (i.e. with agent-theme verbs). 102 Table 3.3: Count of each error type in each condition of Experiment 1 DA SA Error type no-preposition ba bei no-preposition ba bei Element missing 8 0 3 7 6 2 No response 1 0 0 1 1 1 Role reversal 4 1 9 0 0 0 ungrammatical-SVO 14 1 1 14 0 0 ungrammatical-Conjunction 1 0 0 42 0 0 Total 28 2 13 64 7 3 First, the type of element missing (e.g., “The magician …the love affair…I don’t know…”) was more frequent in the no-prep. conditions. In these conditions, participants needed to figure out an event based on two nouns and one verb in the message component and then retrieved a syntactic structure in the grammatical component. To deal with these tasks under time pressure, the probability of error was higher because the production mechanism might be in an unstable state. For instance, on some trials, the speaker was confused and could not decide on the proposition they wanted to covey. On some trials, one noun was missing in the utterance. I suggest that errors of this type originated from difficulty in conceptualization, as they are more akin to a thinking problem, not a talking problem. Second, the type of role reversal (e.g., # “The photographer was grabbed tightly by the camera”) was most common in the DA-bei condition. In this type of error, noun phrases surface in the wrong syntactic position. A correct response in the DA-bei condition requires an inanimate THEME subject (e.g., “The camera BEI the photographer grabbed tightly.”) Due to the effects of animacy and thematic prominence, speakers tended to 103 place an animate AGENT in the subject position. This type of error provides evidence for the claim that sentence production is a highly incremental process based on a “first- retrieved, first-mentioned” principle. Interestingly, it is observed that speakers were usually able to detect an error of grammatical role reversal during production, as indicated by hesitation or pause, or upon the sentence was completed, as indicated by self-correction. This provides evidence for the monitoring component in the perception system proposed by Levelt (1983, 1989). Unlike the element-missing type, the role reversal type may be more akin to a talking problem following from a mismatch between message and structure instead of a thinking problem. Third, the type of ungrammatical-SVO (e.g., “*The employer examined twice the secretary”) was equally common in the DA no-prep. and SA no-prep. conditions. Due to the postverbal constraint, a sentence containing a verb followed by two complements sounds ungrammatical or at best marginal in Mandarin. However, since SVO is the canonical word order, it could serve as the default structure when the participants fail to retrieve a more appropriate structure extemporaneously. Fourth, the type of ungrammatical-Conjunction (e.g., “*The employer and the secretary examined twice”) was most common in the SA-no prep. condition. When both nouns were animate, speakers tended to place both animate nouns in the subject position, resulting in faulty thematic role assignment. Three explanations may account for this phenomenon. First, this could be due to an effect of syntactic persistence because 10 104 fillers were conjoined-NP sentences. Second, if it is the case that the two animate nouns are retrieved equally fast, the “first-retrieved, first-mentioned” principle of the incremental theory predicts the appearance of conjoined-NP sentences. Third, with normal verbs, animate nouns prefer to be AGENT rather than THEME. Such a preference might have also contributed to the high proportion of conjoined-NP sentences in the SA- no prep. condition. Notably, in this condition with theme-experiencer verbs (see Table 3.4 below), the occurrence of conjoined-NP sentences was much fewer. This may result from the fact that theme-experiencer verbs but not agent-theme verbs can license one-place predicate. For example, “The employer and the secretary examined twice” is ungrammatical, but “The employer and the secretary were scared” is well-formed 6 . However, the overall frequency of conjoined-NP sentences including both correct responses and errors in the SA no-prep. condition was also higher with agent-theme verbs (34.7%) than with theme-experiencer verbs (27.8%). Thus, a possible explanation is that THEME or EXPERIENCER are equally likely to be animate, so the tendency to conjoin the two animate nouns is not as strong as it is with agent-theme verbs. Table 3.4 shows the count of each error type in each condition in Experiment 2 (i.e.with theme- experiencer verbs.) 6 Example (1) is an ungrammatical one-place predicate sentence with an agent- theme verb in Chinese. In contrast, example (2) is a grammatical one-place predicate sentence with a theme-experiencer verb. (1) *yuangong han mishu jiancha-le liang-bian. employer and secretary examine twice “*The employer and the secretary examined twice.” (2) yuangong han mishu xia-le yi-tiao. employer and secretary scare quite a bit “The employer and the secretary were scared quite a bit.” 105 Table 3.4: Count of each error type in each condition of Experiment 2 DA SA Error type no-preposition ba bei no-preposition ba bei Element missing 8 3 1 5 2 1 No response 1 0 0 0 0 0 Role reversal 0 8 3 0 0 0 ungrammatical-SVO 4 0 1 5 0 0 ungrammatical-Conjunction 4 0 0 9 0 0 Total 17 11 5 19 2 1 The distribution of error type in Experiment 2 is similar to the distribution in Experiment 1, except that the type of role reversal occurred mostly in the DA-ba condition (e.g., # “The architect BA the developmental project exhausted”). A correct response in the DA-ba condition with theme-experiencer verbs requires an inanimate THEME subject (e.g., “The project BA the architect exhausted”), while the effects of animacy and thematic prominence have led speakers to place an animate EXPERIENCER in the subject position. This again provides evidence for the incremental nature of grammatical encoding. Comparing Table 3.3 and Table 3.4, the patterns of role-reversal between the DA-ba and DA-bei conditions are contrastive depending on verb type, as demonstrated in Figure 3.5. 106 0 1 2 3 4 5 6 7 8 9 10 DA-ba DA-bei condition count of role reversal Agent-theme verbs Theme-exp. Verbs Figure 3.5: Count of role reversal in DA-ba vs. DA-bei conditions with two verb types In summary, the error analysis corroborates the findings from error rate and production latency in two ways. First, the type of ungrammatical-Conjunction in the SA no-prep. condition occurred more frequently with agent-theme verbs than with theme-experiencer verbs, suggesting that the two animate nouns tend to compete for being AGENT, but such competition does not exist as often when the thematic roles to be assigned are THEME and EXPERIENCER. This asymmetry provides evidence for the hypothesis that effects of semantic flexibility can be conditioned by verb type. Second, the type of role reversal occurred more frequently in the DA-bei condition with agent-theme verbs and in the DA- ba condition with theme-experiencer verbs, providing evidence for effects of thematic 107 prominence and animacy such that speakers tend to place a more prominent role in an earlier position. 3.5 General Discussion of Experiments 1 and 2 The Production of BA and BEI Constructions. Experiments 1 and 2 explored the process of message formulation by investigating situations where the speaker has a choice about which proposition to convey. I took as the starting point existing work on syntactic flexibility, which has contrasted a competitive model with an incremental model (V. Ferreira 1996). Existing work on the syntactic level found that syntactic flexibility facilitates production, thus supporting the incremental model. I investigated the effects of semantic flexibility and whether they are sensitive to different thematic role configurations by looking at agent-theme verbs (Experiment 1) and theme-experiencer verbs (Experiment 2). I found that with agent-theme verbs, participants spent longer time producing a sentence when the nouns were both animate (i.e. semantic flexibility is present), as compared to an animate-inanimate noun pair (i.e. without semantic flexibility). However, with theme- experiencer verbs, I did not observe any slowdowns when semantic flexibility was present. Based on this asymmetry, I conclude that message formulation can be carried out competitively or incrementally, depending on linguistic information. In other words, message formulation may behave differently from grammatical encoding, which is 108 regarded as incremental by many researchers. Although speech proceeds in a word-by- word incremental fashion, there is certain amount of parallelism in thinking and in messages. The observation that semantic flexibility can cause competition does not deny that many syntactic structure decisions or NP placements are decided incrementally, but suggests that competition indeed occurs in some proportion of ‘tough decision’ cases (e.g., Stalling et al., 1998). Additionally, the results of Experiment 1 and 2 extended the findings in F. Ferreira (1994) on effects of thematic prominence and animacy to Mandarin Chinese sentence production. In general, speakers preferred to place a more prominent thematic role in an earlier sentence position, and the effect was larger when both nouns differ in animacy. The findings provide cross-linguistic evidence for the incrementality-based account for grammatical encoding, which posits that a more accessible noun is dispatched sooner, i.e. the “first-retrieved, first mentioned” principle. Moreover, as the first attempt to study Mandarin BA and BEI constructions side by side from a production perspective, the results of Experiments 1 and 2 demonstrated that participants produced more ba sentences than bei sentences with agent-theme verbs, and the pattern was opposite with theme-experiencer verbs. Additionally, both error rate and production latency data revealed that ba sentences are easier to produce with agent-theme verbs, while bei sentences are easier to produce with theme-experiencer verbs. Taking these results together, the contrast cannot have resulted from structural difference or 109 frequency effects. Moreover, these effects were larger when nouns differed in animacy. As a whole, our study suggests that to express the same proposition, Mandarin speakers’ choice between the BA and BEI constructions is to a great extent conditioned by the interaction of thematic roles and animacy of nouns. The present work makes contributions to a better understanding of the nature of Mandarin BA and BEI constructions. Adversity in the BEI Construction. A relevant issue related to the semantic property of the BEI construction is the fact that a passive sentence in Mandarin is often accompanied by an adversary meaning. Chinese linguists usually suggest that the subject of a passive sentence can bear the theta role of a neutral EXPERIENCER, or an indirect AFFECTEE. These theta roles refer to an entity that can be altered physically or psychologically by the event in question. Researchers suggest that adversity adds naturalness to the BEI construction. For one special type of passives, termed the adversative passive, adversity is even considered to be a requirement (see discussion in Huang, Li and Li, 2009). In contrast, the BA construction does not feature such a semantic property. Therefore, when a verb denotes an adversary or negative meaning, it is possible that a bei-sentence would be favored. In order to find out whether adversity facilitated the production of the BEI construction, I grouped the verb phrases into negative verbs (e.g., anger) and positive verbs (e.g., praise), and compared the production latencies of bei-sentences between these two groups of verbs. Verbs whose meaning could be not classified as clearly negative positive meaning (e.g., push) was categorized as a neutral verb and excluded 110 from the analysis. If adversity facilitates the naturalness of bei-sentences to some extent, we predict that the production of bei-sentences should be easier with negative verbs than with positive verbs in the bei condition. In addition, speakers would prefer to use bei- sentences more often than ba-sentences with negative verbs when they can choose between alternative structures (i.e. the no-preposition condition). In Experiment 1 which involves agent-theme verbs, 12 verbs were categorized as positive and 11 as negative (7 neutral). Analyses of independent samples t-tests show that adversity did not affect the production latencies of bei-sentences, but to some extent increased the occurrences of bei-sentences. First, in the bei condition where speakers were required to produce bei-sentences, there was no significant difference between the production latencies associated with negative verbs (M=2.999, SE=.135) and positive verbs (M=2.911, SE=.142) (t(106)=.148, p=.655). Second, in the no-preposition condition where speakers freely decided which construction to use, there was a tendency for them to use the BEI construction more often with negative verbs (ba:13, bei: 11) if compared with positive verbs (ba:18, bei: 9). However, ba-sentences still outnumbered bei-sentences regardless of adversity, as predicted by the influence of thematic prominence effects (i.e. AGENT first). Moreover, in the no-prep condition, adversity did not result in any significant difference whether the bei sentences were associated with negative verbs (M= 3.215, SE=.29) or positive verbs (M=3.052, SE=.365) (t(18)=.353, p=.728). 111 In Experiment 2 which involves theme-experiencer verbs, 4 verbs were categorized as positive and 14 as negative. Somewhat different from Experiment 1, independent samples t-tests show that adversity could affect both the production latencies and occurrences of bei-sentences. First, in the bei condition, there was no significant difference between the production latencies associated with negative verbs (M=2.665, SE=.128) and positive verbs (M=2.656, SE=.198) (t(80)=.034, p=.973). This is probably because in general, the bei-condition in Experiment 2 was relatively easy for participants (i.e., EXPERIENCER first). However, in the no-preposition condition, as predicted by the thematic prominence effects (i.e. EXPERIENCER first), speakers used the BEI construction more often than the BA construction both with negative verbs (ba:7, bei: 34) and positive verbs (ba:5, bei: 11). Crucially, this tendency appeared to be stronger when adversary meaning was present. Moreover, when it comes to production latency, adversity resulted in a marginal difference between the bei-sentences associated with negative verbs (M=2.440, SE=.150) and positive verbs (M=2.656, SE=.198) (t(32)=-1.921, p=.064), indicating that when the EXPERIENCER or AFFECTEE undergoes change of psychological states, adversity to some extent adds naturalness to that event, and thus facilitates sentence production, in particular at the message formulation level. In conclusion, the results are compatible with the common observation that the Mandarin Chinese BEI construction is closely related to adversity, especially when the event involves change of psychological states (i.e., theme-experiencer verbs). The semantic factor of adversity, along with the thematic prominence effects, may have contributed to 112 the findings of Experiment 2, which suggest that the BEI construction is favored by theme-experiencer verbs. In contrast, the results from Experiment 1 suggest that the adversity factor may not be very prominent when the event involves an action or change of physical states. Syntactic Flexibility vs. Semantic Flexibility. A final remark concerns the distinction between syntactic flexibility and semantic flexibility. The results of my experimental investigations suggest that effects of semantic flexibility (i.e. having the choice between multiple propositions) and syntactic flexibility (i.e. having the choice between different syntactic structures that encode the same propositional meaning) are not uniform. A natural question is: Why do they differ? I propose that such a difference originates from the distinctive nature of the message component and the grammatical component, where the two kinds of flexibility are related respectively. Syntactic flexibility is assumed to facilitate production under the assumption that grammatical encoding is processed in a piecemeal fashion. The proposal that the construction of syntactic structure is at least partially lexically driven such that words can dictate the eventual syntactic structure is crucial to many sentence production models (e.g. Bock, 1982; Dell, 1986, 1988; Levelt, 1989; F. Ferreira, 2000). For the past decades, the line of research on language production has for the most part evolved around the grammatical component and phonological component because these two levels involve linguistic structures. It is generally agreed that grammatical encoding is incremental 113 based on constraints on processing capacity, though the degree of incrementality is found to vary in different studies As a consequence, the effect of syntactic flexibility should be a uniform phenomenon since it involves a component that operates more or less incrementally. In contrast, the nature of the processes that involve conceptual structures is less tackled. There is no presumption that message formulation needs to be carried out in an incremental way. As the view articulated by Levelt (1989), unlike the autonomous feature of grammatical encoding, message formulation requires planning and therefore can be highly resource-demanding. Consequently, both linguistic and nonlinguistic knowledge can determine the process of message formulation. One may also ask: do the results in Experiments 1 and 2 also show a facilitative effect of syntactic flexibility, given that grammatical encoding is unconditionally incremental? Theoretically speaking, it is possible to tease these effects apart by careful experimental manipulations. However, based on the design of Experiment 1 and 2, it can b problematic if we directly compare the no- prep. condition ([+SemFlex],[+SynFlex]) with the ba or bei condition ([+SemFlex], [-SynFlex]) to explore syntactic flexibility. This is because the tasks performed in a no-prep. condition and a ba/bei condition possessed different degrees of complexity in both the message component and the grammatical component. First, the task in a no-prep. condition engaged the speaker in a complicated process of message formulation. The speaker needed to conceptualize an event based on two nouns and one verb, a quite complex process that demands decision-making in several aspects. Contrastively, the complexity in conceptualization was greatly reduced in the ba or bei 114 condition, since the event was more or less ascribed. Second, in a no-prep. condition, in order to make a complete sentence, participants needed to retrieve a preposition from the mental lexicon. Contrastively, the syntactic structure had been specified in the ba and bei condition. As a result, the effects of syntactic flexibility may be confounded by comparing the no-prep. condition and the ba/bei condition. If put these concerns aside, the effect of syntactic flexibility may be potentially explored by comparing the DA no-prep. condition ([-SemFlex],[+SynFlex]) with DA-ba/DA-bei conditions ([-SemFlex],[-SynFlex]), which are most similar to V. Ferreira’s flexible and inflexible conditions. Analysis of mean production latency of correct responses revealed that with agent-theme verbs, no significant difference in production latency is found between the DA no-prep. condition (M=2.518) and the DA-ba/DA-bei condition (M=2.541, t1(29)=-.236, p=.815; t2(28)=.049, p=.961). In addition, with theme- experiencer verbs, no significant difference in production latency was found between the DA no-prep. condition (M=2.881) and the DA-ba/DA-bei condition (M=2.828, t1(29)=.320, p=.752; t2(16)=-1.226, p=.238). Despite the fact that the DA no-prep. condition demanded greater task complexity in both the message component and the grammatical component than the DA-ba/DA-bei condition, the results revealed that speakers correctly produced a sentence equally fast in the DA-no prep. condition and in the DA-ba/bei condition. This unexpected outcome may be attributed to a facilitative effect of syntactic flexibility in the DA-no prep. condition, which compensates for the 115 detrimental effect of task complexity. In this sense, the results indeed support the claim that syntactic flexibility facilitates production. 116 Chapter 4: English Sentence Production Task Using Three Verb Types 4.1 Introduction The findings from the two Mandarin production tasks—Experiments 1 and 2—lead to the conclusion that effects of semantic flexibility can be conditioned by verb thematic structure. However, one may wonder if this is a language-specific phenomenon, or is restricted to particular constructions such as Mandarin ba and bei sentences. The main aim of Experiment 3 is to investigate effects of semantic flexibility in English sentence production. The research questions to be addressed are: (i) Are English speakers also sensitive to effects of semantic flexibility in the process of sentence production? Or is it just a specific phenomenon in Mandarin? (ii) If these effects exist in English, are they also conditioned by verb type? (iii) Can such effects be extended to other constructions, such as English actives and passives? The findings of Experiment 3 can shed light on the process of message formulation cross-linguistically. Although Experiment 3 is a sentence production task built upon the methodology of Experiments 1 and 2, it differs from these two experiments in three ways. First, Experiment 3 focused on the effects of semantic flexibility, and thus syntactic flexibility was eliminated from the experimental design. Participants were required to produce a specified structure (i.e., passives or actives) in each condition. Second, the results of Experiments 1 and 2 cannot be compared directly because these results were not obtained in the same experiment and the target items were not equal in number (30 agent-theme verbs in Exp.1; 18 theme-experiencer verbs in Exp.2). In Experiment 3, I treated the 117 factor of verb type as a within-subject variable such that each participant saw different verb types in one experiment. Third, in addition to agent-theme verbs and theme- experiencer verbs, a third verb type, experiencer-theme verbs, was added to the target items. The reason for including experiencer-theme verbs is two-fold. On one hand, experiencer-theme verbs are associated with the same thematic roles as theme- experiencer verbs. Thus, the two verb types are predicted to condition the effects of semantic flexibility in the same way. On the other, experiencer-theme verbs resemble agent-theme verbs in their grammatical-thematic role mappings. Thus, they are predicted to show identical patterns for effects of thematic prominence 7 . Due to these properties, in Experiment 3, experiencer-theme verbs serve as a baseline for comparison with the other two verb types when we examine the effects of semantic flexibility and thematic prominence. More detail will be provided in 4.2. The chapter is outlined as follows. I will first introduce the methodology of Experiment 3 and explain the rationale underpinning my experimental design, followed by the proposed hypothesis and predictions. Next I will report the results obtained from production latency and error rate and discuss about the theoretical implications of the findings. I will then present the error analysis of Experiment 3, followed by a general discussion. 7 In F. Ferreira (1994), she grouped agent-theme and experiencer-theme verbs into the same verb type which she termed normal verbs. 118 4.2 Experiment 3 4.2.1 Methodology Method Like the sentence production tasks in Experiments 1 and 2, participants saw words on a computer screen and were instructed to create a complete sentence using those words (see F. Ferreira 1994; V. Ferreira, 1996 for other work using a similar task). On critical trials, the screen showed (i) a verb, (ii) two nouns and (iii) sometimes a preposition by, depending on the condition. Participants were instructed to plan their sentence before speaking. Producing a sentence requires speakers to come up with an event based on the given entities, and to arrange the words in a proper order. Thus, this task engages both the message component and the grammatical component.. Participants Sixteen graduate/undergraduate students from the University of Southern California participated. All participants were native speakers of English. They were paid $5 for participating in this experiment. Design I manipulated the factors of VOICE (i.e. the active or passive voice) and VERB TYPE. On half of the critical trials, two animate nouns and one verb were given (i.e. the active 119 condition). Participants were told explicitly not to add any preposition on their own, so they could only produce actives on these trials. In contrast, on the other half of the trials, the preposition by was present (i.e. the passive condition). Participants were also told explicitly that they had to use every word on the screen to make a sentence, so they could only produce passives on these trials. In this way, syntactic flexibility is eliminated from Experiment 3. Moreover, the animate-animate noun pair ensures the availability of semantic flexibility because in real world knowledge, either noun could potentially be the subject or the object. For example, it seems plausible for a director to criticize an actress or vice versa. In order to avoid reading bias, noun positions were counterbalanced. On half of the trials, the first noun in the pair was positioned on the left; on the other half trials, the first noun was positioned on the right. Crucially, the verb on a critical trial belonged to one of the three verb types: (i) agent- theme verbs, (ii) theme-experiencer verbs, or (iii) experiencer-theme verbs. The rationale underpinning the inclusion of these three verb types is explained as follows. First, based on the findings of Experiment 1 and 2, when semantic flexibility is available, the two animates would compete for AGENT, but would not compete for THEME and EXPERIENCER. Thus, theme-experiencer and experiencer-theme verbs contrast with agent-theme verbs when it comes to message formulation. 120 Second, based on the findings of F. Ferreira (1994) and my two Mandarin experiments, it is evident that effects of thematic prominence play a role in grammatical encoding such that speakers tend to mention a more prominent thematic role earlier. As a result, agent- theme and experiencer-theme verbs would favor the active voice while theme- experiencer verbs would favor the passive voice. Since syntactic flexibility is not available, production difficulty is expected in conditions where the participants are required to produce the less preferred structure. In this sense, agent-theme and experiencer-theme verbs form a contrast with theme-experiencer verbs in grammatical encoding. Presumably, observed production latency comprises the sum of times spent on both the message component and grammatical component 8 . Therefore, taking effects of semantic flexibility and thematic prominence together, the three verb types are expected to display a three-way distinction in production latency. The conditions depending on VOICE and positional counterbalance are summarized in Table 4.1 8 Undoubtedly, the time spent on phonological encoding is also part of the formulation time of a sentence. However, the phonological component is ignored in the present work, assuming that the phonological encoding process is controlled in my experiments because the conditions involve the production of the same materials as a whole. 121 Table 4.1: The experimental conditions of Experiment 3 POSITION VOICE first noun on the left first noun on the right Active voice (by is not given) words given director, criticize, actress actress, criticize, director possible sentences (1) The director criticized the actress. (1) The director criticized the actress. (2) The actress criticized the director. (2) The actress criticized the director. Passive voice (by is given) words given director, criticize, actress, by actress, criticize, director, by possible sentences (1) The director was criticized by the actress. (1) The director was criticized by the actress. (2) The actress was criticized by the director. (2) The actress was criticized by the director. Materials Targets. Twelve agent-theme verbs, twelve experiencer-theme verbs, and twelve theme- experiencer verbs were selected from Levin (1993) and F. Ferreira (1994). Each verb was assigned a noun pair consisting of occupation names. The two nouns in each pair were matched in syllabic numbers and word length to avoid phonological bias. In addition, they were chosen to be as semantically neutral as possible (e.g. “the actor assisted the actress” vs. “the actress assisted the actor”) to minimize any potential semantic bias. Each target item resulted in four frames based on the interaction of VOICE and positional counterbalance. The resulting 144 frames were assigned into four lists and rotated through the four conditions based on a Latin square design. Participants saw each verb once in one of the four conditions, but encountered all four conditions across the 36 targets. Overall, each participant saw 9 different targets per condition, i.e. 3 different targets per verb. A total of 144 trials across participants, i.e. 48 trials per verb type, were 122 collected for each condition. See Appendix III for a list of the target items used in Experiment 3. The verb phrase was placed at the top, the two nouns were in the middle of the screen, on the left and right side, and the preposition (if present) was at the bottom, as shown in Figure 4.1. The fillers had the same layout as the targets. Figure 4.1: Layout of one experimental trial in Experiment 3 Fillers. In addition to the 36 target items, Experiment 3 also contained 42 filler items. The fillers ranged from 3 to 5 words. To add variability to the sentence types participants needed to create, eight different sentence types were added as fillers. The first type included 4 fillers, consisting of an intransitive reciprocal verb (ex. meet), two animate nouns, and the conjunctive word and. The second type included 6 fillers, consisting of an 123 intransitive verb, an animate noun, and an adverb. The third type consisted of an animate noun and two inanimate nouns, a preposition, and a verb of the spray/load alternation (Levin, 1993). There were 4 fillers of this type. Participants would produce something like “The waitress sprayed perfume on the napkin.” The fourth type included 6 fillers of adjectival passives. Participants would produce something like “The hall was decorated with flowers.” The fifth type consisted of an animate noun, two inanimate nouns, a transitive verb and a locative preposition. There were four fillers of this type. Participants would produce something like “The carrier dropped a package at the door.” The sixth type was double object/dative alternation. There were six fillers of this type. In half of this type, the preposition to was present and the dative construction was expected. In the other half, to was not given and the double object construction was expected. The seventh type was negation, including 3 active-voice sentences and 3 passive-voice sentences, consisting of two animate nouns, a verb belonging to one of the three verb types in the targets, and not. The eighth type included 3 active-voice sentences and 3 passive-voice sentences, consisting of one animate noun, one inanimate noun and a verb. The purpose of the last two types is to add variations to the active or passive voice, so the targets won’t stand out in the experiment. Moreover, of all the fillers, 24 fillers did not have semantic flexibility and 18 fillers had semantic flexibility. In those which did not have semantic flexibility, 12 fillers required participants to start a meaningful sentence with the word on the left side, and 12 fillers 124 required them to start with the word on the right side. The purpose is to prevent participants from using an “always starting from the left” strategy. Procedure Participants sat in front of an ASUS U6 laptop. Spoken responses were recorded with a head-mounted USB microphone. The experiment was run using Paradigm developed by [Perception Research Systems]. Each trial was initiated and ended by pressing the space bar. The words appeared on the screen simultaneously. The experiment began with instructions and 12 practice trials, followed by 78 test trials. Participants were instructed to create well-formed sentences that contained every word shown on the screen. They were told that sometimes it might be necessary to add a function word, including the definite article the, helping verbs did or was, and past-tense suffix –ed. However, they were not informed when to add function words or which function words to add. In addition, they were told explicitly not to add new content words or prepositions. Participants were also instructed that they could freely arrange word order. Importantly, participants were requested to produce their utterances as quickly as they could without sacrificing accuracy. To impose time pressure, after the words appeared on the screen for 5 seconds, a 10000 Hz tone sounded for 350 ms. Participants were encouraged to complete their utterances before the onset of the tone. 125 The entire experiment took approximately twenty minutes. The recordings were then transcribed and analyzed. The dependent variables are production latency and error rate as they serve as indicators of production difficulty level. However, an error did not occur very often in Experiment 3 because the nouns were both animate in each condition, and whichever noun is mentioned first did not affect the well-formedness or plausibility of a sentence. Therefore, production latency may be a better indicator than error rate for this experiment. 4.2.2 Hypotheses and Predictions The design in Experiment 3 allows us to investigate how effects of semantic flexibility and thematic prominence interact with verb type. Importantly, effects of semantic flexibility and thematic prominence may influence different processes of sentence production. On one hand, in the message component, I hypothesize that when the thematic roles to be assigned are AGENT and THEME, competition between the two animates would compete for AGENT. In contrast, when the thematic roles to be assigned are THEME and EXPERIENCER, the competition between animates would not take place. On the other, in the grammatical component, effects of thematic prominence predict that a more prominent thematic role would be more likely to be mentioned earlier. Thus, agent-theme verbs and experiencer-theme verbs would favor the active voice, while theme-experiencer verbs would favor the passive voice. The overall additive effects of 126 thematic prominence and semantic flexibility thus predict a three-way distinction of difficulty level depending on verb type. The predictions are shown in Table 4.2. Table 4.2: The predicted difficulty level of each verb type in Experiment 3 Hypothesized effect verb type Semantic flexibility Thematic prominence overall effect Agent-theme verbs active voice -1 +1 0 passive voice -1 -1 -2 Experiencer-theme verbs active voice 0 +1 +1 passive voice 0 -1 -1 Theme-experiencer verbs active voice 0 -1 -1 passive voice 0 +1 +1 Note: A minus sign indicates an inhibitory effect on production; a plus sign indicates a facilitative effect; a zero indicates no effect. The overall effect is the sum of these effects in each condition. The value is for comparison only and does not reflect magnitude. Table 4.2 shows that both semantic flexibility and thematic prominence are hypothesized to have an impact on sentence production. First, based on the findings of Experiments 1 and 2, I expect that effects of semantic flexibility would be conditioned by verb thematic structure. Specifically, the predictions are: (i) for agent-theme verbs, semantic flexibility would have an inhibitory effect, and (ii) for theme-experiencer and experiencer-theme verbs, semantic flexibility would have no effects. Second, based on the findings on effects of thematic prominence, I predict that in the active condition, thematic prominence would have a facilitative effect when the verb is an agent-theme verb or an 127 experiencer-theme verb (i.e., AGENT or EXPERIENCER first), but would have an inhibitory effect when the verb is theme-experiencer verb (i.e. THEME first). In contrast, in the passive condition, the expected thematic prominence effects are opposite: thematic prominence would have an inhibitory effect when the verb is an agent-theme verb or an experiencer-theme verb (i.e., THEME or THEME first), but would have a facilitative effect when the verb is theme-experiencer verb (i.e., EXPERIENCER first). Presumably, we can look for effects of semantic flexibility by comparing agent-theme and experiencer-theme verbs, and look for effects of thematic prominence by comparing experiencer-theme and theme-experiencer verbs. Moreover, assuming the effects of semantic flexibility and thematic prominence are independent of each other because they involve different components of the language production mechanism, I predict that the overall effects can be reflected in participants’ error rates and production latencies, because the two measures potentially reflect the difficulty level of sentence production. The predictions are summarized as follows: (i) For the production of active voice, the production latency/error rate would be longest/highest for theme-experiencer verbs, followed by agent-theme verbs and experiencer-theme verbs. (ii) For the production of passive voice, the production latency/ error rate would be longest/ highest for agent-theme verbs, followed by experiencer-theme verbs and theme-experiencer verbs. 128 4.2.3 Results 4.2.3.1 Data Preparation Exclusions and Replacements Production latencies more than 2.5 standard deviations from the mean in each condition were removed from subsequent analyses (4.1% of correct responses). No empty cells were obtained in the item analysis or the subject analysis. Coding the Responses Different response types were coded according to the content of the actual utterance to evaluate the difficulty participants experienced in their production. Participants’ utterances were grouped into two main categories: correct responses and errors. A correct response includes the following subcategories: (1) Perfect responses: the sentence contained all given words, was well-formed, fluently produced and semantically plausible. (2) Correct but imperfect responses: the sentence contained all the words and was well-formed and semantically plausible, but accompanied by speech disfluency such as hesitation, stuttering, repetition, or self-correction (e.g., “uh…the gardener was hurt by the sailor.”) An error is defined by one of the following criteria: (1) an ill-formed sentence: the participant failed to add a helping verb ‘was’ in the passive-voice condition. (2) wrong 129 voice: the participant produced the passive voice instead of the active voice or vice versa. (3) word addition: the participant added a word not given on the trial. (4) fail: the participant produced an incomplete sentence or did not say anything. Calculating Production Latencies Production latencies were calculated manually in Praat. Production latencies were measured from the onset of the presentation of a trial to the onset of the first content word of the correct sentence (e.g., the onset of gardener in “Sailor…the gardener hit the sailor”), including the time of hesitation, repetition, self-correction or stuttering, and the duration of the definite article the. Following Experiments 1 and 2, the time of speech disfluency was included in the production latency because it may indicate ongoing speech planning. The reason for including the duration of the definite article is two-fold. First, in 12 tokens of correct responses, the participants began the sentence without the definite article (e.g., “Principal liked the professor.”) Second, it is obvious that in some cases the duration of ‘the’ is lengthened (e.g. “The…gardener hit the sailor.”) This may be due to the incremental nature of grammatical encoding. Participants are likely to initiate their speech as soon as part of the message is encoded and continue constructing the remaining part of the message while building up the structure at the same time. In addition, it may be a strategy to use disfluency as a means of coordinating the transition from thought to speech (Brown-Schmidt & Konopka, 2008). Therefore, it may pause a problem to compare the production latency including a short ‘the’ with that containing a lengthened 130 ‘the’ 9 . Based on the two reasons, I decided that for Experiment 3, the onset of the first content word may serve as a better ending point for calculating production latencies. 4.2.3.2 Results Three sets of results are reported. The first set concerns the proportions of different sentence types in each condition. The second set concerns the error rate for each condition. The third set concerns the mean production latency for correct responses in each condition. Proportions of Sentence Types The proportions of different sentence types produced in each condition are reported in Table 4.3. As can be seen, in conditions where by was present, participants almost always produced passives. In contrast, in conditions where by was not given, participants almost always produced actives. This result verified that syntactic flexibility was not available in Experiment 3. 9 This problem is not present in the data analysis of Experiments 1 and 2. Bare nouns can be definite in Mandarin Chinese, so the participants almost always began a sentence with a bare noun 131 Table 4.3: Number and percentage of sentence types in each condition of Experiment 3 first noun position absence/presence of by left right without by active voice 138 (95.8%) 135 (93.8%) passive voice 5 (3.5%) 7 (4.8%) other 1 (0.7%) 2 (1.4%) Total 144 (100%) 144 (100%) with by active voice 0 (0%) 1 (0.7%) passive voice 139 (96.5%) 141 (97.9%) other 5 (3.5%) 2 (1.4%) Total 144 (100%) 144 (100%) Note: Total number of observation in each cell is 144. Beyond doubt, whether semantic flexibility existed throughout the experiment is an empirical issue. 134 tokens of the 552 correct responses (24%) start with the noun on the right and 418 tokens (76%) start with the noun on the left. This shows that although participants indeed had a tendency to start with the noun on the left side based on the “first-retrieved, first-mentioned” principle, they still exploited the availability of semantic flexibility. Moreover, in an informal interview after the experiment, almost every participant commented that on some of the trials, they realized there could be two possible answers. 132 Error Rate The error rates of each condition are shown in Figure 4.2. ANOVAs were conducted on each condition. The figure shows that POSITION and VOICE did not interact (F1(2,15)=1.154, MSE=.003, p=.300; F2(1,35)=1, MSE=.007, p=.324). The active voice resulted in more errors when the first noun was on the right, while the passive voice resulted in more errors when the first noun was on the left. The main effect of VOICE was marginally reliable by subjects (F1(1,15)=4.286, MSE=.012, p=.056; F2(1,35)=2.059, MSE=.028, p=.160). In general, the active voice incurred a higher error rate than the passive voice. This is surprising at first sight because the passives are assumed to have a more complicated syntactic structure than the actives. I will discuss this issue in 4.4. 0 0.01 0.02 0.03 0.04 0.05 0.06 0.07 first noun on the left first noun on the right Position Production error rate active voice passive voice Figure 4.2: Mean error rate as a function of position and voice in Experiment 3 133 Additionally, in order to find out whether the effects of semantic flexibility and thematic prominence are sensitive to verb type, paired samples t-tests and independent samples t- tests were conducted in the subject and item analyses, respectively. As mentioned above, the predictions are: (i) For the active voice, the error rate would be highest for theme- experiencer verbs, followed by agent-theme verbs and experiencer-theme verbs. (ii) For the passive voice, the error rate would be highest for agent-theme verbs, followed by experiencer-theme verbs and theme-experiencer verbs. Figure 4.3 shows that the predicted pattern was found in the active condition but not in the passive condition. Now let’s take a closer look at the results from paired samples t-tests by subjects and independent samples t-tests by items in the active and passive condition. 0 0.02 0.04 0.06 0.08 0.1 0.12 0.14 active voice passive voice Voice Production error rate agent-theme exp.-theme theme-exp. Figure 4.3: Mean error rate depending on verb type in the active and passive conditions in Experiment 3 134 Active V oice. For the active voice, theme-experiencer verbs had the highest error rate (M=.145, SE=.037), followed by agent-theme verbs (M=.02, SE=.014) and experiencer- theme verbs (M=0, SE=0). Statistically, the error rate for theme-experiencer verbs was significantly higher than that for agent-theme verbs by items (t1(15)=.910, p=.377; t2(22)=3.146, p<.005), and significantly higher than that for experiencer-theme verbs (t1(15)=4.341, p<.005; t2(22)=3.924, p<.005). However, the error rates for agent-theme and experiencer-theme verbs did not differ significantly (t1(15)=.167, p=.869; t2(22)=1.483, p=152). This pattern conforms to the first prediction in that participants make errors most often with theme-experiencer verbs. Moreover, the significant difference between theme-experiencer and experiencer-theme verbs suggests that such difficulty can be attributed to an inhibitory effect of thematic prominence on theme- experiencer verbs. However, since no significant difference was found between agent- theme and experiencer-theme verbs, no evidence was provided for an inhibitory effect of semantic flexibility with agent-theme verbs. Passive V oice. For the passive voice, the error rates for the three verb types did not differ from one another. The error rate for theme-experiencer verbs (M=.031, SE=.016) was not different from the error rate for experiencer-theme verbs (M=.031, SE=.016) (t1(15)=0, p=1; t2(22)=0, p=1) or agent-theme verbs (M=.020, SE=.020) (t1(15)=.075, p=.941; t2(22)=.394, p=.698). The error rates for agent-theme and experiencer-theme verbs did not differ significantly, either (t1(15)=.082, p=.936; t2(22)=.394, p=.698). Thus, the 135 prediction that agent-theme verbs would be hardest to produce in the passive condition due to the inhibitory effects of both semantic flexibility and thematic prominence was not supported by the error rate patterns. Since in general, the error rates in the passive condition were very low, (around 2-3%), it is likely that these error rates exhibited ceiling effects. Therefore, production latency may serve as a better indicator than error rate for the passive condition. Mean Production Latency The production latencies of each condition are shown in Figure 4.4. ANOVAs were conducted on each condition. The figure shows that POSITION and VOICE did not interact (F1(2,15)=1.364, MSE=.068, p=.261; F2(1,35)=.169, MSE=.133, p=.684). As expected of a control variable, there was no main effect of POSITION (M=(F1(1,15)=.264, MSE=.028, p=.615; F2(1, 15)=.095, MSE=.057, p=.760.) In contrast, the main effect of VOICE was highly reliable (F1(1,15)=6.496, MSe=3.678, p<.05; F2(1,35)=26.316, MSe=8.326, p=0). Across the board, the passive voice incurred longer production latencies than the active voice, which is expected because the passive voice is more complicated than the active voice structurally. 136 1600 1800 2000 2200 2400 2600 first noun on the left first noun on the right Position Mean production latency (ms) active voice passive voice Figure 4.4: Mean production latency as a function of position and voice in Experiment 3 Additionally, in order to examine effects of semantic flexibility and thematic prominence, paired samples t-tests and independent samples t-tests were conducted in the subject and item analyses, respectively. As already mentioned, the predictions are: (i) For the active voice, the production latency would be longest for theme-experiencer verbs, followed by agent-theme and experiencer-theme verbs. (ii) For the passive voice, the production latency would be longest for agent-theme verbs, followed by experiencer-theme verbs and theme-experiencer verbs. As Figure 4.5 shows, the patterns in both conditions do not conform to the predictions. Now let’s take a closer look at the different verb types in the 137 active and passive conditions to further examine effects of semantic flexibility and thematic prominence. 0 400 800 1200 1600 2000 2400 active voice passive voice Voice Mean production latency (ms) agent-theme exp.-theme theme-exp. Figure 4.5: Mean production latency depending on verb type in the active and passive conditions in Experiment 3 Active V oice. For the active voice, the three verb types did not differ from one another. The mean production latency of theme-experiencer verbs (M=1.780, SE=.051) is not different from that of experiencer-theme verbs (M=1.795, SE=.068)(t1(15)=-.611, p=.550; t2(22)=-.181, p=.858) or agent-theme verbs (M=1.815, SE=.044) (t1(15)=-.516, p=.611; t2(22)=-.815, p=.428). The production latencies for agent-theme and experiencer-theme 138 verbs did not differ significantly either (t1(15)=.440, p=.666; t2(22)=.240, p=.813). Thus, neither effects of semantic flexibility nor effects of thematic prominence were found in the active condition. An account for the lack of significant differences here will be provided in 4.4. Passive V oice. For the passive voice, experiencer-theme verbs had the longest production latency (M=2.295, SE=.112), followed by agent-theme verbs (M=2.112, SE=.104) and theme-experiencer verbs (M=2.097, SE=.057). Statistically, the production latency of experiencer-theme verbs was significantly longer than that of theme-experiencer verbs by subject (t1(15)=2.506, p<.05; t2(22)=1.569, p=.131), but not significantly higher than that of agent-theme verbs (t1(15)=1.283, p=.219; t2(22)=1.196, p=.245). In addition, the production latencies of agent-theme and theme-experiencer verbs did not differ significantly (t1(15)=.125, p=.901; t2(22)=.137, p=.893). This pattern does not support the second prediction that passives with agent-theme verbs would be the hardest to produce. Moreover, semantic flexibility effects were not found because agent-theme verbs and experiencer-theme verbs did not differ. The only significant difference was found between experiencer-theme and theme-experiencer verbs: Participants spent longer time producing the passive voice with experiencer-theme verbs than with theme- experiencer verbs, providing evidence for a thematic prominence effect. An interesting question is why experiencer-theme verbs have resulted in a numerically longer production latency than agent-theme verbs, given the hypothesis that semantic flexibility 139 has an inhibitory effect on the latter. I suggest this can be attributed to the composition of the subtypes of correct responses pertaining to the experiencer-theme verbs. A detailed explanation will be provided in 4.3.2. 4.3 Error Analysis of Experiment 3 In addition to analyzing production latency and error rate, I also conducted an analysis on the errors as well as the subtypes of correct responses. These data can look into the process of sentence production and illuminate some phenomena that may be overlooked by error rate and production latency, especially when they fail to reflect potential differences between conditions due to ceiling effects. In 4.3.1, I report the occurrence frequency of different types of errors depending on verb type. In 4.3.2, I report the occurrence frequency of the subtypes of correct responses depending on verb type. In both sections, a comparison is made between the three verb types, along with theoretical accounts for the observed patterns. 4.3.1 Error types The overall error rate of Experiment 3 was very low (4.1%). Of the three verb types, theme-experiencer verbs incurred the most errors (X 2 =13.654, p<.005). Table 4.4 shows the count of different error types in the active and passive conditions depending on verb type. Due to the small number of observations, the comments are somewhat speculative. 140 Table 4.4: Count of different error types depending on verb type in Experiment 3 Agent-theme Experiencer-theme Theme-experiencer error type active passive active passive active passive Ill-formed sentence 0 1 0 2 0 0 Wrong voice 1 0 0 0 11 1 Word addition 1 0 0 1 2 2 Fail 0 1 0 0 1 0 Total 2 2 0 3 14 3 The first type is an ill-formed sentence (e.g., “The clergyman … cherished by the detective”). In these errors, participants did not add a helping verb ‘was’ to form a passive sentence, resulting in ungrammaticality. Errors of this type are more akin to a talking problem, not a thinking problem. When the speaker blurted out a sentence under time pressure, s/he may fail to retrieve a word of grammatical function, but the meaning of the proposition remains intact. The second type is use of the wrong voice. For example, in the active condition, the speaker says passives instead. This is the most frequent error type. Notably, 11 of the 13 tokens of this type occurred in the active condition with theme-experiencer verbs. Due to effects of thematic prominence, participants would prefer to use passives instead of actives with theme-experiencer verbs (e.g., “the cashier was confused by the trainer” rather than “the cashier confused the trainer”). Notably, this type of error rarely occurred with agent-theme or experiencer-theme verbs. Although the preferred structure of these two verb types is the active voice, participants never produced actives in the passive 141 condition. This is probably because the presence of ‘by’ in the passive condition provides a strong clue for the specified sentence type. Therefore, it is also not surprising that the passive condition exhibited a ceiling effect in error rate, as we already saw in 4.2.3.2. The third type is word addition. In these errors, participants added a conjunction word ‘and’, a negation word ‘not’ or another preposition such as ‘at’ to make a sentence. I suggest this may due to a priming effect from the fillers because the fillers contain these sentence types. Finally, the fourth type is what we have termed ‘fail’. Participants either said nothing or did not finish the sentence. There were only two observations of this type. In conclusion, the error analysis shows that participants encountered more difficulty when they were asked to produce active sentences with theme-experiencer verbs, which can be attributed to the thematic prominence effect. However, participants were less error-prone with agent-theme and experiencer-theme verbs in producing passive sentences. This asymmetry may be attributed to the presence of by in the passive condition, which greatly reduces the chance of making errors. 4.3.2 Subtypes of Correct Responses There are five subtypes of correct responses: (i) perfect responses, (ii) hesitation, (iii) correction, (iv) stuttering, and (v) repetition. Except for being a perfect response, a correct response may fit into one or more categories. Table 4.5 shows the count of each 142 subtype for each verb type. By carefully inspecting the subcomponents of correct responses, we are able to gain insight into what may have caused a particularly long or short mean production latency for a certain verb type. Table 4.5: Count of the subtypes of correct responses depending on verb type in Experiment 3 subtype of Agent-theme Experiencer-theme Theme-experiencer correct responses active passive active passive active passive Perfect response 71 61 77 39 55 68 Correction 1 4 0 19 1 6 Hesitation 18 26 13 32 22 16 Stuttering 4 3 3 5 4 2 Repetition 0 1 3 3 0 1 Total 94 95 96 98 82 93 The first type is a perfect response. In this type, participants produce the correct answer without any speech disfluencies. In general, there were more perfect responses in the active condition with agent-theme and experiencer-theme verbs, and more perfect responses in the passive condition with theme-experiencer verbs. This contrast is expected due to the thematic prominence effect. The remaining four types are variants of speech disfluency. I will focus on two of them that had a dominant number of observations: hesitation and correction. Hesitation is the most common type, suggesting that participants sometimes initiated an utterance before speech planning was completed, especially when they felt time pressure (e.g., “The principal…was liked by the professor.”) This phenomenon can be attributed to the 143 incremental nature of grammatical encoding. As Table 4.5 shows, speakers hesitated more in the passive condition with agent-theme and experiencer-theme verbs, while they hesitated more in the active condition with theme-experiencer verbs. This again provides evidence for the thematic prominence effect. In correction, participants either corrected the subject noun or the voice halfway. Such a behavior usually reflects the speaker’s change of mind or uncertainty in message. Notably, in the passive condition, experiencer-theme verbs have the fewest counts of perfect response (X 2 =8.179, p<.05), but the most counts of correction (X 2 =12.7, p<.005). This indicates that participants were often uncertain about the intended message when using experiencer-theme verbs in passive voice. The unstable state may be caused by the special properties of experiencer-theme verbs. In message formulation, semantic flexibility is hypothesized to have no effects on thematic role assignment with an experiencer-theme verb, since EXPERIENCER or THEME are both likely to be animate. This explains why participants sometimes switched between nouns due to multiple options (e.g. “The clergyman…the detective was cherished by the clergyman.”) Moreover, in grammatical encoding, the preferred structure for experiencer-theme verbs is the active voice (i.e. EXPERINCER first). The relative ease of thematic role assignment may lead the speaker to initiate speech before the structure is fully activated, which is often the case of a highly incremental process. This could explain why participants often changed the voice right after they realized the specified structure was the passive voice (e.g. “The student hated…was hated by the plumber.”) As we saw, experiencer-theme verbs had a 144 numerically longer production latency than the other two verb types in the passive condition. The larger proportions of correction and hesitation may have contributed to the relatively long mean production latency. 4.4 Discussion The results of Experiment 3 are summarized as follows. To begin with, when we consider error rates, we see that for the production of the active voice, participants made errors more often with theme-experiencer verbs than with experiencer-theme verbs. Such a difference might be solely attributed to the effect of thematic prominence. In contrast, no significant difference was found between the error rates for agent-theme and experiencer- theme verbs. In other words, an inhibitory effect of semantic flexibility on agent-theme verbs was not found. For the production of the passive voice, the error rates of the three verb types did not differ from one another. That is, neither effects of semantic flexibility nor thematic prominence were found. As discussed above, the lack of effect might be due to a ceiling effect in the passive condition caused by the presence of ‘by’. Note that the error rates were very low in the passive condition across the board (i.e., 2% ~3%). In fact, the error rate of the passive condition was lower than the error rate of the active condition. This seems unexpected given passives are normally harder to produce than actives due to structural complexity and occurrence frequency. Recall that in the error analysis, we saw that participants were more likely to make errors in the active condition with theme- experiencer verbs. However, they never made errors in the passive condition with agent- 145 theme and experiencer-theme verbs. A possible account for this asymmetry could be that the presence of by in the passive condition excluded the possibility of producing actives because the participants were explicitly told to use every word on the screen. Now, turning to the production latencies, we see that for the production of the active voice, the three verb types did not differ from one another. This could be explained by a ceiling effect because the active voice consists of the simplest SVO structure in English, leading to an ease of production. In contrast, for the production of passive voice, the production latency for experiencer-theme verbs was significantly higher than that for theme-experiencer verbs by subjects, again suggesting an effect of thematic prominence. However, like the results of error rates, an inhibitory effect of semantic flexibility was not found because there was no significant difference between agent-theme verbs and experiencer-theme verbs. As a whole, the active condition seems to exhibit a ceiling effect in production latency, while the passive condition seems to exhibit a ceiling effect in error rate. Moreover, only effects of thematic prominence but not semantic flexibility were found in Experiment 3. These results have brought about two speculations: First, is Experiment 3 too simple to show real effects? Second, do semantic flexibility effects exist in English or are they just a language-specific phenomenon in Mandarin? 146 To address the first question, let’s take a closer look at the differences existing in the Mandarin and English experiments. In Experiments 1 and 2, effects of semantic flexibility were shown to interact with verb type: With agent-theme verbs, participants were slower to produce a sentence when the nouns were both animate ([+SemFlex]), compared with situations where the nouns differed in animacy ([-SemFlex]). In contrast, with theme-experiencer verbs, slowdowns were not observed in the animate-animate condition. Moreover, this asymmetry was consistent with or without syntactic flexibility. Therefore, it is reasonable to suspect that such an effect can be extended to Experiment 3 which targets on English. However, although Experiment 3 was built on the methodology of Experiments 1 and 2, two experimental distinctions can be identified. First, the English experiment only had sixteen participants, while the Mandarin experiments had thirty participants each. The small number of participants in Experiment 3 might not suffice to demonstrate statistical differences. Second, the SA-ba and SA-bei conditions of Experiments 1 and 2 resemble the active and passive conditions of Experiment 3 because all of these conditions are [+SemFlex] and [-SynFlex]. That is, in these conditions, speakers had options between two propositions by but they could only produce the specified structure. If we compare the production latencies in these conditions, it seems fair to conclude that the Mandarin experiments were harder than the English experiment. As Table 4.6 shows, in general, participants responded much faster in English (i.e.the shaded cells). In an informal interview after Experiment 3, many participants commented that among the various 147 sentence types of all the trials (including targets and fillers), those requiring them to produce the active or passive voice were the most straightforward. Moreover, as noted earlier, the error rate of the passive condition and the production latency of the active condition seemed to exhibit ceiling effects and failed to show any differences among the three verb types. As a result, it is reasonable to suspect that the small number of participants (n=16) and the simple nature of the task have prevented Experiment 3 from showing real effects of semantic flexibility. Table 4.6: A comparison of the mean production latency of the Mandarin SA-ba/SA-bei conditions and the English active/passive conditions (shaded) verb type SA-ba condition Active condition SA-bei condition Passive condition Agent-theme verbs 2847 ms 1815ms 2881 ms 2112ms Theme-experiencer verbs 2549ms 1780ms 2701ms 2097ms In addition to the problems of a small sample and task simplicity, another potential problem of Experiment 3 may be the slow alarm signal. Following the design of Experiments 1 and 2, participants heard a tone five seconds after the target words appeared on the screen. The purpose of the tone is to prevent participants from sacrificing speed for accuracy, because normal speech is usually not delayed. However, based on the fact that the mean production latency for each verb type ranged from 1780ms to 2295 ms, the 5-second alarm signal might have been too slow. In other words, the participants might have been allowed too much time for making a response on a trial. Thus, to 148 uncover real effects, a follow-up experiment with a larger sample and enhanced complexity is desirable. With regard to the second question, the lack of evidence for semantic flexibility effects does not necessarily lead to the conclusion that English speakers are not sensitive to semantic flexibility in message formulation. As discussed in the error analysis, the proportion of perfect responses in the passive condition shows a significant difference between agent-theme verbs and experiencer-theme verbs. In specific, although both agent-theme verbs and experiencer-theme verbs are less preferred with passive sentences, participants made fewer perfect responses but more corrections with experiencer-theme verbs than with agent-theme verbs. This shows that speakers were sensitive to the interaction between different thematic roles and animate nouns. To further investigate the effects of semantic flexibility in English, follow-up experiments with improvements were conducted (i.e. Experiments 4 and 5) and will be reported in Chapter 6. Before turning to those experiments, I will first analyze the perfect responses of Experiments 1-3 to find out if the results would be divergent from those obtained from correct responses. 149 Chapter 5: Analyzing Perfect Responses 5.1 Introduction In this chapter, I will present analyses on the perfect responses in Experiments 1 and 2 (on Chinese) and Experiment 3 (on English) by studying two sets of data: (i) the perfect- response rate, and (ii) the mean production latency. The primary goal is to explore whether speakers behaved differently when making a perfect response and a correct response, as these two are categorized by distinct criteria in my coding system. If the two ways of coding have resulted in divergent production patterns, a natural question is: What may have contributed to this variability? This attempt provides us with an opportunity to evaluate the validity of different criteria for measuring the dependent variables (i.e. response type and production latency), and potentially opens a new window for understanding the processes of message formulation and grammatical encoding. As a reminder, Experiments 1-3 involved sentence production tasks. Participants were given a few words on the screen and asked to make a complete sentence. In the Mandarin experiments, I manipulated the ANIMACY and PREPOSITION CONSTRAINT factors, such that participants either saw a verb together with an animate-animate noun pair (the SA condition) or an animate-inanimate noun pair (the DA condition). In addition, they either saw a preposition such as ba or bei, or no preposition. In the English experiment, I manipulated the VERB TYPE and PREPOSITION CONSTRAINT factors. Participants saw a verb belonging to one of three different types (i.e., agent-theme verbs, experiencer- 150 theme verbs or theme-experiencer verbs), and saw the preposition by in half of the trials (i.e. the passive condition). In my coding system, correct responses are further divided into three subcategories: (i) perfect response, (ii) correct but imperfect response, and (iii) correct but unexpected response. As we have seen, the design of the Mandarin experiments included conditions where syntactic flexibility was present (i.e. the no-preposition condition) or absent (i.e. the BA or BEI condition). The no-prep. condition only presented two nouns and a complex verb phrase, as displayed in (1) 10 . This is because in Mandarin, SVO word order is usually ungrammatical when a sentence contains a complex verb phrase. Thus, in the no-prep.condition, participants were expected to add ba or bei or other prepositions to make a grammatical sentence. In my coding system, only ba and bei sentences without any speech disfluencies was coded as a perfect response. If with some sort of speech 10 For (1), it is ungrammatical (i.e. an error) if the participant produces an SVO sentence. Ex. * Nanhai jinbao zai-huai-li nuhai. boy hold tightly in the arms girl Intended meaning: “The boy held the girl tightly in his arms.” In contrast, a correct response may belong to one of the following subcategories: 1. Perfect response (BA or BEI) : Ex. Nanhai BA nuhai jinbao zai-huai-li. boy BA girl hold tightly in the arms “ The boy held the girl tightly in his arms.” Ex. Nanhai BEI nuhai jinbao zai-huai-li. boy BEI girl hold tightly in the arms “ The boy was held tightly by the girl in her arms.” 2. Correct but imperfect response (with speech disfluencies) : Ex. Hm…nanhai BA nuhai jinbao zai-huai-li. Hm…boy BA girl hold tightly in the arms “ Hm…the boy held the girl tightly in his arms.” 3. Correct but unexpected response (other prepositions): Ex. Nanhai JIANG nuhai jinbao zai-huai-li. boy JIANG girl hold tightly in the arms “ The boy held the girl tightly in his arms.” 151 disfluencies, it was then coded as a correct but imperfect response. If the response belonged to other sentence types, such as NP-conjoined sentences or sentences containing a preposition other than ba and bei, it was then categorized as a correct but unexpected response. (1) Clearly, perfect responses are a subset of correct responses. The reason to analyze perfect responses independently is based on the speculation that the nature of correct responses may be different from that of perfect responses. Although the nature of correct responses is presumbaly typical of natural speech since speech disfluencies are not unusual in sentence production, the unequal proportions of speech disfluencies in each condition could have confounded the attempt to evaluate the difficulty level between conditions. Moreover, since correct responses include responses with or without speech disfluencies, 男孩男孩男孩男孩 boy 女孩女孩女孩女孩 girl 緊抱在懷裡緊抱在懷裡緊抱在懷裡緊抱在懷裡 hold tight in the arms 152 it is hard to identify the scenarios which may induce distinct behaviors of the language production system, under the assumption that the production system may be adaptable to the contexts or circumstances. In contrast, the analysis of perfect responses potentially allows us to investigate the behaviors of the language production mechanism in scenarios of homogeneity. To be coded as a perfect response, a well-formed plausible sentence needed to contain every given word on a trial without speech disfluencies. Conceivably, a perfect response is likely to occur in two scenarios which vary in the degree of incrementality of advance linguistic planning. The first scenario is assumed to depict a highly incremental process, where a perfect response may be produced extemporaneously due to easy formulation. This scenario may be characteristic of a relatively short mean production latency compared with correct responses. The second scenario is assumed to depict a less incremental process, where a perfect response may be ‘planned’ before it is uttered. In such cases, the speaker may consciously monitor their speech till accuracy is confirmed, resulting in a tradeoff between speed and accuracy 11 . As a result, ‘prepared’ perfect responses are assumed to have a relatively long mean production latency compared with a less prepared response. Crucially, we need to identify in which scenario perfect responses occurred before interpreting the data. 11 For example, in his language production model, Levelt (1983, 1989) proposed a monitoring component for recognizing pre-articulatory and overt errors. The monitoring processes are hypothesized to occur thoughout spontaneous speech. First, during conceptualization, the monitoring component checks the appropriateness of an intended message. Second, when message formulation is finished but before articulation, the monitoring component is responsible for parsing the inner speech resulting from speech planning. Third, one can listen to what one says (i.e. hearing monitoring). 153 Existing work investigating the flexibility of sentence production processes posits that these processes are to some extent susceptible to linguistic contexts or speakers’ strategies. For instance, in message formulation, results from the sentence production tasks in van Nice & Dietrich (2003) suggested that when speeh is less prepared, the process of referent processing seems more ‘sequential’, whearas in planned speech, referent processing seems more ‘compact’ (i.e., more bits of information are processed overlappingly). Additionally, eye-tracking data also suggested that ‘‘planned’’ picture descriptions involve a more compact processing of referents than do ‘‘spontaneous’’ descriptions of pictures that remain in view (Griffin, 1998; Griffin & Bock, 2000). In grammatical encoding, F. Ferreira & Swets (2002) proposed that the degree to which the production is incremental is under strategic control. Their studies showed that under time pressure, speakers would tend to initiate speech before the utterances are fully planned, whereas speakers would postpone speaking until an utterance is fully planned if they are allowed more time. Based on these findings, under the assumption that perfect responses may take place in a scenario distinct from correct responses, we expect to see certain discrepancies in speakers’ production behaviors. I have presented the results from the correct responses in the two Mandarin experiments (Chapter 2 and 3) and the English experiment (Chapter 4). In this chapter, I first report the analysis on perfect-response rates and the production latencies of perfect responses in these experiments, and then discuss what these results may suggest for effects of semantic flexibility and thematic prominence. The findings from perfect responses are 154 then compared with those of correct responses in search of systematic variation. As will be discussed in more detail below, the findings suggest that both effects of semantic flexibility and thematic prominence may be weakened from perfect responses. I argue that these findings provide additional evidence for the account that linguistic contexts or speakers’ strategies may influence the processes of sentence production. 5.2 Perfect Responses in Experiments 1 and 2 (Chinese) Tables 5.1 and 5.2 show the count of each subtype of correct responses in Experiment 1 (with agent-theme verbs) and Experiment 2 (with theme-experiencer verbs). As shown in tables 5.1 and 5.2, the proportion of perfect responses was generally lower in a no-prep. conditions than in a ba or bei condition, most likely because the no-prep conditions required the speaker to figure out an event without structural cues (i.e., the preposition). As a result, when comparing the results obtained from correct responses and perfect responses, it is worth keeping in mind that discrepancies are more likely to exist in the SA and DA no-prep. conditions than in the ba and bei conditions. Table 5.1: Count of each subtype of correct responses in Experiment 1 DA SA Subtype of correct responses no-preposition ba bei no-preposition ba bei Perfect response 73 (60.8%) 131 (88.5%) 103 (76.9%) 48 (58.5%) 117 (82.4%) 117 (82.4%) Correct but imperfect response 32 (26.7%) 17 (11.5%) 30 (22.4%) 19 (23.2%) 24 (16.9%) 19 (13.4%) Correct but unexpected response 15 (12.5%) 0 1 (0.7%) 15 (18.3%) 1 (0.7%) 6 (4.2%) Total no. of correct responses 120 (100%) 148 (100%) 134 (100%) 82 (100%) 142 (100%) 142 (100%) Note: The percentage refers to the proportion of perfect responses of the correct responses. 155 Table 5.2: Count of each subtype of correct responses in Experiment 2 DA SA Subtype of correct responses no-preposition ba bei no-preposition ba bei Perfect response 56 (77.8%) 55 (69.6%) 66 (82.5%) 41 (57.8%) 79 (89.8%) 71 (79.8%) Correct but imperfect response 14 (19.4%) 24(30.4%) 14 (17.5%) 15 (21.1%) 9 (10.2%) 16 (18%) Correct but unexpected response 2 (2.8%) 0 0 15 (21.1%) 0 2 (2.2%) Total no. of correct responses 72 (100%) 79 (100%) 80 (100%) 71 (100%) 88 (100%) 89 (100%) Note: The percentage refers to the proportion of perfect responses of the correct responses. 5.2.1 Experiment 1 Exclusions and replacements For the analyses reported in this chapter, the item analysis and subject analysis were conducted on the perfect-response rates and production latencies of perfect responses. In the item analysis, one item was excluded due to programming error. In the subject analysis, one out of thirty participants was excluded due to a very low perfect-response rate, with only one instance of perfect response out of 30 trials. In addition, the stringent criterion for a perfect response resulted in a greater number of empty cells. A total of 10 empty cells in the item analysis out of 174 cells and 8 in the subject analysis out of 174 cells were replaced with row and column means as in Winer (1971). Moreover, for each condition, any production latency greater than 2.5 standard deviations above the mean in each condition was trimmed from the analysis, amounting to 7 observations being excluded (1.1%). Two sets of data are reported here. The first set concerns the percentage of perfect responses (i.e. perfect-response rate) in each condition. The second set concerns the mean 156 production latency of perfect responses in each condition. A higher perfect-response rate and a shorter mean production latency are assumed to indicate a lower difficulty level of producing sentences in a condition. Perfect-response rate The perfect-response rate in each condition is shown in Figure 5.1. In my analysis, perfect-reponse rate refers to the percentage of perfect responses out of all responses (including correct responses and errors) in each condition. In other words, the perfect- response rate indicates the probability for a perfect response to take place in a condition. The figure shows that the two factors ANIMACY and PREPOSITION CONSTRAINT were interacting. The SA and DA conditions had the lowest perfect-response rate when the preposition was not given, but the patterns were mixed in the ba and bei conditions: the DA-ba condition has a higher perfect-response rate than the SA-ba condition, while the DA-bei condition has a lower perfect-response rate than the SA-bei condition. ANIMACY appeared to have some effect because both the SA-no prep. condition and the SA-ba condition had a lower perfect-response rate than their DA counterparts. PREPOSITION CONSTRAINT also had some effect because the perfect-response rate was much lower without a preposition in both the DA and SA conditions. ANOVAs demonstrated that the main effect of ANIMACY was reliable by subjects (F1(1,28)=4.976, MSE=.187, p<.05; F2(1, 28)=.876, MSE=.055, p=.357). The main 157 effect of PREPOSITION CONSTRAINT was highly reliable (F1(2,58)=72.860, MSE=3.167, p<.001; F2(2,56)=88.396, MSE=4.357, p<.001). The interaction between the two factors was also highly reliable (F1(2,58)=8.890, MSE=.371, p<.001; F2(2,56)=15.360, MSE=.355, p<.001). 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 no preposition ba bei Preposition constraint Perfect-response rate Different animacy Same animacy Figure 5.1: Mean perfect-response rate as a function of animacy and preposition constraint in Experiment 1 158 Additionally, to gain a better understanding of what differences are driving the interaction between ANIMACY and PREPOSITION CONSTRAINT, I took a closer look at the different conditions by means of planned comparisons. Semantic Flexibility. To begin with, to test the effects of semantic flexibility, planned comparisons were conducted on the perfect-response rates between (i) the SA no-prep. condition ([+SemFlex, +SynFlex]) and the DA no-prep. condition ([-SemFlex, +SynFlex]), and (ii) the SA-ba / SA-bei conditions ([+SemFlex, -SynFlex]) and the DA- ba/ DA-bei conditions ([-SemFlex, -SynFlex]). The prediction is the SA conditions would have a lower perfect-response rate than their DA counterparts based on the hypothesized inhibitory effect of semantic flexibility with agent-theme verbs. As predicted, the SA-no prep. condition had a significantly lower perfect-response rate (M=.32, SE=.050) than the DA-no prep. condition (M=.487, SE=.049) (t1(28)=-2.504, p<.05; t2(28)=-2.693, p<.05). Additionally, the SA-ba condition had a significantly lower perfect-response rate (M=.78, SE=.040) than the DA-ba condition (M=.873, SE=.022) (t1(28)=-2.674, p<.05; t2(28)=-4.894, p<.005). These results support the hypothesis that having semantic flexibility increases production difficulty with agent-theme verbs. However, an opposite pattern was observed between the SA-bei condition (M=.78, SE=.029) and the DA-bei condition (M=.687, SE=.038): The SA-bei condition had a significantly higher perfect-response rate than the DA-bei condition (t1(28)= 2.588, 159 p<.05; t2(28)=2.108, p<.05). The less preferred inanimate THEME subject of a bei sentence (e.g. “the movie BEI the director criticized quite a bit”) may have contributed to the lower perfect-response rate in the DA-bei condition. Thematic Prominence. Second, to test the effects of thematic prominence, planned comparisons were conducted on the perfect-response rates between (i) the SA-ba condition and the SA-bei condition, and (ii) the DA-ba condition and the DA-bei condition. The prediction is the ba conditions would have a higher perfect-response rate than the bei conditions based on the hypothesis that speakers would prefer to mention a more prominent thematic role earlier (i.e. AGENT first in Experiment 1), and thus ba sentences would be easier to produce than bei sentences. As predicted, the perfect-response rate of the DA-ba condition was significantly higher than that of the DA-bei condition (t1(28)=4.474, p<.001; t2(28)=4.157, p<.001), indicating that it is indeed easier for speakers to produce ba sentences than bei sentences with agent-theme verbs. However, the SA-ba and SA-bei conditions did not differ significantly in perfect-response rate probably due to the ceiling effect (t1(28)=-.384, p=.704; t2(28)=-.509, p=.616). 160 Animacy. Finally, an effect of animacy was also found. The difference in the perfect- response rates between the DA-ba condition and the DA-bei condition was greater than that between the SA-ba and SA-bei conditions (18.6% vs. 0%). The result suggests that effects of animacy and thematic prominence are independent of each other. In sum, on the basis of analysis on the perfect-response rate in Experiment 1, we can conclude that semantic flexibility posed certain difficulty when the sentence contrained an agent-theme verb, because perfect responses occurred less often in the SA conditions. In addition, due to thematic prominence effects, perfect ba-sentences were produced more often than perfect bei-sentences, as demonstrated by higher rates of perfect responses in the ba conditions. Mean Production Latency The mean production latency of perfect responses in each condition is shown in Figure 5.2. The figure shows that the two factors ANIMACY and PREPOSITION CONSTRAINT did not interact. The patterns in the SA conditions and the DA conditions looked similar. ANIMACY appeared to have some effect such that the SA conditions had longer latencies than the DA conditions across the three preposition constraint conditions. PREPOSITION CONSTRAINT also had some effect such that ba sentences were produced fastest and bei sentences were produced slowest in both the SA and DA conditions. Notably, in 161 Experiment 1, the general patterns of the mean production latency of perfect responses are almost identical with those of correct responses (c.f. Figure 2.5). ANOVAs demonstrated that the main effect of ANIMACY was reliable (F1(1,28)=22.957, MSE=3.920, p<.001; F2(1,28)=5.476, MSE=1.955, p<.05). The main effect of PREPOSITION CONSTRAINT was highly reliable (F1(2,56)=9.811, MSE=1.598, p<.001; F2(2,56)=7.561, MSE=2.191, p<.005). The interaction between the two factors was not reliable (F1(2,56)=1.038, MSE=.142, p=.358; F2(2,56)=1.956, MSE=.752, p=.159). 2000 2200 2400 2600 2800 3000 no preposition ba bei Preposition constraint Mean production latency (ms) Different animacy Same animacy Figure 5.2: Mean production latency of perfect responses as a function of animacy and preposition constraint in Experiment 1 162 Additionally, to test the predictions regarding effects of semantic flexibility, thematic prominence and animacy, independent samples t-tests and paired samples t-tests were conducted on the production latency of different conditions. Semantic Flexibility. First, to explore the effect of semantic flexibility, independent samples t-tests were conducted on production latencies between the SA-no prep. and DA- no prep. conditions. 12 The hypothesis is that semantic flexibility would hinder production with agent-theme verbs. Thus, participants are predicted to spend longer production latencies producing both ba and bei sentences in the SA-no prep. condition than in the DA- no prep. condition. Participants’ perfect responses were grouped by sentence type and the mean production latency of ba sentences and bei sentences were compared respectively between these two conditions. The mean production latencies of the perfect responses depending on sentence type are shown in Figure 5.3. 12 Independent samples t-tests instead of paired t-tests were conducted for the no-prep.conditions due to the unequal numbers of the ba and bei sentences. 163 2469.6 2335.2 2525.3 2822 2000 2200 2400 2600 2800 3000 ba bei sentence type production latency (ms) Different animacy Same animacy Figure 5.3: Mean production latency of ba and bei sentences categorized as perfect responses in the SA-no prep. condition vs. the DA-no prep. condition in Experiment 1 For ba-sentences, the SA-no prep. condition had a numerically longer production latency (M=2.525, SE=.141) than the DA-no prep. condition (M=2.469, SE=.133), but this difference was not significant (t(77)=.287, p=.775). Similarly, for bei-sentences, the SA- no prep. condition had a numerically longer production latency (M=2.822, SE=.240) than the DA-no prep. condition (M=2.335, SE=.163), but this difference was not significant either (t(39)=1.671, p=.105). Unlike the results of correct responses, the results did not reveal an inhibitory effect of semantic flexibility with agent-theme verbs. Instead, the results suggested that when a perfect response was produced in a no-prep. condition, whether the two nouns were both animate or differ in animacy did not have an effect on production. 164 In addition, paired samples t-tests were conducted on the production latencies between (i) the SA-ba condition and the DA-ba condition, and (ii) the SA-bei condition and the DA- bei condition. Again, the prediction is that the two SA conditions would have a longer mean production latency than their DA counterpart. As shown in Figure 5.4, the predictions are confirmed: The SA-ba condition (M=2.612, SE=.101) had a significantly longer production latency than the DA-ba condition (M=2.210, SE=.070) (t1(28)=5.166, p<.005; t2(28)= 3.521, p<.005). In addition, the SA- bei condition (M=2.882, SE=.142) had a significantly longer production latency than the DA-bei condition (M=2.589, SE=.101) by subjects (t1(29)=2.575, p<.05; t2(28)= 1.204, p=.239). The results are consistent with the results of correct responses. 2210.6 2589.2 2612.6 2882.1 2000 2200 2400 2600 2800 3000 ba bei sentence type production latency (ms) Different animacy Same animacy Figure 5.4: Mean production latency of perfect responses in the SA-ba vs. DA-ba conditions and the SA-bei vs. DA-bei conditions in Experiment 1 165 Thematic Prominence. Second, to test the effects of thematic prominence, planned comparisons were conducted on production latencies between (i) the SA-ba condition and the SA-bei condition, and (ii) the DA-ba condition and the DA-bei condition. Based on the prediction that ba sentences would be easier to produce than bei sentences with agent- theme verbs, participants would spend shorter time in the SA-ba and DA-ba conditions than in the SA-bei and DA-bei conditions. Paired samples t-tests revealed that the SA-ba condition (M=2.612) had a significantly shorter production latency than the SA-bei condition (M=2.882) by subjects (t1(28)=-2.287, p<.05; t2(28)=-1.823, p=.079). In addition, the DA-ba condition (M=2.210) had a significantly shorter production latency than the DA-bei condition (M=2.589) (t1(28)=-4.266, p<.001; t2(28)=-4.592, p<.001). Thus, effects of thematic prominence were confirmed by the production latencies of perfect responses. Animacy. Finally, an effect of animacy was also found by production latency. The difference in the mean production latency between the DA-ba and the DA-bei conditions was greater than that between the SA-ba and SA-bei conditions (379 ms vs. 270 ms). This result corroborates the evidence found by error rate, indicating that as well as thematic prominence, animacy also makes a noun more prominent. In sum, basis on the analysis on the production latencies of perfect responses, an inhibitory effect of semantic flexibility with agent-theme verbs was not found by perfect 166 responses in the no-prep. conditions, but was found in the ba and bei conditions. Moreover, effects of thematic prominence were confirmed by the shorter production latencies of perfect ba-sentences than those of perfect bei-sentences. As a whole, consistent with the results of correct responses reported in Chapter 2, the perfect-response rates and production latencies of perfect responses also provide evidence for my hypotheses about effects of semantic flexibility and thematic prominence, but with one exception: we did not find an inhibitory effect of semantic flexibility with agent- theme verbs in the no-prep. condition. This issue will be further discussed in 5.2.3. Next, let’s turn to the analysis of perfect responses in Experiment 2 with theme- experiencer verbs. 5.2.2 Experiment 2 Exclusions and replacements For the analyses reported here, the item analysis and subject analysis were conducted on the perfect-response rates and production latencies of perfect responses. In the item analysis, one item was excluded out of 18 due to programming error. In the subject analysis, one participant out of 30 was excluded due to a very low perfect-response rate, with only six observations of perfect responses out of 30 trials. A total of 4 empty cells out of 102 cells in the item analysis and 11 empty cells out of 174 cells in the subject 167 analysis were replaced by row and column means. Furthermore, for each condition, any production latency greater than 2.5 standard deviations above the mean was trimmed from the analysis, amounting to 11 responses being excluded (3%). Two sets of data are reported. The first set concerns the perfect-response rate in each condition. The second set concerns the mean production latency of perfect responses in each condition. Perfect-response rate The perfect-response rate in each condition is shown in Figure 5.5. The figure shows that the two factors ANIMACY and PREPOSITION CONSTRAINT were interacting. In the SA conditions, the perfect-response rate was highest when ba was present and lowest when there was no preposition. In contrast, in the DA conditions, the perfect-response rate was highest when bei is present and lowest when ba was present. ANIMACY appeared to have no effects because the SA-ba condition and the SA-bei condition had a higher perfect- response rate than their DA counterpart, but the DA no-preposition condition had a higher perfect-response rate than its SA counterpart. PREPOSITION CONSTRAINT appeared to have some effect: in general, the perfect-response rate was lower when there is no preposition than when ba or bei was present. ANOVAs demonstrated that the main effect of ANIMACY was not reliable (F1(1,28)=1.054, MSE=.077, p=.313; F2(1,16)=.871, MSE=.056, p=.364). The main effect of PREPOSITION CONSTRAINT was highly reliable (F1(2,56)=11.370, 168 MSE=.898, p<.001; F2(2,32)=14.917, MSE=.598, p<.001). The interaction between the two factors was also highly reliable (F1(2,56)=8.489, MSE=.663, p<.005; F2(2,32)=5.654, MSE=.385, p<.01). 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 no preposition ba bei Preposition constraint Perfect-response rate Different animacy Same animacy Figure 5.5: Mean perfect-response rate as a function of animacy and preposition constraint in Experiment 2 Additionally, to gain a better understanding of what differences were driving the interaction between ANIMACY and PREPOSITION CONSTRAINT, I took a closer look at the different conditions by means of planned comparisons. 169 Semantic Flexibility. To begin with, to test the effects of semantic flexibility, paired samples t-tests were conducted between (i) the SA no-prep. condition ([+SemFlex, +SynFlex]) and the DA no-prep. condition ([-SemFlex, +SynFlex]), and (ii) the SA-ba / SA-bei conditions ([+SemFlex, -SynFlex]) and the DA-ba/ DA-bei conditions ([-SemFlex, -SynFlex]). The prediction is the SA conditions would not differ from their DA counterparts in perfect-response rate based on the hypothesis that effects of semantic flexibility would not have negative effects with theme-experiencer verbs. Contrary to the prediction, the SA-no prep. condition had a significantly lower perfect- response rate (M=.456, SE=.058) than the DA-no prep. condition (M=.622, SE=.057) by subjects (t1(28)=2.450, p<.05; t2(16)=-1.54, p=.143). This seems to suggest that it is harder to produce a perfect response when the trial contains two animates and a theme- experiencer verb, a distinctive finding from the error rate result. In contrast, the SA-bei condition (M=0.789, SE=.047) and the DA-bei condition (M=.733, SE=.054) did not differ in perfect-response rate (t1(28)=.902, p=.375; t2(16)=.368, p=.718), providing evidence for the hypothesis that semantic flexibility has no effects with theme- experiencer verbs. However, the SA-ba condition (M=.878, SE=.038) had a significantly higher perfect-response rate than the DA-ba condition (M=0.611, SE=.056) (t1(28)=3.660, p<.01; t2(16)=2.749, p<.05), probably due to the less preferred inanimate THEME subject (e.g. “the movie BA the child scared a lot”) in the DA-ba condition. 170 Thematic Prominence. Second, to test the effects of thematic prominence, paired samples t-tests were conducted between (i) the SA-ba condition and the SA-bei condition, and (ii) the DA-ba condition and the DA-bei condition. The prediction is the bei conditions would have a higher perfect-response rate than the ba conditions based on the hypothesis that speakers would prefer to mention a more prominent thematic role earlier (i.e. EXPERIENCER first in Experiment 2), and thus bei sentences would be easier to produce than ba sentences. However, the perfect-response rate of the DA-bei condition was not significantly different from that of the DA-ba condition (t1(28)=1.491, p=.147; t2(16)=1.731, p=.103). In addition, no significant difference was found between the SA-ba and SA-bei conditions (t1(28)=1.126, p=.270; t2(16)=1.646, p=.119). Contrary to the error rates in the DA-ba and DA-bei conditions, which revealed effects of thematic prominence, the results here suggest that when a preposition is present with theme-experiencer verbs, it seems equally easy for the speaker to produce a perfect ba sentences and a perfect bei sentence. Animacy. Finally, an effect of animacy was also found. The difference in the perfect- response rate s between the DA-ba condition and the DA-bei condition was greater than that between the SA-ba and SA-bei conditions (12.2% vs. 8.9%). The result suggests that effects of animacy and thematic prominence are independent of each other. 171 In sum, based on the analysis of perfect-response rates, the hypothesis about the null effects of semantic flexibility with theme-experiencer verbs was not supported. Results showed that it was harder to produce a perfect response when the trial contained two animates and a theme-experiencer verb, a distinctive finding from the error rate result. Moreover, contrary to the results of error rates, the results here did not support the hypothesis about effects of thematic prominence, either. When a preposition was present with theme-experiencer verbs, it was equally easy for the speaker to produce a perfect ba sentences and a perfect bei sentence. Mean production latency In addition to perfect-response rates, I also examined the production latency data. The mean production latency of perfect responses in each condition is shown in Figure 5.6. The figure shows that the two factors ANIMACY and PREPOSITION CONSTRAINT did not interact. The patterns in the SA conditions and the DA conditions looked similar except that in the no-prep. conditions, ANIMACY appeared to have some effect such that the SA conditions had longer latencies than the DA conditions across the three preposition constraint conditions. PREPOSITION CONSTRAINT appeared to have no effects because the DA no-prep. condition had the shortest production latency of the three DA conditions, while the SA no-prep. condition had the longest production latency of the three SA conditions. It merits attention that in Experiment 2, in general, the SA 172 conditions had longer production latencies than the DA conditions, the opposite pattern of what was found by the correct responses. ANOVAs demonstrated that the main effect of ANIMACY was reliable by subjects (F1(1,28)=4.819, MSE=2.257, p<.05; F2(1,16)=1.023, MSE=.605, p=.327). The main effect of PREPOSITION CONSTRAINT was not reliable (F1(2,56)=1.001, MSE=.409, p=.374; F2(2,32)=.165, MSE=.071, p=.849). The interaction between the two factors was not reliable either (F1(2,56)=1.351, MSE=.514, p=.267; F2(2,32)=1.393, MSE=.246, p=.263). 2400 2600 2800 3000 no preposition ba bei Preposition constraint Mean production latency (ms) Different animacy Same animacy Figure 5.6: Mean production latency of perfect responses as a function of animacy and preposition constraint in Experiment 2 173 Additionally, to test the predictions concerning effects of semantic flexibility, thematic prominence and animacy, independent samples t-tests and paired samples t-tests were conducted on the production latency of different conditions. Semantic Flexibility. To begin with, to explore effects of semantic flexibility, an independent samples t-test was conducted between the SA-no prep. and DA-no prep. conditions. Additionally, paired samples t-tests were conducted between (i) the SA-ba and DA-ba conditions, and (ii) the SA-bei and DA-bei conditions. The hypothesis is that semantic flexibility would have no effects on production with theme-experiencer verbs. Thus, participants are predicted to spend equal production latencies in the SA and DA conditions. For the no-prep. conditions, participants’ perfect responses were grouped by sentence type and the mean production latency of ba sentences and bei sentences were compared respectively between these two conditions. The mean production latencies of the perfect responses depending on sentence type are shown in Figure 5.7. As predicted, the ba-sentences in the SA-no prep. condition had a numerically longer production latency (M=3.056, SE=.514) than those in the DA-no prep. condition (M=2.541, SE=.313), but this difference was not significant (t(13)=.881, p=.394). Similarly, the bei-sentences in the SA-no prep. condition had a numerically longer 174 production latency (M=2.827, SE=.231) than the DA-no prep. condition (M=2.402, SE=.115), but this difference was not significant either (t(76)=1.787, p=.078). Figure 5.7: Mean production latency of ba and bei sentences categorized as perfect responses in the SA-no prep. vs. the DA-no prep. conditions in Experiment 2 In addition, as shown in Figure 5.8, the SA-ba condition (M=2.623, SE=.154) did not differ significantly from the DA-ba condition (M=2.503, SE=.117) (t1(28)=.764, p=.451; t2(16)=.148, p=.884). Likewise, the SA-bei condition (M=2.648, SE=.164) did not differ significantly from the DA-bei condition (M=2.526, SE=.114) (t1(28)=.694, p=.493; t2(16)= .526, p=.606). 2 5 1 4 . 6 2 5 1 4 . 6 2 5 1 4 . 6 2 5 1 4 . 6 2 4 0 2 . 4 2 4 0 2 . 4 2 4 0 2 . 4 2 4 0 2 . 4 3 0 5 6 . 9 3 0 5 6 . 9 3 0 5 6 . 9 3 0 5 6 . 9 2 8 2 7 . 4 2 8 2 7 . 4 2 8 2 7 . 4 2 8 2 7 . 4 2200 2400 2600 2800 3000 3200 ba bei sentence type production latency (ms) Different animacy Same animacy 175 2 5 0 3 . 9 2 5 0 3 . 9 2 5 0 3 . 9 2 5 0 3 . 9 2 5 2 6 . 6 2 5 2 6 . 6 2 5 2 6 . 6 2 5 2 6 . 6 2 6 2 3 . 6 2 6 2 3 . 6 2 6 2 3 . 6 2 6 2 3 . 6 2 6 4 8 . 5 2 6 4 8 . 5 2 6 4 8 . 5 2 6 4 8 . 5 2000 2200 2400 2600 2800 ba bei sentence type production latency (ms) Different animacy Same animacy Figure 5.8: Mean production latency of perfect responses in the SA-ba vs. DA-ba conditions and the SA-bei vs. DA-bei conditions in Experiment 2 As a whole, despite that participants spent longer time in SA conditions than in the DA conditions, the statistical results of perfect responses are consistent with those of correct responses in support of the hypothesis that semantic flexibility does not affect production with theme-experiencer verbs. Thematic Prominence and Animacy. Second, to test the effects of thematic prominence, paired samples t-tests were conducted between (i) the SA-ba condition and the SA-bei condition, and (ii) the DA-ba condition and the DA-bei condition. Based on the prediction that bei sentences would be easier to produce than ba sentences with theme- 176 experiencer verbs, participants are expected to spend shorter times in the bei conditions than in the ba conditions. However, no significance was found between the mean production latency of the SA-ba condition (M=2.623) and the SA-bei condition (M=2.648) (t1(28)=-.167, p=.868; t2(16)=-.002, p=.998). In addition, no significance was found between the mean production latency of the DA-ba condition (M=2.503) and the DA-bei condition (M=2.526) (t1(28)=-.161, p=.873; t2(16)=-.316, p=.756). In other words, the production latencies of perfect responses did not show effects of thematic prominence, consistent with the findings of correct responses. Moreover, an animacy effect was not salient. The difference between the DA-ba and DA-bei conditions was very close to the difference between the SA-ba and SA-bei conditions (23ms vs. 25ms). In sum, based on the analysis of production latencies, the results of perfect responses are consistent with the hypothesis that semantic flexibility does not affect production with theme-experiencer verbs. However, just like the results from correct responses, effects of thematic prominence were not found here. As a whole, based on the analysis on the perfect responses in Experiment 2, the results of perfect-response rates differed from those of error rates in two ways. First, against our hypothesis, an inhibitory effect of semantic flexibility was found with theme-experiencer verbs in perfect-response rates. Second, unlike error rates, perfect-response rates failed to reveal effects of thematic prominence. These findings are somewhat unexpected and will be discussed in 5.2.3. In contrast, analysis of production latencies showed congruent 177 findings in both correct and perfect responses. Production latencies in both cases did not find effects of semantic flexibility and thematic prominence with theme-experiencer verbs. Possible accounts for the lack of findings have been addressed in Section 3.3 and will not be repeated here. Next, I will discuss the incongruent findings of correct and perfect responses in Experiments 1 and 2, and provide possible accounts for these discrepancies. 5.2.3 Comparing Correct Responses and Perfect Responses in Experiments 1 and 2 Summarizing the findings so far, the results based on perfect-response rates and production latencies of perfect responses mostly agree with what we have found from correct responses, except for one discrepancy in Experiment 1 and two discrepancies in Experiment 2, as shown in Tables 5.3 and 5.4 and indicated by grey shading. The ‘yes’ and ‘no’ in the SemFlex column indicate whether effects of semantic flexibility were found to be significant. 178 Table 5.3: Comparison between correct responses and perfect responses of Experiment 1 Experiment 1 condition Dependent variable SemFlex ThemProm No-prep. error rate yes RT yes correct responses ba error rate yes error rate: yes in DA, no in SA RT yes RT: yes in both DA and SA bei error rate no RT yes No-prep. perfect-response rate yes RT no perfect responses ba perfect-response rate yes perfect rate: yes in DA, no in SA RT yes RT: yes in both DA and SA perfect-response rate no bei RT yes Table 5.4: Comparison between correct responses and perfect responses of Experiment 2 Experiment 2 condition Dependent variable SemFlex ThemProm No-prep. error rate no RT no correct responses ba error rate yes error rate: yes in DA, no in SA RT no RT: no in both DA and SA bei error rate yes RT no No-prep. perfect-response rate yes RT no perfect responses ba perfect-response rate yes perfect rate: no in DA and SA RT no RT: no in both DA and SA perfect-response rate yes bei RT no First, in Experiment 1 where the target items contained agent-theme verbs, the only discrepancy between correct and perfect responses was shown by production latency: When a preposition was not present, semantic flexibility appeared to have no effects on the production latency of perfect responses, whereas it had an inhibitory effect on the production latency of correct responses. In the case of perfect responses, participants spent equally time producing perfect sentences whether the nouns being both animate or 179 differed in animacy, whereas in the case of correct responses, participants spent longer time producing correct sentences when the nouns were both animate. Second, in Experiment 2 where the target items contained theme-experiencer verbs, two discrepancies resulted from perfect-response rates. In the no-prep. conditions, semantic flexibility appeared to have an inhibitory effect on the production of perfect responses: Participants produced less perfect responses in the SA-no prep. condition than in the DA- no prep. condition, whereas with correct responses, the probability to produce a correct response was equal in the SA-no prep. and DA-no prep. conditions. Moreover, in the ba and bei conditions, the proportion of perfect responses was not influenced by thematic prominence effects: Participants produced perfect ba-sentences as often as bei-sentences when either preposition was given. This contrasts with the proportion of correct responses, which was sensitive to thematic prominence, showing a pattern where correct responses occurred more often in the bei condition (i.e., EXPERIENCER first) than in the ba condition (i.e., THEME first). In general, these discrepancies occurred more often in the no-prep. conditions, because the perfect responses in these conditions comprise a smaller subset of correct responses than in the ba or bei conditions. Conceivably, a smaller subset may not be as representative of the whole set compared with a larger subset. Moreover, it seems that for perfect responses, effects of semantic flexibility are no longer conditioned by verb type in the predicted way, particularly in the no-prep. condition. For example, in Experiment 1, 180 the production latency data of perfect responses show that semantic flexibility no longer hindered production with agent-theme verbs. Additionally, in Experiment 2, the perfect- response rate shows that semantic flexibility increased difficulty on production with theme-experiencer verbs. Moreover, effects of thematic prominence seemed to be weakened in the perfect responses, particularly in the ba and bei conditions, where a preposition was given. These observations led me to contemplate on the correlation between the behaviors of the production mechanism and the nature of perfect responses and correct responses. As argued earlier, a perfect response is likely to occur in two scenarios: The first scenario depicts a highly incremental process, characteristic of a relatively short mean production latency compared with correct responses. The second scenario depicts a less incremental process, characteristic of a relatively long mean production latency compared with correct responses. Bearing this in mind, let’s turn to the discrepancy found between perfect responses and correct responses in Experiment 1. As mentioned, we found an inhibitory effect of semantic flexibility on trials where people gave correct responses, but not on trials were people produced perfect responses. Table 5.5 compares the mean production latencies of the correct responses and perfect responses in the SA-no prep. and DA-no prep. conditions depending on sentence type (i.e. ba or bei sentences). In general, the mean production latencies of perfect responses were shorter than those of correct responses in 181 all conditions, in particular the perfect ba sentences produced in the SA no-prep. conditions, with a decrease from 2847 ms to 2525 ms. This suggests that these perfect responses were probably produced in the first scenario, where production is easy and carried out in a highly incremental fashion. Importantly, in such circumstances, the inhibitory effect of semantic flexibility associated with agent-theme verbs was somewhat weakened. Table 5.5: Mean production latencies of correct responses and perfect responses in the SA no-prep. and DA no-prep. conditions in Experiment 1 Produced sentence type Condition Response type Correct responses Perfect responses BA SA no-prep. [+SemFlex, +SynFlex] 2847 (ms) 2525 (ms) DA no-prep. [-SemFlex, +SynFlex] 2493 (ms) 2469 (ms) ( p<.05) (p=.775) BEI SA no-prep. [+SemFlex, +SynFlex] 2881 (ms) 2822 (ms) DA no-prep. [-SemFlex, +SynFlex] 2418 (ms) 2335 (ms) (p=.067) (p=.105) Now, let’s turn to the discrepancies in Experiment 2. In the no-prep. conditions of Experiment 2, participants produced less perfect responses in the SA-no prep. condition than in the DA-no prep. condition. This seems unexpected according to the hypothesis that effects of semantic flexibility do not hinder production with theme-experiencer verbs. However, as we saw in Chapter 3, the error rates of the SA-no prep. and DA-no prep. `conditions did not differ significantly. Thus, there being fewer perfect responses in the SA-no prep. condition may not be influenced by effects of semantic flexibility, but due to 182 the high proportion of correct but unexpected responses (21.1%), as shown in Table 5.2. This is because an NP-conjoined sentence containing two animates and a theme- experiencer verb is plausible and grammatical in Mandarin (e.g., Mingdaoyan he numingxing xia-le-yi-tiao “the director and the actress were scared”), but coded as a correct but unexpected response in my analysis. In contrast, the DA-no preposition condition had a much lower proportion of correct but unexpected responses (2.8%), because it is not plausible to say an NP-conjoined sentence containing an animate- inanimate noun pair (e.g., # Mingdaoyan he dianyin xia-le-yi-tiao “#The director and the movie were scared.”) Finally, the second discrepancy in Experiment 2 is the fact that effects of thematic prominence were weakened in the ba and bei conditions: Participants produced perfect ba-sentences as often as bei-sentences when ba or bei was given. The insignificance between the perfect-response rates of the SA-ba and SA-bei conditions is very likely to be a consequence of the ceiling effects existing in these two conditions, as the error rates for these two conditions did not differ significantly, either. However, the error rate was marginally significantly higher in the DA-ba condition than in the DA-bei condition by subjects (p=.056), but the perfect-response rates of these two conditions did not differ significantly. Moreover, the production latency data of perfect responses also failed to show any effects of thematic prominence. In other words, it seems reasonable to claim that effects of thematic prominence were less salient when these perfect responses were 183 produced. As shown in Table 5.6, in the DA-ba and DA-bei conditions, the mean production latencies of perfect responses were shorter than those of correct responses. Table 5.6: Mean production latencies of correct responses and perfect responses in the DA-ba and DA-bei conditions in Experiment 2 Condition Response type Correct responses Perfect responses DA-ba 2861(ms) 2503(ms) DA-bei 2794(ms) 2526(ms) Again, this suggests that these ‘fast’ perfect responses were probably produced in the first scenario where speech production is assumed to be highly incremental. In such a scenario, just as the inhibitory effect of semantic flexibility with agent-theme verbs was weakened, the thematic prominence effect which favors one structure over the other became less salient . So far, the observations point to one speculation: When the production process is executed in a highly incremental way, the effects of semantic flexibility and thematic prominence may lose their impacts on individual components of the process. Next, I turn to analyze the perfect responses of the English experiment and examine whether those results are compatible with this speculation. 184 5.3 Perfect Responses of Experiment 3 (English) Following the Mandarin experiments, a correct response in the English experiment must be a well-formed plausible sentence containing every given word. Sample displays of the trials are given in (2) (i.e., the active condition) and (3) (i.e., the passive condition). In the active condition, only active sentences were considered as a correct response. Likewise, in the passive condition, only passive sentences were considered as a correct response. When a correct response was accompanied without speech disfluencies, it was subcategorized as a perfect response; otherwise, it was subcategorized as a correct but imperfect response. The count of each subtype of correct responses depending on verb type is shown in Table 5.7. As can bee seen, among the three verb types, the perfect responses of experiencer-theme verbs comprise the smallest subset of correct responses. (2) Sample display of the active condition 185 (3) Sample display of the passive condition Table 5.7: Count of each subtype of correct responses depending on verb type in Experiment 3 Subtype of correct responses Agent-theme verbs Experiencer-theme verbs Theme-experiencer verbs Perfect response 132 (69.8%) 116 (61.4%) 123 (70.3%) Correct but imperfect response 57 (30.2%) 73 (38.6%) 52 (29.7%) Total no. of correct responses 189 (100%) 189 (100%) 175 (100%) The primary goal of the analysis on perfect responses of Experiment 3 was to further explore how the effects of semantic flexibility and thematic prominence correlate with the scenarios where speakers produce perfect responses. Importantly, analyzing the perfect responses in Experiment 3 offers an opportunity to verify the speculation cross- 186 linguistically by exploring whether systematic variability between correct and perfect responses also exists in another language. 5.3.1 Results Exclusions and replacements For the analyses reported here, the item analysis and subject analysis were conducted on the perfect-response rates and production latencies of perfect responses. In the subject analysis, one participant out of 16 was excluded due to a very low perfect-response rate, with only 5 tokens of perfect responses out of 36 trials. In addition, two empty cells out of 144 cells in the item analysis were replaced with row and column means. Two sets of data are reported here as an indication of the difficulty level of a condition. The first set concerns the perfect-response rate of each verb type. The second set concerns the mean production latency of perfect responses of each verb type. Perfect-response Rate of Each V erb Type To test whether the effects of semantic flexibility and thematic prominence are sensitive to verb type, paired samples t-tests and independent samples t-tests were conducted on the three verb types in the subject and item analyses, respectively. Contrary to the error rate, a higher perfect-response rate indicates a lower difficulty level of production. Following the rationale in Experiment 3, the perfect-response rate and production latency 187 are assumed to show the overall additive effects of semantic flexibility and thematic prominence. The hypotheses and predictions of Experiment 3 are repeated here as a reminder: First, effects of semantic flexibility would hinder the production with agent- theme verbs but have no effects on experiencer-theme verbs and theme-experiencer verbs. Second, effects of thematic prominence are hypothesized to be sensitive to the interaction of thematic structure and sentence type. Thus, sentences with agent-theme verbs and experiencer-theme verbs would prefer the active voice while sentences with theme- experiencer verbs would prefer the passive voice. The predictions are: (i) For active voice, the perfect-response rate of experiencer-theme verbs would be highest, followed by that of agent-theme verbs and theme-experiencer verbs. (ii) For passive voice, the perfect- response rate of theme-experiencer verbs would be highest, followed by experiencer- theme verbs and agent-theme verbs. Additionally, taking the production of experiencer- theme verbs as the baseline for comparison, effects of semantic flexibility predict that experiencer-theme verbs would have a higher perfect-response rate than agent-theme verbs. In contrast, effects of thematic prominence predict that for the active voice, experiencer-theme verbs would have a higher perfect-response rate than theme- experiencer verbs, and the pattern is reverse for the passive voice. The perfect-response rates in the active and passive conditions depending on verb type are shown in Figure 5.9. 188 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 active voice passive voice Voice Perfect rate agent-theme exp.-theme theme-exp. Figure 5.9: Mean perfect-response rate depending on verb type in the active and passive conditions in Experiment 3 Active V oice. For the active voice, experiencer-theme verbs had the highest perfect- response rate (M=.802, SE=.032), followed by agent-theme verbs (M=.739, SE=.032) and theme-experiencer verbs (M=.572, SE=.069). This pattern conforms to the first prediction. Statistically, the perfect-response rate of experiencer-theme verbs was not significantly higher than that of agent-theme verbs (t1(14)=.725, p=.481; t2(22)=1.360, p=.188), which means effects of semantic flexibility was not found. In contrast, the perfect-response rate of experiencer-theme verbs was significantly higher than that of theme-experiencer verbs (t1(14)=5.602, p<.001; t2(22)=2.984, p<.05). Additionally, the perfect-response rate of agent-theme verbs was also significantly higher than that of 189 theme-experiencer verbs by item (t1(14)=1.086, p=.296; t2(22)=2.170, p<.05). These two differences verify effects of thematic prominence. As a whole, the results of perfect responses and correct responses corroborate the same conclusions regarding the production of active voice. Passive V oice. For the passive voice, as the second prediction expects, theme-experiencer verbs had the highest perfect-response rate. The perfect-response rate of theme- experiencer verbs (M=.708, SE=.047) was significantly higher than that of experiencer- theme verbs (M=.406, SE=.069) (t1(14)=5.511, p<.001; t2(22)=3.617, p<.005), again verifying an effect of thematic prominence. It is worth mentioning that contrary to the prediction, the perfect-response rate of agent-theme verbs was significantly higher than that of experiencer-theme verbs (t1(14)=2.662, p<.05; t2(22)=3.067, p<.05), suggesting that semantic flexibility has a facilitative effect on sentences with agent-theme verbs. However, this result was not surprising. In the error analysis presented in Chapter 4, I have pointed out that experiencer-theme verbs had the fewest counts of perfect response but most counts of correction and hesitation (cf. Table 4.5) and provided an explanation therein. Notably, while the error rates for the passive condition exhibited ceiling effects, the perfect-response rate for the passive voice was able to reveal effects of thematic prominence. 190 In sum, based on the anlaysis of perfect-response rates, effects of semantic flexibility were not found in both active and passive conditions. In contrast, effects of thematic prominence were found in both conditions. Mean production latency In addition to perfect-response rates, the production latencies of perfect responses were also analyzed. Following the hypotheses mentioned above, the predictions are: (i) For the active voice, the production latency would be shortest for experiencer-theme verbs, followed by agent-theme and theme-experiencer verbs. (ii) For the passive voice, the production latency would be shortest for theme-experiencer verbs, followed by experiencer-theme verbs and agent-theme verbs. Additionally,, taking experiencer-theme verbs as the baseline for comparison, effects of semantic flexibility predict that experiencer-theme verbs would have a shorter production latency than agent-theme verbs. Moreover, effects of thematic prominence predict that for the active voice, experiencer- theme verbs would have a shorter production latency than theme-experiencer verbs, and vice versa for the passive voice. The mean production latencies in the active and passive conditions depending on verb type are shown in Figure 5.10. 191 0 400 800 1200 1600 2000 active voice passive voice Mean production latency (ms) agent-theme exp.-theme theme-exp. Figure 5.10: Mean production latency of perfect responses depending on verb type in the active and passive conditions in Experiment 3 Active V oice. For the active voice, the three verb types did not differ from one another. The mean production latency of theme-experiencer verbs (M=1.725, SE=.090) was not different from that of experiencer-theme verbs (M=1.721, SE=.068)(t1(14)=.013, p=.990; t2(22)=.036, p=.972) or agent-theme verbs (M=1.851, SE=.073) (t1(14)=-1.187, p=.255; t2(22)=-1.085, p=.290). Moreover, although the mean production latency of agent-theme was numerically longer than that of experiencer-theme verbs, they did not differ significantly (t1(14)=1.840, p=.087; t2(22)=1.301, p=.207). That is, neither effects of semantic flexibility nor effects of thematic prominence were found. Just as the correct responses, the production latency data of the perfect responses may also exhibit ceiling effects, probably due to the simple SVO structure of active voice. As Figure 5.10 shows, 192 the mean production latencies of perfect responses across the three verb types are short (i.e. around 1800 ms), almost identical to the production latencies of correct responses. Passive V oice. For the passive voice, although the tendency was consistent with the prediction: theme-experiencer verbs had the shortest production latency while agent- theme verbs had the longest production latency, statistically the three verb types did not differ from one another. The mean production latency of theme-experiencer verbs (M=1.898, SE=.089) was not different from that of experiencer-theme verbs (M=2.037, SE=.122)(t1(14)=-.982, p=.343; t2(22)=.-.919, p=.368) or agent-theme verbs (M=2.054, SE=.121) (t1(14)=-1.006, p=.332; t2(22)=-1.029, p=.315). Additionally, the production latencies for agent-theme and experiencer-theme verbs did not differ significantly, either (t1(14)=.555, p=.588; t2(22)=..095, p=.925). As a result, neither effects of semantic flexibility nor effects of thematic prominence were found in the passive condition. Notably, the production latency of correct responses revealed significant differences between experiencer-theme verbs and theme-experiencer verbs for the passive voice. However, perfect responses failed to do so. In sum, based on the analysis of the production latencies of pefect responses, effects of semantic flexibility or thematic prominence were not found, either in the active or passive condition. 193 As a whole, perfect-response rates showed effects of thematic prominence for the active condition, whereas error rates exhibited ceiling effects for the active condition. In contrast, the production latencies of correct responses revealed effects of thematic prominence for the passive condition, wheareas the production latencies of perfect responses failed to do so. These two discrepancies will be discussed further in 5.3.2. 5.3.2 Comparing Correct Responses and Perfect Responses in Experiment 3 As just mentioned, the results of perfect responses in Experiment 3 differed from the results of correct responses in two important ways. First, for the active voice, though the error rates did not reveal any effects among the three verb types, the perfect-response rates revealed effects of thematic prominence between experiencer-theme and theme- experiencer verbs. Second, for the passive voice, the production latencies of correct responses revealed effects of thematic prominence, while the production latencies of perfect responses failed to do so. These discrepancies are shown in Table 5.8, as indicated by grey shading. 194 Table 5.8: Comparison between correct responses and perfect responses of Experiment 3 Experiment 3 condition Dependent variable SemFlex ThemProm active error rate no no correct responses RT no no passive error rate no no (ceiling effect) RT no yes active perfect-response rate no yes perfect responses RT no no passive perfect-response rate no no RT no no Let us first consider the discrepancy in the active condition. When we look at the proportion of trials on which participants produced correct responses, we see no effects of thematic prominence, but when we restrict our analysis to only those trials where participants produced perfect responses, we find significant effects of thematic prominence. This may be attributed to the more stringent criterion for a perfect response. Overall, Experiment 3 had a low error rate (4.2 %). In other words, 95.8% of the total responses were ‘correct’. In contrast, only 64.4% of the total responses were considered ‘perfect’. This may account for why the perfect-response rate is more likely to reveal differences among verbs. Now let us turn to the second discrepancy in the passive condition: The production latencies of correct responses seemed to be more ‘powerful’ than those of perfect responses because the former were able to show effects of thematic prominence for the passive voice. In response, I suggest that the proportion of perfect responses of each verb type is closely related to the second discrepancy. As Table 5.3 shows, experiencer-theme 195 verbs have the smallest proportion of perfect responses. If we restricted our analysis on perfect responses after discarding the data of correct but imperfect responses, the mean production latency of experiencer-theme verbs decreased from 2295 ms to 2037 ms. Just like the perfect responses in Experiments 1 and 2, the perfect responses in Experiment 3 might also belong to the scenario where production is fast and easy, for they resulted in a shorter mean production latency compared with correct responses. Therefore, it seems reasonable to claim that the production latencies of perfect responses failed to reveal differences among verb types due to a highly incremental process. 5.4 Theoretical Implications The results presented in this chapter so far imply that effects of semantic flexibility and thematic prominence may be conditioned by how incremental the production process is. Under the assumption that the production of correct responses resembles the production of natural speech because speech disfluencies are included, effects of semantic flexibility seem to be sensitive to the interaction of thematic roles and animacy, and effects of thematic prominence can influence speakers’ syntactic choice between alternative structures. However, closer examinations of the production of perfect responses in the three experiments show a different picture: effects of semantic flexibility and thematic prominence become less salient. In other words, how language production proceeds may be adjusted in response to variation in linguistic settings or contexts. Importantly, recent studies and theoretical accounts within the language production framework, in particular 196 those concerning how extra-linguistic factors such as speakers’ task sensitivity and strategic control affect the process of message formulation and grammatical encoding, can serve as the stepping stones for understanding why effects of semantic flexibility and thematic prominence would potentially be influenced by the incremental nature of speech production. To begin with, effects of semantic flexibility are related to the referent processing in the conceptual or message formulation level, which involves assigning thematic roles to an animate referent in a mental model. To investigate under what circumstances thematic role and animacy tend to interact, Van Nice & Dietrich (2003) conducted three parallel German picture-description tasks: writing from memory, speaking with pictures in view, and speaking from memory by manipulating the factors of thematic role (agent/patient) and voice (active/passive). The first two are expected to involve ‘sequential processing’, while the third task should favor what they call ‘compact processing’. By comparing the mean production latencies of the passive sentences in these experiments, Van Nice & Dietrich concluded that the interaction of thematic roles and animacy only occurs when speakers overlap referent processing or ‘cram in’ more bits of information during speech planning. The reason why speakers opt for compact processing is because activating more bits of information in a mental model can ‘speed’ the time course of planning. In contrast, an interaction of role and animacy does not exist in more sequential processing such as tasks involving writing from memory and speaking with pictures in view, where pre- utterance planning is not demanded. In these tasks, speech planning and utterances may 197 be executed parallely. Additional support for compact processing comes from eye- tracking data (Griffin, 1998; Griffin & Bock, 2000). ‘‘Planned’’ picture descriptions (those uttered after the picture disappears from view) involve a more compact processing of referents than do ‘‘spontaneous’’ descriptions of pictures that remain in view. In planned utterances, attention shifts from the first to the second referent picture at a very minimal lag, suggesting a partially simultaneous (‘‘overlapped’’) processing of successive referents. Following this rationale, effects of semantic flexibility would be obscured in a highly incremental process since thematic roles and animacy do not interact. Turning to the grammatical component, effects of thematic prominence are related to constituent assembly in grammatical encoding. The tendency to mention a more prominent thematic role earlier would render the less preferred structure more difficult to produce. The thematic prominence effects in sentence production are argued to provide evidence for the views of incremental models, which assume the first conceptualized referent to be first lexicalized and to appear first in word order (Bates & MacWhinney, 1982; Bock, 1982; Tomasello, 2000; also see Garnsey, Pearlmutter, Myers, & Lotocky, 1997). Of great relevance to the present issue, F. Ferreira & Swets (2002) proposed that the degree to which the production is incremental is under strategic control. In their studies, participants were asked to calculate the answers to arithmetic problems (e.g. 53+5) and 198 answered in the format “The answer is …”. In one experiment, they were allowed to answer the question when they felt ready. The data did not support incremental production: the initiation times were longer for more difficult problems, while the initiation durations were unaffected, indicating the utterances were fully planned before articulation. In another experiment, participants were required to start speaking before a deadline (beep). Interestingly, accuracy was not sacrificed though initiation times were largely reduced. The data showed that initiation times were still affected by problem difficulty, so was the duration of the earlier part of an utterance, suggesting that participants postponed some planning until they were actually speaking. Thus, F. Ferreira & Swets concluded that the production system becomes more incremental if a premium is placed on beginning to speak quickly, but speakers seem to prefer to plan if they have the opportunity to do so. As we already saw in this chapter, in Experiment 1, effects of semantic flexibility were found from correct responses but not from perfect responses in the no-preposition conditions. In addition, in Experiment 3, effects of thematic prominence were found from correct responses but not from perfect responses in the passive condition. The assumption is that perfect responses in these experiments were produced more incrementally than correct responses because on average the former were produced with relatively short production latencies. Following these threads of thought, let’s imagine what might happen when a response is being produced at fast speed. The speaker would tend to initiate an utterance as soon as one piece of syntactic information (i.e. the subject) 199 becomes available. On a trial where both nouns are animate and can take either thematic role, the first retrieved noun, most likely the noun that is read first, would grab the subject position right away, even before it is assigned a thematic role. In the same vein as the proposal of Van Nice & Dietrich (2003), there would not be interactions between thematic roles and animacy in a sequential planning. As a result, effects of semantic flexibility become insignificant. Moreover, with a short deadline, the most efficient way to produce a sentence is to dispatch the first retrieved word right away by inserting it into the subject position, and then build on the rest of the sentence on a specified structure. In other words, both message planning and linguistic planning are still in process while speaking. In this way, effects of thematic prominence would not play a role. An important thing to keep in mind: although in general the perfect responses in my experiments were suggested to be produced in fast speech, a perfect response can also occur in prepared speech. I predict effects of semantic flexibility and thematic prominence would both exist in prepared speech. In prepared speech, the production process is assumed to be less incremental. Thus, referent processing would be more compact, allowing thematic roles and animacy to interact with each other. Consequently, inhibitory effects of semantic flexibility are expected to be found when the interaction of thematic roles and animacy pose certain difficulty for figuring out a relational meaning. In addition, if speakers are allowed to speak when they are prepared, they would have 200 enough time to plan their sentence. To maximize accuracy and fluency, a possible strategy is to generate the specified structure first, and then map a message onto the linguistic structure. However, if the specified structure is the less preferred one (e.g., THEME was criticized by the actress AGENT”), it would be harder to retrieve the sentence-initial word, resulting in effects of thematic prominence. To summarize, the observed discrepancies between the results of correct responses and perfect responses have led to the hypothesis that effects of semantic flexibility and thematic prominence, or more broadly, the processes of message formulation and grammatical encoding, may be conditioned by extra-linguistic factors such as speakers’ strategic control, because speakers’ strategy may influence the incrementality of speech production. I further hypothesize that effects of semantic flexibility and thematic prominence would vary for fast speech and prepared speech, Without doubt, whether or not a response is produced highly incrementally is an empirical issue. In order to further explore the correlation between strategic control and processes of sentence production, two follow-up experiments were designed by manipulating the strategic control factor, and will be presented in Chapter 6. 201 Chapter 6: English Production Tasks Investigating Strategic Control 6.1 Introduction The main aim of Experiments 4 and 5 is to investigate how the extra-linguistic factor of speakers’ strategic control influences the process of sentence production. A large numbers of experimental work have provided evidence for the incrementality hypothesis of sentence production. As Konopka (2012:144) pointed out, while it is desirable to maintain both fast and fluent production (Clark, 2002), adult speakers often strategically choose to begin communicating quickly and plan subsequent sentence chunks during production itself ( e.g., Bock, Irwin, Davidon & Levelt, 2003; Clark & Wasow, 1998; Griffin, 2003; Levelt & Meyer, 2000; Martin, Crowther, Knight, Tamborello & Yang, 2010). Although most accounts of language production assume that message planning and linguistic planning proceed incrementally, that is, speakers plan an utterance at different levels while speaking at the same time, the scope of linguistic planning (i.e., the amount of linguistic information that speakers prepare in advance) appears to be highly variable. In some studies, radical incrementality is found such that utterances can be initiated as soon as a sub-phrasal unit such as a phonological word is available (e.g., Allum and Wheeldon, 2009; Brysbaert, Fias & Noel, 1998;Griffin, 2001; Griffin & Spieler, 2006; Meyer, Roelofs, & Levelt, 2003; Wheeldon & Lahiri, 1997). In other studies, speakers are more cautious and do not speak until they have planned a reasonably large chunk of an upcoming utterance, such as a clausal or phrasal unit (e.g., Allum & Wheeldon, 2007; Lindsley, 1975; Meyer, 1996, 1997; Smith & Wheeldon, 1999, 2004; Stallings, MacDonald & O’Seaghdha, 1998; Martin, Crowther, Knight, Tamborello & 202 Yang, 2010; Wagner, Jescheniak & Schriefers, 2010). Due to the variability of advance planning scope, Bock et al. (2003) have proposed that there are two types of incrementality, lexical incrementality and structural incrementality, under the assumption that either the structural process or the lexical process can guide early sentence planning. In lexically incremental processes, words may play a role in selecting the necessary structural frame. A speaker might retrieve the word corresponding to one message element at the outset of sentence formulation and continue building the sentence word by word from that point (e.g., Bates & MacWhinney, 1982; Bock, 1982; Garnsey, Pearlmutter, Myers, & Lotocky, 1997; Gleitman, January, Nappa & Trueswell, 2007). In contrast, in structurally incremental processes, sentence formulation begins with the generation of a sentence plan that captures the relationships between various message elements. Structural information may control the timing of lexical retrieval, with words becoming active when a structural frame calls for them (e.g., Bock, 1990; Bock, Irwin, Davidon & Levelt. 2003; Christianson & F. Ferreira, 2005; Dell, 1986; V. Ferreira & Dell, 2000; Fisher, 2002; Konopka & Bock, 2009; Kuchinsky & Bock, 2010). Since there is little consensus as to the size of advance planning scope, the most ‘parsimonious’ conclusion is that production can be incremental to different degrees in different contexts. Existing experimental work attributed the adjustments of planning scope to a variety of pre-linguistic, linguistic and extra-linguistic factors. For example, it has been proposed that speakers may narrow the planning scope under time pressure, such as when responding before a deadline (Ferreira & Swets, 2002, 2005), or when the 203 condition has increased cognitive load (Wagner, Jescheniak & Schriefers, 2010). Individual differences such as working memory spans also influence the planning scope (Swets, Jacovina, & Gerrig, 2008). Even during the course of a single experimental session, adjustments in planning scope can occur as speakers develop strategies to perform the task efficiently (Schriefers & Teruel, 1999). Among these factors which can potentially affect the scope of advance message and linguistic planning, the strategic control factor is of great relevance to the present study. Broadly speaking, the effects of ‘strategic control’ factors have generated extensive discussions in a research exploring a wide range of human cognitive activities, including information processing (e.g., Bartholow et. al., 2005; Gratton, Coles & Donchin, 1992; Meiran, 1996), word recognition (e.g., Baluch & Besner, 1991; Jared, 1997; Kello & Plaut, 2000, 2003), memory performance (e.g., Gronlund & Shiffrin, 1986; Goldsmith, Koriat & Pansky, 2005; Goldsmith, Koriat & Weinberg-Eliezer, 2002; Reder, 1982, 1987; Reder & Ross, 1983). reading comprehension (e.g., Boulware-Gooden, Carreker, Thornhill & Malateshe, 2007; Jarrod et al., 2011; Magliano, Trabasso & Graesser, 1999), and sentence production (e.g., F. Ferreira & Swets, 2002, 2005; Oomen & Postma, 2001; van Nice & Dietrich, 2003; Wagner, Jescheniak & Schriefers, 2010). The present study manipulated speakers’ strategic control in an attempt to condition the degree of incremenality during sentence production by causing variability in the advance planning scope. In this work, speakers’ strategic control refers to the extra-linguistic 204 factor that involves explicit control mechanisms, which might be invoked by the instructions given to a participant. Consider a situation where the speaker is under time pressure and aims at speed against another situation where the speaker is told to prepare his/her speech as accuracy is crucial. Under the assumption that advance speech planning is flexible and time pressure tends to narrow the planning scope, sentence production is expected to proceed more incrementally with a narrower planning scope in the speed- oriented setting than in the accuracy-oriented setting. In other words, the speakers would presumably develop different strategies in response to the instruction they receive in an experiment. The primary goal is to investigate whether or not the processes of message formulation and grammatical encoding are susceptible to speakers’ strategic control within the framework of the incremental theory. An important question is: do the processes of message formulation and grammatical encoding have different behaviors under speakers’ strategy control, since they involve individual components in the production system? The findings of Experiments 1 and 2 showed that effects of semantic flexibility and syntactic flexibility are not uniform, suggesting that the two components may involve different processes. Consequently, further examining the effects of semantic flexibility and thematic prominence allows us to pursue this theoretical inquiry because it offers a unique window for looking into the processes that take place in these components. The current work may offer insight into the operation of the production system by indicating to what extent the production system is automatic vs. controllable. 205 The research questions are two-fold. First, existing studies have proposed that speakers’ strategic control might adjust the scope of linguistic planning such that grammatical encoding may proceed more or less incrementally. Since effects of thematic prominence involve the grammatical component, are they sensitive to different degrees of incrementality in grammatical encoding? Second, assuming that the message component and grammatical component may not be operated in the same way, can speakers’ strategic control affect the message component as well? To investigate these questions, I conducted two experiments designed to create an accuracy-oriented and a speed-oriented setting, respectively. Experiments 4 and 5 were built on the experimental design and materials of Experiment 3 with some modifications in the wording of the instructions and the screen display. In Experiment 4, participants were allowed to read the words at their own pace and were told to prepare their speech before initiating an utterance. In contrast, in Experiment 5, participants were instructed to produce a sentence before a three-second progress bar filled up. Following the rationale of Experiment 3, I examined effects of semantic flexibility and thematic prominence by comparing the production of three verb types— agent-theme verbs, theme-experiencer verbs, and experiencer-theme verbs. Notably, as we saw in Chapter 4, the findings regarding effects of semantic flexibility are not congruent in the Mandarin and English experiments. Unlike the Mandarin experiments (ie., Experiments 1 and 2), which demonstrated that effects of semantic flexibility were sensitive to verb type in Mandarin Chinese, the results of the English experiment (i.e., Experiment 3) did not reveal effects of semantic flexibility in English. 206 As argued earlier, two possible reasons might account for the absence of semantic flexibility effects in Experiment 3. First, the alarm signal may have appeared too slowly (i.e.,five seconds after the words appeared). Second, the number of participants was relatively small (i.e., n=16). Thus, the follow-up experiments (i.e., Experiments 4 and 5) were improved by using a faster alarm signal (i.e., three seconds) to increase task difficulty. Moreover, the sample size was increased to twenty-four participants. Clearly, we know very little about the scenarios where natural speech usually occurs, and it is very likely that both speed and accuracy are equally important for the speaker in most cases. It is worth keeping in mind that Experiments 4 and 5 were not meant to ‘simulate’ a setting where natural speech takes place. Instead, Experiments 4 and 5 were designed to contrast with each other in order to highlight how strategic control factor may change speakers’ production behaviors, by placing speakers in two somewhat extreme cases: one focusing more on accuracy and the other emphasizing more on speed. Therefore, these two experiments should be conceptualized as two sides of the same coin, and their results should ultimately be interpreted in comparison to/relative to each other. As an overview, Chapter 6 is organized as follows. First I will introduce the methodology of Experiments 4 and 5 and explain the rationale underpinning my experimental design, followed by the proposed hypothesis and predictions. Next I will report the production latency results and discuss about the theoretical implications of the findings. I will then 207 present the across-experiment analysis and the error analysis, followed by a general discussion. 6.2 Experiment 4 : Accuracy-oriented Setting 6.2.1 Methodology Method, participants, design, Materials, procedure Experiment 4 had the same methodology as Experiment 3. Twenty-four new people took part in this experiment. The experimental design and materials were the same as those in Experiment 3. The procedure was the same as the one in Experiment 3, except that an alarm signal beeped three seconds instead of five seconds after the words appeared on the screen in attempt to simulate naturalistic circumstances for producing speech, as interlocutors are usually allowed very little formulation time before they initiate speech in daily conversations. Importantly, Experiment 4 differed from Experiment 3 in the wording of the instructions and the screen display of a trial. The purpose of Experiment 4 was to create an accuracy- oriented setting, where the participants may produce sentences when they feel ready. In the instruction, participants were told explicitly to focus on accuracy. To achieve an accurate performance, they were encouraged to prepare their response before initiation. Moreover, the procedure of a trial was different from Experiment 3. On each trial, participants first read the words on the screen. The words disappeared from the screen as 208 soon as they pressed the space bar to trigger recording, in the same vein as the ‘speaking from memory’ scenario in Van Nice & Dietrich (2003), which they assumed to encourage a more compact referent processing in the mental model. In other words, participants could not read the words and say the sentence simultaneously. In this way, the scope for advance linguistic planning before production is likely to be larger than a sub-phrasal unit, resulting in a less incremental process. The procedure of Experiment 4 could potentially prevent participants from initiating an utterance simply with a piece of syntactic information. 6.2.2 Hypotheses and Predictions To begin with, let’s look at the message component. The first hypothesis concerns effects of semantic flexibility. As we saw in the comparison between the correct and perfect responses of Experiment 1, an inhibitory effect of semantic flexibility with agent-theme verbs was found by the production latency data of correct responses, but not by the perfect responses. As mentioned above, Experiment 4 resembled the ‘speaking from memory’ scenario in Van Nice & Dietrich (2003), which they proposed to involve a more compact processing with an interaction between role an animacy. In the accuracy- oriented setting, where participants are encouraged to be more cautious and prepared, they might strategically postpone grammatical encoding till a full representation of relational meaning is conceptualized (i.e. after both thematic roles are assigned). As a result, the two animate nouns would compete for the AGENT role and thus delay 209 production. Thus, I hypothesize that in Experiment 4, an inhibitory effect of semantic flexibility would be observed with agent-theme verbs. However, this prediction is faced with a potential challenge. As we already saw, in Experiment 3, which Experiment 4 was based on, effects of semantic flexibility were not found either from correct responses or perfect responses. Therefore, it is also possible that effects of semantic flexibility is a language-specific phenomenon that only exists in Mandarin, or the design of Experiment 3 was unable to demonstrate such effects in English. If this is the case, results of Experiment 5 may not show any effects of semantic flexibility either. Next, let’s turn to the grammatical component. The second hypothesis concerns effects of thematic prominence. The prediction is again based on the hypothesis that thematic prominence effects are sensitive to the time course of lexical-structural integration. In an accuracy-oriented setting, speakers may prefer to maximize fluency by planning their utterances more extensively before speaking, so they may figure out the relational meaning between the referents and finish thematic role assignment before the message is sent for grammatical encoding. Moreover, to ensure an accurate response, the speaker may strategically generate the specified sentence frame before lexical retrieval begins and then call for words in the order specified by that sentence frame. Imagine that in a condition, the speaker is required to produce a passive sentence containing an agent- theme verb. Strategically, the speaker would first make sure that the to-be-produced sentence is a passive sentence. Meanwhile, based on the thematic structure of the verb, the passive sentence frame specifies a PA TIENT (THEME) role for the subject slot. As 210 thematic role assignment has been completed in the message component and conceptually, a PA TIENT referent is less accessible than an AGENT referent, it would take longer production latency for the sentence-initial word to be retrieved, resulting an effect of thematic prominence. In other words, the effects of thematic prominence found in Experiment 5 could be a consequence of the precedence of structural processing (i.e. the positional stage) over the content processing (i.e. the functional stage). This view is compatible with the view of structural incrementality. Taking these predictions together, the predicted difficulty levels of each verb type are shown in Table 6.1. Notably, in this chapter, I mainly focus on the production latency data of correct responses because the results of Experiment 3 suggest that the error rate might not be a very informative indicator of difficulty level in the English experiment. For the production of active voice, the production latency would be longest for theme- experiencer verbs, followed by agent-theme verbs and experiencer-theme verbs. For the production of passive voice, the production latency would be longest for agent-theme verbs, followed by experiencer-theme and theme-experiencer verbs. Additionally, the mean production latency of agent-theme verbs would be longer than that of experiencer- theme verbs due to an inhibitory effect of semantic flexibility. Moreover, due to effects of thematic prominence, the mean production latency of theme-experiencer verbs would be longer than that of experiencer-theme verbs in the active voice condition, and vice versa in the passive voice condition. 211 Table 6.1: The predicted difficulty level of each verb type in an accuracy-oriented setting Hypothesized effect verb type Semantic flexibility Thematic prominence overall effect Agent-theme verbs active voice -1 +1 0 passive voice -1 -1 -2 Experiencer-theme verbs active voice 0 +1 +1 passive voice 0 -1 -1 Theme-experiencer verbs active voice 0 -1 -1 passive voice 0 +1 +1 6.2.3 Results 6.2.3.1 Data Preparation Exclusions and replacements Production latencies more than 2.5 standard deviations from the mean in each condition were removed from subsequent analyses, amounting to 1.2 % of the correct responses in Experiment 5. There were no empty cells in the item and subject analyses of Experiment5. Coding the Responses Following the coding system of Experiment 3, participants’ utterances were grouped into two main categories: correct responses and errors. The subtypes of correct responses remained the same as those in Experiment 3: (i) perfect response, and (ii) correct but imperfect response, including hesitation, stuttering, correction, repetition. Contrastively, an error was defined by one of the following criteria: (1) Incorrect noun: the participant 212 did not use the noun on the screen, but used another instead. (2) Wrong voice: the participant produced the passive voice instead of the active voice or vice versa. (3) Word addition: the participant added a word not given on the trial. (4) Fail: the participant produced an incomplete sentence or did not say anything. Calculating Production Latencies The production latencies for Experiment 4 were composed of the sum of two parts. The first part was the reading time of the words on the screen. The second part was measured from the onset of the disappearance of the words from the screen (i.e. when the participants pressed the space bar) to the onset of the first content word of a correct sentence, including the duration of speech disfluency if any and the duration of the definite article ‘the’. 13 6.2.3.2 Results Mean Production Latency To investigate whether the factors of VERB TYPE and VOICE affected participant’s production in an accuracy-oriented setting, ANOVA was conducted on the production 13 Same statistical analyses were also conducted on the production latency data of Experiment 4, based on two other benchmarks for calculating production latencies. One measurement was to include only the first part (i.e. the reading time). The other was to include the reading time plus the speech onset time of the definite article “the” in each response. The results regarding the effects of semantic flexibility and thematic prominence did not differ from what is reported here. 213 latencies by participants. 14 . As shown in Figure 6.1, in general, participants spent longer production times in the passive condition than in the active condition. However, the differences between the two voices were smallest for theme-experiencer verbs than for the other two verb types. As a result, a marginally main effect of VERB TYPE was found (F(2,46)=3.178, MSE=.224, p=.056). A main effect of VOICE was also found (F(1,23)=48.911, MSE=6.788, p<.001). Unlike the results of Experiment 4, however, there was an interaction between these two factors (F(2,46)=4.211, MSE=.180, p<.05). 2000 2200 2400 2600 2800 3000 3200 agent-theme exp.-theme theme-exp. Verb type Mean production latency (ms) active voice passive voice Figure 6.1: Mean production latency as a function of verb type and voice in Experiment 4 In addition, to explore what drives the interaction between verb type and voice, and to examine the effects of semantic flexibility and thematic prominence, paired samples t- 14 Note that ANOV A cannot be conducted by items because the items of each verb type are different. 214 tests and independent samples t-tests were conducted on the three verb types in the subject and item analyses, respectively. The mean production latencies of correct responses depending on verb type are shown in Figure 6.2. 200 600 1000 1400 1800 2200 2600 3000 active voice passive voice Voice Mean production latency (ms) agent-theme exp.-theme theme-exp. Figure 6.2: Mean production latency depending on verb type in the active and passive conditions in Experiment 4 For the active voice, agent-theme verbs had the longest production latency (M=2.577, SE=.113), followed by theme-experiencer verbs (M=2.540, SE=.098) and experiencer- theme verbs (M=2.464, SE=.101). However, statistically, the three verb types did not differ from one another. Although the mean production latency for agent-theme verbs was numerically longer than that for experiencer-theme verbs, the difference did not reach statistical significance (t1(23)=1.392, p=.177; t2(22)=1.122, p=.274). Therefore, effects 215 of semantic flexibility were not found. Moreover, the mean production latency for agent- theme verbs was not significantly different from that for theme-experiencer verbs either (t1(23)=.577, p=.569; t2(22)=.353, p=.728). In addition, although the mean production latency for theme-experiencer verbs was longer from that for experiencer-theme verbs, the difference did not reach statistical significance (t1(15)=-1.217, p=.236; t2(22)=-.814, p=.425), indicating that effects of thematic prominence were not found in the active condition. For the passive voice, agent-theme verbs had the longest production latency (M=3.062, SE=.144), followed by experiencer-theme verbs (M=2.987, Mse=.126) and theme- experiencer verbs (M=2.834, Mse=.109). The general pattern conformed to the prediction. Nevertheless, the mean production latency for agent-theme verbs was not significantly different from that for experiencer-theme verbs (t1(23)=1.249, p=.224; t2(22)=.532, p=.600). Again, effects of semantic flexibility were not found. Contrastively, the mean production latency of agent-theme verbs was significantly longer than that of theme- experiencer verbs by subjects (t1(23)=2.934, p<.05; t2(22)=1.578, p=.129). Additionally, the mean production latency of experiencer-theme verbs was significantly longer than that of theme-experiencer verbs by subjects (t1(23)=2.427, p<.05; t2(22)=1.021, p=.318). That is, effects of thematic prominence existed in the passive condition. As a whole, analyses of production latencies showed no effects of semantic flexibility in either the active or the passive conditions. Nevertheless, the tendency for agent-theme 216 verbs to be slightly harder to produce than experiencer-theme verbs in both conditions was consistent with the hypothesis that semantic flexibility hinders production with agent-theme verbs. In addition, the second prediction that there would be effects of thematic prominence in a less incremental process was partly confirmed in the passive condition. Crucially, compared with the findings of Experiment 3, the findings of Experiment 4 are more consistent with those from the correct responses, but different from those from the perfect responses. 6.3 Experiment 5: Speed-oriented Setting 6.3.1 Methodology Method, Participants, Design, Materials and procedure Experiment 5 had the same methodology as Experiment 3. Twenty-four new people took part in this experiment. The experimental design and materials were the same as those in Experiment 3. The procedure was the same as the one in Experiment 3, except that an alarm signal beeped three seconds instead of five seconds after the words appeared on the screen in attempt to simulate naturalistic circumstances for producing speech, as interlocutors are usually allowed very little formulation time before they initiate speech in daily conversations. Importantly, Experiment 5 differed from Experiment 3 in the wording of the instructions and the screen display of a trial. The purpose was to manipulate speakers’ strategic 217 control over the trials in order to motivate a highly incremental sentence production process. In the instruction, participants were told explicitly to focus on their speed, and to make a response before the 3-second tone sounded. The words remained on the screen throughout a trial, so it was possible for participants to read and speak simultaneously. Van Nice & Dietrich (2003) used a similar design in their ‘speaking with pictures in view” experiment to create sequential referent processing, though their work involved picture-description tasks. Additionally, in Experiment 5, a progress bar which gradually filled up in three seconds was displayed at the central bottom of the screen on each trial (in addition to the words also shown on the screen). The progress bar has been used in previous studies and is considered an effective way of motivating faster responses (e.g. Ferreira & Swets, 2002). Crucially, the progress bar was used in a speed-oriented setting (i.e., Experiment 5), but not in an accuracy setting (i.e., Experiment 4). 6.3.2 Hypotheses and Predictions Based on the finding of Experiments 1 and 2, in the message component, effects of semantic flexibility are hypothesized to be conditioned by verb thematic structure. Specifically, when the thematic roles to be assigned are AGENT and THEME, competition is likely to take place between two animate nouns because agenthood is closely related to animacy. In contrast, no competition would arise when the thematic roles to be assigned are THEME and EXPERIENCER, because both roles are equally likely to be animate. As mentioned above, the speed-oriented setting resembles a 218 ‘speaking with pictures in view’ setting, which Van Nice & Dietrich (2003) proposed to involve a sequential referent processing without interaction between role and animacy. Therefore, I predict that role and animacy information would not get a chance to interact in a highly incremental process, and the inhibitory effect of semantic flexibility due to the interaction between agent-theme verbs and noun animacy would not be found. Moreover, as we already saw in Chapter 5, the results of perfect responses in Experiment 3 did not show effects of semantic flexibility, either. It is also possible that under time pressure, participants would not have enough time to exploit the availability of multiple propositions, as evidenced by participants’ tendency to start a sentence with the left noun in most cases. Since the purpose of Experiment 5 is to create the scenario for producing a fast response, in parallel with the perfect responses found in Experiment 3, I predict that effects of semantic flexibility would not be observed in Experiment 5. Turning to the grammatical component, effects of thematic prominence are hypothesized to cause a more prominent thematic role to be mentioned earlier in a sentence. Such effects would potentially make the less preferred structure more difficult to produce. However, in Experiment 3, effects of thematic prominence were found from correct responses but not from perfect responses in the passive condition. I propose that thematic prominence effects are sensitive to the time course of lexical-structural integration. In a highly incremental process, lexical retrieval would take precedence over structural retrieval. Presumably, in a speed-oriented setting, the scope of linguistic planning is likely to be narrowed to a sub-phrasal unit (i.e., a word). That is, speakers aiming at fast 219 speed are likely to form a strategy of initiating an utterance as soon as one piece of lexical information becomes available. On a trial where both nouns are animate and can take either thematic role, the speaker may immediately dispatch the first retrieved word, most likely the noun on the left, even before it is assigned a thematic role. More recent accounts for serial models generally assume that different components of the model can interact with one another during the process of production. Thus, speakers are able to continue planning their sentences while speaking. In other words, after one noun is retrieved from the mental lexicon, it is then sent to the structural processing and claims the subject position followed by phonological encoding, with message and sentence formulation still in process. Meanwhile, since there is a specified structure in each condition, the structural frame may intergrate with the lexical retrieval shortly after the sentence-intial word is articulated. What follows is the rest of the sentence will simply develop on the ‘blueprint’ by inserting lemmas into the already specified ‘slots’. In other words, the lack of effects of thematic prominence in Experiment 5 could be a consequence of the precedence of the content processing (i.e., the functional stage) over structural processing (i.e., the positional stage). This view is compatible with the view of lexical incrementality. As a whole, I predict that neither effects of semantic flexibility nor thematic prominence would be observed when speakers are under pressure to initiate speech quickly, at least in a condition where there is no syntactic flexibility. Table 6.2 shows the hypothesized effects of semantic flexibility and thematic prominence on sentence production in 220 Experiment 5. Assuming the effects of semantic flexibility and thematic prominence are independent of each other, the overall effects can be reflected by the difficulty level of each condition. Notably, here, I mainly focus on the production latency data of correct responses because the results of Experiment 3 suggest that error rate might not be a very informative indicator of difficulty level in the English experiment. In summary, taking these predictions together, there would be no differences among the mean production latencies of the three verb types. It is crucial to note that these predictions differ fundamentally from those for Experiment 4, where effects of semantic flexibility and thematic prominence are both expected to exist due to a less incremental process. Thus, by looking at these two experiments together, we can see whether degree of incrementality may condition the effects on message formulation and grammatical encoding processes. Table 6.2: The predicted difficulty level of each verb type in a speed-oriented setting Hypothesized effect verb type Semantic flexibility Thematic prominence overall effect Agent-theme verbs active voice 0 0 0 passive voice 0 0 0 Experiencer-theme verbs active voice 0 0 0 passive voice 0 0 0 Theme-experiencer verbs active voice 0 0 0 passive voice 0 0 0 Note: zero means no effect. 221 6.3.3 Results 6.3.3.1 Data Preparation Exclusions and Replacements Production latencies more than 2.5 standard deviations from the mean in each condition were removed from subsequent analyses, amounting to 2.5 % of the correct responses. There were no empty cells in the subject and item analyses. Coding the Responses The coding system was the same as that of Experiment 4. Calculating Production Latencies Following the method of Experiment 3, the production latency in Experiment 5 was measured from the onset of the presentation of a trial to the onset of the first content word of a correct response, including the time of speech disfluency if any and the duration of the definite article ‘the’. 6.3.3.2 Results Mean Production Latency To investigate whether the factors of VERB TYPE and VOICE affected participant’s production in a speed-oriented setting, ANOVA was conducted on the production latencies by participants. As shown in Figure 6.3, in general, participants spent longer 222 production times in the passive condition than in the active condition. Additionally, they spent shorter production times with theme-experiencer verbs in both conditions. As a whole, the two patterns are identical. Thus, a marginally main effect of VERB TYPE was found (F(2,46)=3.218, MSE=.070, p=.055). A main effect of VOICE was also found (F(1, 23)=18.128, MSE=.785, p<.001). However, there was no interaction between these two factors (F(2, 46)=.212, MSE=.006, p=.807). 1100 1150 1200 1250 1300 1350 1400 1450 agent-theme exp.-theme theme-exp. Verb type Mean production latency (ms) active voice passive voice Figure 6.3: Mean production latency as a function of verb type and voice in Experiment 5 To further examine the effects of semantic flexibility and thematic prominence, paired samples t-tests and independent samples t-tests were conducted on the production latency 223 of the three verb types in the subject and item analyses, respectively. Following the rationale in Experiment 3, effects of semantic flexibility can be investigated by comparing agent-theme verbs and experiencer-theme verbs, whereas effects of thematic prominence can be investigated by comparing experiencer-theme and theme-experiencer verbs. The mean production latencies of correct responses depending on verb type are shown in Figure 6.4. 0 400 800 1200 1600 active voice passive voice Voice Mean production latency (ms) agent-theme exp.-theme theme-exp. Figure 6.4: Mean production latency depending on verb type in the active and passive conditions in Experiment 5 For the active voice, no differences were found among the three verb types. The mean production latency for agent-theme verbs (M=1.303, SE=.059) was not different from the production latency for experiencer-theme verbs (M=1.250, SE=.046) (t1(23)=1.749, p=.094; t2(22)=.575, p=.571), or that for theme-experiencer verbs (M=1.223, MSE=.042) 224 (t1(23)=1.549, p=.135; t2(22)=1.022, p=.318). In addition, the mean production latency for experiencer-theme verbs was not significantly different from that for theme- experiencer verbs either (t1(23)=.529, p=.602; t2(22)=.658, p=.518). Similarly, for the passive voice, no differences were found among the three verb types. The mean production latency for agent-theme verbs (M=1.434, SE=.060) was not different from the production latency for experiencer-theme verbs (M=1.423, MSE=.079) (t1(23)=.190, p=.851; t2(22)=.342, p=.736), or that for theme-experiencer verbs (M=1.362, SE=.066) (t1(23)=1.682, p=.106; t2(22)=.768, p=.451). Additionally, there was no significant difference between experiencer-theme verbs and theme-experiencer verbs either (t1(23)=1.474, p=.154; t2(22)=1.237, p=.229). As a whole, the production latency results in both the active and passive conditions did not show effects of semantic flexibility or thematic prominence, conforming to the prediction that in a highly incremental process, both effects of semantic flexibility and thematic prominence might not play a role. However, it is worth mentioning that in both conditions, the mean production latencies of agent-theme verbs were numerically longer than experiencer-theme verbs, suggesting that it was harder to construct a sentence with agent-theme verbs than with experiencer-theme verbs. Crucially, the findings here replicated those from perfect responses in Experiment 3, implying that they probably both involved a highly incremental process. 225 6.4 Cross-Experiment Analysis Between Experiments 4 and 5 6.4.1 Production Latency Analysis A post-hoc cross-experiment analysis on production latency was conducted to further investigate whether participants employed different strategies in these two experiments. Since Experiments 4 and 5 were based on the same materials and design with equal numbers of participants, ANOVA with repeated measures was conducted using EXPERIMENT as a between-participants variable. The goal was to explore whether or not the experimental design led participants to behave in different ways under the influence of distinct experimental settings and instructions. Figure 6.5 shows a comparison between these two experiments. First, the different instructions and procedure of the two experiments had an impact on production. There was a main effect of EXPERIMENT (F(1,23)=242.667, MSE=143.467, p<.001). As a whole, production latencies were longer in Experiment 4 (accuracy-oriented), which was expected given the design differences between the two experiements. Second, in general, participants were sensitive to the factors of verb type and voice. The main effects of VERB TYPE (F(2, 46)=4.841, MSE=.295, p<.05) and VOICE (F(1, 23)=54.058, MSE=6.094, p<.001) were both reliable. In general, participants spent longer time with agent-theme verbs in both experiments. They also spent longer time in the passive condition in both experiments. Third, participants’ behaviors in the active and passive conditions changed with experiments: As a whole, the difference between the active and passive conditions was larger in Experiment 4 in Experiment 5. Although ANOVA found an interaction between VOICE and EXPERIMENT (F(1,23)=21.327, MSE=1.479, p<.001), the effects of verb 226 type on production remain constant across experiments. There was no interaction between VERB TYPE and EXPERIMENT (F(2,46)=.763, MSE=.034, p=.452). 1000 1500 2000 2500 3000 3500 Agent-theme Experiencer- theme Theme- experiencer Mean production latency (ms) Active voice (Exp.5) Passive voice (Exp.5) Active voice (Exp. 4) Passive voice (Exp.4) Figure 6.5: Mean production latency as a function of verb type and voice in Experiments 4 and 5 6.4.2 Error Analysis In addition to analyzing production latencies, a closer look at the error types and subtypes of correct responses can also provide useful information for understanding the sources of difficulty during online production and participants’ behavioral patterns in each 227 experiment. Moreover, although the production latency data demonstrated an effect of EXPERIMENT between participants, it remains a question whether speakers employed different strategies in these two experiments. To investigate this assumption, I compared Experiments 4 and 5 with three indicators: (i) error rates (i.e., how often an error occurs in a condition), (ii) perfect-response rates (i.e., how often a perfect response occurs in a condition), and (iii) the left/right ratios (i.e., correct responses starting with the noun on the left side of the screen vs. the noun on the right side of the screen). Additionally, I also compared the counts of different error types and the subtypes of correct responses. The goal was to explore whether these two settings engaged participants in distinct production behaviors and how these behaviors differed. Error Rate and Perfect-response Rate As mentioned earlier, in Experiments 4 and 5, a correct response belonged to either of the two categories: (i) Perfect responses: the sentence contained all given words, was well- formed, fluently produced and semantically plausible. (ii) Correct but imperfect responses: the sentence contained all the words and was well-formed and semantically plausible, but accompanied with speech disfluencies (e.g., “uh…the gardener was hurt by the sailor”). The error rate of Experiment 4 with an accuracy-oriented setting was 3.8 % (i.e., n of errors=22/576), while the error rate of Experiment 5 with a speed-oriented setting was 6.6 % (i.e., n of errors=38/576). That is, despite the same materials, participants were less error-prone in the accuracy-oriented setting. In addition, the 228 perfect-response rate of Experiment 4 was 90.8 % (i.e., n of perfect responses=523/576), while the perfect-response rate of Experiment 5 was 68.5 % (i.e., n of perfect responses=395/576). The strikingly high proportion of perfect responses in Experiment 4 indicates that as they were instructed, participants focused on accuracy and were well- prepared for their speech most of the time. In contrast, participants did not aim at accuracy as much in Experiment 5, so the perfect-response rate was much lower. The Ratio of Sentences Starting from the Left Noun vs. the Right Noun In order to measure the degree of incrementality,, I posit that the ratio of correct responses starting with the left noun against those starting with the right noun in each experiment might serve as an indicator of how much is planned before an utterance is made. Based on the “first-retrieved, first-mentioned” principle in a radically incremental model, a higher ratio of left noun against right noun might indicate a smaller planning scope. Given the fact that reading proceeds from left to right in English, the left noun is assumed to be retrieved first since both nouns were animate. In Experiment 4 where participants aimed at accuracy, the left/right ratio was 2.35 (i.e., n of left noun=389 (70.2%); n of right noun=165 (29.8%); X 2 =90.57, p<.001). In Experiment 5 where participants aimed at speed, the left/right ratio was increased to 5.81 (i.e., n of left noun=459 (85.3%); n of right noun = 79 (14.7%); X 2 =268.4, p<.001). In general, participants tended to start with the left noun. However, they started with the left noun much more often in a speed-oriented setting than in an accuracy-oriented setting. In other 229 words, the contrast in ratios indicate that sentence production was largely lexically driven in a speed-oriented setting, but this tendency was less strong in an accuracy-oriented setting: Presumably, around 30% of the correct responses were not initiated till both nouns were retrieved. That is, the scope of linguistic planning seemed to vary in these two settings. Error types In Experiments 4 and 5, an error belonged to one of the following types: (1) Incorrect noun: the participant failed to use the noun on the screen, but used another instead. (2) Wrong voice: the participant produced the passive voice instead of the active voice or vice versa. (3) Word addition: the participant added a word not given on the trial. (4) Fail: the participant produced an incomplete sentence or did not say anything. Table 5.3 shows the count of each error type in Experiments 4 and 5. Note that the comments are speculative due to small numbers. Table 6.3: Count of each error type in Experiments 4 and 5 Experiment 4 Experiment 5 error type (accuracy-oriented) (speed-oriented) Incorrect noun 8 2 Wrong voice 13 40 Word addition 5 4 Fail 10 10 Total 36 56 230 Since these two experiments used the same materials and design, it is not surprising that the errors made by participants have resulted in almost identical patterns. In general, wrong voice was the most frequent error type: 13 out of 36 in Experiment 4 (X 2 =3.778, p=.286); 40 out of 56 in Experiment 5 (X 2 =66.857, p<.001). Moreover, a closer inspection of these wrong voice errors showed that they occurred mostly with theme- experiencer verbs in the active condition. This tendency suggests that participants sometimes prefer to use theme-experiencer verbs in the passive voice, especially in the speed-oriented setting. As discussed in Chapter 4, the reason why it was easier to make a passive sentence in the active condition but not vice versa may be due to the relative ease to add a preposition rather than delete it. Moreover, this kind of errors appeared more often in Experiment 4 because according to the monitoring theory, when under time pressure, if the wrong structure is activated from the very beginning, there is little chance for the speaker to make corrections (Oomen & Postma, 2001). Subtypes of correct responses Depending on its content, a correct response was further categorized as one of the following types: (i) perfect responses, (ii) hesitation, (iii) correction, and (iv) repetition. Except for being a perfect response, a correct response may fit in one or more categories (e.g., when the speaker hesitated and self-corrected in one single sentence). Table 6.4 and 6.5 summarize the count of each subtype of correct responses depending on verb type in Experiments 4 and 5. A comparison between these two tables leads to two observations. 231 First of all, perfect responses were evenly distributed across verb types (all ps>.05). More importantly, participants hesitated much more often in the speed-oriented setting, suggesting uncertainty in the intended message or incomplete speech planning. In contrast, in the accuracy-oriented setting, prefect responses were dominant in numbers and speech disfluencies rarely occurred. Table 6.4: Count of the subtypes of correct responses depending on verb type in Experiment 4 subtype of Agent-theme Experiencer-theme Theme-experiencer Total correct responses verbs verbs verbs Perfect response 264 250 256 770 Correction 4 3 0 7 Hesitation 1 2 5 8 Repetition 0 1 1 2 Table 6.5: Count of the subtypes of correct responses depending on verb type in Experiment 5 subtype of Agent-theme Experiencer-theme Theme-experiencer Total correct responses verbs verbs verbs Perfect response 200 201 193 594 Correction 19 20 9 48 Hesitation 55 55 55 165 Repetition 1 6 1 8 To conclude, the between-participants analysis and error analysis corroborated that speakers employed different strategies for making responses in these two experiments. In 232 the accuracy-oriented setting, speakers tended to prepare their speech before speaking, resulting in a lower proportion of speech disfluencies, a lower error rate and a much higher rate of perfect responses. In contrast, in the speed-oriented setting, speakers tended to initiate an utterance earlier, usually before speech planning is complete, thus resulting in a higher proportion of speech disfluencies, a higher error rate and a lower rate of perfect responses. 6.5 Discussion The main aim of this chapter was to investigate how speakers’ strategic control may condition the effects of semantic flexibility in the message component and the effects of thematic prominence in the grammatical encoding during online production. As a whole, effects of semantic flexibility were not found in both experiments. Thus, the hypothesis concerning a sequential vs. compact referent processing in the message component is not verified. In contrast, effects of thematic prominence were found to vary in the two experiments, providing evidence for the hypothesis that grammatical encoding is susceptible to speakers’ strategy control. To begin with, although in Experiments 4 and 5, we saw a tendency for sentences containing an agent-theme verb to be slightly harder to produce than those containing an experiencer-theme verb in both the active and passive conditions, significant effects of semantic flexibility were not found. Arguably, this finding might suggest that speakers’ strategic control did not have an impact on the process of message formulation, probably 233 because message formulation is an automatic process that is beyond control. However, this claim needs to be dealt with caution since significant effects of semantic flexibility were not found in all three English experiments. It is possible that effects of semantic flexibility are a language-specific phenomenon in Mandarin Chinese, resulting from the interaction of verb thematic structure and the special semantic properties pertaining to BA and BEI construction. Specifically, as discussed in Chapter 3, the BA and BEI constructions have distinct requirements associated with the ‘affectedness’ and ‘causality’ of its noun arguments. These requirements may cause conceptual complexity when the speaker is assigning AGENT and THEME roles to the animate referents. In contrast, English active and passive voices do not have these semantic properties, so the difficulty resulting from conceptual complexity is not expected. Without doubt, a direction worth pursuing in my future work is to investigate the factor of strategic control on effects of semantic flexibility in Mandarin, based on the materials of Experiments 1 and 2 and the design of Experiments 4 and 5. Second, the findings concerning effects of thematic prominence support the hypothesis that strategic control can potentially influence the process of grammatical encoding. In the speed-oriented setting, effects of thematic prominence were not found in both conditions. In contrast, effects of thematic prominence were found in the passive condition of the accuracy-oriented setting. This asymmetry may be attributed to the variability of advance planning scope in these two experiments, which involves distinct time courses of lexical-structural integration in early sentence formulation: In Experiment 234 4, a broader linguistic planning scope might enable the specified sentence frame to be generated before lexical retrieval. In contrast, in Experiment 5, a smaller linguistic planning scope was adjusted under time pressure and the specified structure ‘kicks in’ shortly after the first word is retrieved. Consequently, I argue that grammatical encoding in an accuracy-oriented setting may be executed with structural incrementality. As proposed in Bock et al. (2003, 2004), early sentence formulation may begin with a simple sentence plan that captures the relationships between various message elements. Support for structural incrementality comes from existing work. For example, eye-tracking data show longer planning times for the first content word than later words, suggesting the possible generation of information about both the sentential start point and a continuation of the utterance from that point (Bock et al., 2003; Griffin & Bock, 2000). Additional evidence comes from production tasks investigating the effects of syntactic priming (Bock, 1986a; V. Ferreira & Bock, 2006, Konopka, 2012). Briefly, speakers tend to use a previously processed structure persistently on later trials, suggesting that structural information is accessible in early stage of sentence formulation. In an accuracy-oriented setting, speakers may strategically favor structural incrementality in order to maximize fluency. By engaging in extensive planning, they are able to map the message onto a linguistic structure more easily, and to reduce the possibility of speech errors and speech disfluencies. 235 In contrast, I argue that grammatical encoding in a speed-oriented setting may be executed with lexical incrementality. In this setting, speakers may strategically favor the ‘first retrieved, first mentioned principle’ in order to minimize speech onset times. By retrieving the sentence-initial word first and building a structure accordingly, there is little need for anticipatory activation for the upcoming message elements, though at the expense of a higher probability of producing errors and disfluencies. It needs to be stressed that presumably, the speed-oriented setting and the accuracy- oriented setting differ in two fundamental aspects. First, thematic role assignment may not be complete in the speed-oriented scenario. Under the pressure for speeding up their responses, speakers may strategically start an utterance with the left noun without assigning a thematic role to this referent. In contrast, in planned speech, speakers may be willing to spend more time figuring out the relational meaning between the two referents. In other words, in the former situation, the lexical representation of a referent may be retrieved without carrying the information of its thematic role before it is sent to grammatical encoding, while in the latter, a thematic role has been assigned to the referents before their lexical representations are retrieved. Second, the time course of lexical-structure integration may vary in these two scenarios. In the speed-oriented setting, once a word is retrieved, it grabs the sentence position right away. Then, the structural information would kick in and provide a blueprint for the remaining structure. To a large extent, this is because the design of Experiment 5 allows speakers to speak and read the words at the same time. When exposed to time pressure, speakers in Experiment 5 may 236 develop a strategy of ‘dispatching the first word right away’ and look for the structural clue (i.e. the presence/absence of by ) instantly by reading on. Since the nature of the task was very straightforward and the target items only involved the production of active and passive voices, this kind of strategy seemed to work out successfully most of the time. It is worth noting, however, it seems that this strategy did not work that well on the filler trials, which generally involved more words and more complicated sentence, as evidenced by frequent disfluencies and errors on the filler trials. However, it remains to be explained why in Experiement 4, effects of thematic prominence were not found in the active condition. This finding could be attributed to a general ceiling effect in the active condition, as indicated by the high perfect-response rate in the active condition of Experiment 5 (95 %). In English, the active voice dominantly outnumbers the passive voice (Hopper & Thompson, 1980; Shintani, 1979; Svartvik, 1966). It is also easier to produce an active sentence than a passive sentence, as evidenced by the generally shorter mean production latencies across verb type in the active condition. Therefore, I argue that it is difficult to observe effects of thematic prominence by comparing experiencer-theme verbs and theme-experiencer verbs in the active condition, because in general an active sentence is easy to produce regardless of verb type. To conclude, the two experiments reported in this chapter shed light on the important question of whether language production is an incremental and automatic process. The 237 answer most consistent with the results reported here and in previous work (e.g. F. Ferreira & Swets, 2002) is that the system is not architecturally incremental. Instead, the scope of planning is at least partly under speakers’ control, and it depends on the communicative goal speakers aim to achieve: speed or accuracy. However, although it is clear that this kind of flexibility is available in grammatical encoding, little is known about whether the processes of message formulation are also adaptable to planning strategies based on the findings of the present work. Thus, it would be a promising attempt to extend the research direction to Mandarin Chinese since effects of semantic flexibility were found in Experiments 1 and 2. The findings regarding thematic prominence effects in this chapter suggest that the operations of grammatical encoding are flexible and controllable. As Konopka (2012) suggested, the extent of advance planning scope forms a continuum between lexical incrementality and structural incrementality and follows from the speaker’s strategies. Therefore, at different times, the sequence of encoding operations may fall somewhere on this continuum, being compatible with either. Moreover, the dynamic time course of lexical-structural integration implies that the two stages in the grammatical component are interacting with each other during the process of sentence formulation. Although in a serial model the three components are hierarchically structured, most recent sentence production theories generally assume that the three components can be operated in parallel, and it is possible to have bi-directional information flows between these levels. As V. Ferreira & Slevc (2007:458) pointed out, “this is an important processing detail of 238 the consensus model that is relevant to its staged character, but not a fundamental challenge to that staged character.” 239 Chapter 7: Conclusions 7.1 Summary of the Current Work The study of sentence production investigates how speakers turn thoughts into utterances. In my dissertation, I focused on the nature and processes of the message component and the grammatical component within the framework of existing language production theories. The three primary goals of my dissertation are to investigate: (i) the effects of freedom of choice on the process of message formulation, (ii) the effects of thematic prominence and animacy on speakers’ syntactic choices in Mandarin Chinese, and (iii) speakers’ strategic control in response to distinct communicative goals (i.e. aiming at speed or accuracy), which potentially affects the processes of sentence production. In the hope of exploring the nature of the message component, the first key component of my research asks whether having freedom of choice in the message component (termed semantic flexibility) facilitates or complicates the process of language production. Given limited processing capacity, will message formulation slow down if a speaker is faced with choices between more than one possible message to be conveyed? This research builds on work looking into the effects of flexibility on the level of grammatical encoding (V. Ferreira, 1996). Intuitively, having a choice (e.g., the give-type sentence) appears detrimental for decision-making, since competition for selection potentially causes delays. However, contrary to intuition, Ferreira’s findings suggest that syntactic flexibility facilitates production, as indicated by shorter speech-onset times. He attributed this result to the highly incremental nature of the grammatical encoding process. In contrast, since 240 our thoughts are abstractly represented and may not be temporally organized, it is worth investigating how flexibility may influence message formulation. To explore this issue, I conducted two experiments on Mandarin Chinese using BA and BEI constructions. I investigated whether ease of production would be influenced by whether the two nouns are both animate or differ in animacy. For example, when a speaker is given two animate nouns (e.g., director, actress) and a verb (e.g., criticize), s/he can say either “the actress criticized the director” or “the director criticized the actress.” In contrast, semantic flexibility is not available if the two nouns differ in animacy (e.g., director , movie) because a movie cannot criticize a director. Importantly, I also explored effects of verb type by looking at agent-theme verbs (e.g., AGENT criticized THEME) and theme-experiencer verbs (e.g., THEME surprised EXPERIENCER). Participants were asked to generate sentences using nouns and verbs displayed on the computer screen. Their production latencies (i.e. how quickly people begin to speak after seeing the nouns on the computer screen) as well as error rates were analyzed. While one might at first glance expect the availability of semantic flexibility to result in difficulties/slowdowns, my experimental results suggest a more complex pattern: Message formulation processes are highly sensitive to the interaction between thematic role and animacy. Semantic flexibility hinders production with agent-theme verbs but has no effects on production with theme-experiencer verbs. Based on this asymmetry, I 241 hypothesize that message formulation can be carried out competitively or incrementally, depending on linguistic information. In other words, message formulation may be executed differently from grammatical encoding, which is generally regarded to be incremental. The asymmetry in effects of semantic flexibility may be explained by the observation that agenthood is closely tied to animacy semantically (Gruber, 1976; Jackendoff, 1972). Thus, the two animates would compete for the AGENT role. In contrast, with theme- experiencer verbs, since an animate noun can plausibly be THEME or EXPERIENCER, there is no competition between the two animates. As a result, semantic flexibility has no effects on the production of sentences containing a theme-experiencer verb. Moreover, the inhibitory effect of semantic flexibility with agent-theme verbs may be strengthened by the special semantic properties of the Mandarin BA and BEI constructions, which impose semantic features such as “affectedness” and “causality” on their arguments, making the assignment of AGENT and THEME roles even more difficult in the mental model. In a broader sense, these results indicate that subtle linguistic information guides language production very early on. Relevant to the second key component of my research, the results of Experiment 1 and 2 also extended the findings in F. Ferreira (1994) on effects of thematic prominence and animacy to Mandarin Chinese sentence production. In general, speakers preferred to place a more prominent thematic role in an earlier sentence position, and the effect was 242 larger when both nouns differed in animacy. The findings provide cross-linguistic evidence for the incrementality-based account of grammatical encoding, which posits that a more accessible noun is dispatched sooner based on the “first-retrieved, first mentioned” principle. Moreover, as the first attempt to study Mandarin BA and BEI constructions side by side from a production perspective, the findings of Experiments 1 and 2 demonstrated that participants produced more ba sentences than bei sentences with agent-theme verbs, and the pattern was opposite with theme-experiencer verbs. In addition, both the error rate and production latency data revealed that ba sentences are easier to produce with agent-theme verbs, while bei sentences are easier to produce with theme-experiencer verbs. Taking these results together, the contrast in ease of production between the BA and BEI constructions cannot simply result from structural difference or frequency effects. As a whole, my findings indicate that to express the same proposition, Mandarin speakers’ choice between the BA and BEI constructions is to a large extent conditioned by verb thematic structure. The present work helps shed light on a better understanding of the nature of Mandarin BA and BEI constructions. Moreover, in order to further explore whether semantic flexibility is conditioned by the interaction of thematic role and animacy, and to see whether the findings from Mandarin can be extended cross-linguistically, I conducted Experiment 3 in English to investigate effects of semantic flexibility on three different verb types: agent-theme verbs (e.g., criticize), theme-experiencer verbs (e.g., surprise), and experiencer-theme verbs (e.g., adore). Based on the same hypothesis as Experiments 1 and 2, I predict that semantic 243 flexibility hinders language production with agent-theme verbs, but has no effects on language production with theme-experiencer and experiencer-theme verbs. With two animate nouns and a verb in view on the screen, participants were instructed to produce English actives when the preposition by is absent, or to produce passives when by is present. However, only effects of thematic prominence but not effects of semantic flexibility were found in the English experiment. The absence of semantic flexibility effects in English may seem surprising, given that we found them in Mandarin. This difference may be due to the relative ease of the task used in the English experiment -- which failed to reveal subtle differences in the message-formulation level -- or it may point to a semantic distinction between the Mandarin BA/BEI constructions and the English active/passive voices. The third key component of my research is to investigate how the extra-linguistic factor, namely, speakers’ strategic control, affects sentence production in both the message and grammatical component. Since systematic discrepancies between the results of correct responses and perfect responses from Experiments 1 to 3 were observed, I hypothesize that effects of semantic flexibility and thematic prominence, or more broadly, the processes of message formulation and grammatical encoding, may be conditioned by speakers’ strategic control. This hypothesis is based on existing findings which suggest that speakers’ strategy has the potential to affect the scope of advance linguistic planning, i.e. the incrementality of speech production. I further hypothesize that for fast speech and prepared speech, effects of semantic flexibility and thematic prominence would result in 244 different patterns. In order to test these hypotheses, Experiments 4 and 5 were designed by manipulating the strategic control factor, attempting to create a speed-oriented setting vs. an accuracy-oriented setting. Though a large body of research agrees that grammatical encoding proceeds in a piecemeal fashion, it is not clear yet just how incremental this process is. The scope of linguistic planning can vary from the smallest bit of linguistic structure such as a phonological word (Brysbaert et al., 1998) to a large chunk of an upcoming utterance (Stallings et al., 1998). Based on the conflicting results, Bock et al. (2003, 2004) propose two types of incrementality in sentence production: structural incrementality and lexical incrementality. The critical distinction lies in the order of the operations of lexical retrieval and structural retrieval in early sentence formulation. Moreover, it has been proposed that the scope of advance planning can be adjusted under time pressure (F. Ferreira & Swets, 2002). To be more specific, the planning unit may be smaller in a fast response than in a prepared response, because under time pressure, speakers tend to initiate articulation earlier. To explore to what extent the production system is flexible, I conducted two experiments to investigate whether speakers aiming at speed or accuracy behaved differently when producing speech. Consider a situation where a speaker is under time pressure and aims at speed vs. another situation where there is enough planning time and accuracy is crucial: Will speech production be more incremental in one case and less incremental in the other? 245 Previous research mainly focuses on the scope of linguistic planning in the grammatical component. My work aims at both the grammatical component and the message component and asks whether both components are susceptible to strategic control, or are operated in a more automatic way. This line of research can shed light on the architecture of the production system by helping us understand to what extent the system is automatic vs. controllable. The results of Experiments 4 and 5 showed that effects of semantic flexibility were not found in both experiments. Thus, the hypothesis that effects of semantic flexibility would vary due to a sequential vs. compact referent processing in different settings (Van Nice & Dietrich, 2003) is not verified. In contrast, effects of thematic prominence were only found in the accuracy-oriented setting but not in the speed-oriented setting, suggesting that speakers’ strategic control can affect the processes of grammatical encoding. In this sense, the production system is not architecturally incremental. Instead, the scope of planning is at least partly under speakers’ control, depending on whether the goal speakers aim to achieve is speed or accuracy. In conclusion, my dissertation is dedicated to issues at the interface of theoretical linguistics and cognitive psychology, focusing on the process of message formulation and grammatical encoding in sentence production. My findings suggest that message formulation is sensitive to the interaction between linguistic information such as verb thematic structure and conceptual accessibility such as noun animacy. Moreover, the language production system is susceptible to extra-linguistic factors such as strategic control, in particular the grammatical component. By exploring the nature and processes 246 of the individual components as well as the interactions between them, my research makes contributions to the understanding of how people produce speech. 7.2 Concluding Remarks on Future work I propose that the current work can be extended along four distinct directions. First, the results from the two Mandarin experiments suggest that regardless of frequency effects or structural differences, speakers prefer to produce ba-sentences with agent- theme verbs and bei-sentences with theme-experiencer verbs during online production under the influence of thematic prominence and animacy effects. An interesting question is whether thematic prominence and noun animacy are only of importance from the perspective of language production theories. Can we find support for these cognitive effects from other sources, such as oral and written corpora? A corpus study aiming to explore how distributions of BA and BEI construction are correlated with verb type and animacy may provide an answer for this question. Second, as we saw from the discussions about the English experiments, participants’ performances seemed to have reached a ceiling in the English sentence production tasks, very possibly due to task simplicity for producing an active or passive sentence. What also concerns us is the strikingly low error rate in the passive condition. A possible account for the low error rate is the presence of by in the passive condition provided a very strong structural clue for the speaker. Uhfortunately, ceiling effects potentially 247 obscured significant differences and rendered it difficult to discover semantic flexibility and thematic prominence effects. One possible way to improve the design is to use get- passives in place of be-passives. In syntactic theories, a prepositional phrase (e.g., by someone) occupies a peripheral position of a syntactic tree and is considered an adjunct. Moreover, by-phrase can often be omitted. In contrast, get is considered part of the core verbal complex of the sentence, and is NOT optional in the passive voice. Another advantage of using get-passives instead of be-passives is because English get-passives share some common features with Mandarin BEI construction. In the syntactic dimension, the structures of English get-passives and Mandarin bei-sentences are generally assumed to be base-generated, comprising a main verb and its arguments or predicates. In contrast, English be-passives are generally argued to be derived via movement. In the semantic dimension, both get-passives and bei-sentences often indicate a situation with an adverse or beneficial effect on the EXPERIENCER or PA TIENT subject (e.g., “The boy got frightened by the story.”). Contrastively, be-passives usually denote a neutral meaning (e.g., “The boy was frightened by the story”) (e.g., Carter & McCarthy, 1999; Chappell, 1980; Givón & Yang, 1994). Therefore, it seems more appropriate to compare the production of English get-passives rather than be-passives with Mandarin bei-sentences from a cross-linguistic point of view. Third, in Chapter 6, the arguments developed from the views of structural incrementality and lexical incrementality focus on situations where syntactic flexibility is not available (i.e., Experiments 4 and 5). Specifically, in Experiments 4 and 5, structural information 248 (i.e., the absence or presence of by) is accessible to the speaker at the beginning of a trial, which enables structural retrieval to be triggered in early sentence formulation. However, under naturalistic circumstances, syntactic structures are usually not explicitly specified or pre-identified by the speaker. Therefore, based on the incremental nature of grammatical encoding, thematic prominence effects may play a role regardless of degrees of incrementality. Experimental settings that are similar to a naturalistic scenario are the no-prep. conditions in Experiments 1 and 2, or the experiments in F. Ferreira (1994), where speakers can freely choose between alternative structures. It is worth mentioning that both the no-prep. conditions of my Chinese experiments and Ferreira’s experiments demonstrated effects of thematic prominence. Thus, future work is desirable to explore whether effects of thematic prominence are also sensitive to strategic control when syntactic flexibility is available. In other words, it merits further investigation whether grammatical encoding is susceptible to strategic control if structural information is not accessible in early sentence formulation. Fourth, as indicated in Chapter 6, the results of Experiments 4 and 5 did not verify whether effects of semantic flexibility are conditioned by strategic control. This is probably because the English experiments in the current work are incapable of finding effects of semantic flexibility in the first place. In contrast, results from the Mandarin experiments demonstrated significant effects of semantic flexibility. Therefore, it would be a promising effort to extend my research direction into Mandarin Chinese and 249 investigate how strategic control may influence message formulation by means of Chinese production tasks. 250 References Ahrens K. (2003). Verbal integration: The interaction of participant roles and sentential argument structure. Journal of Psycholinguistic Research, 32:497–516. Allum, P. H. & Wheeldon, L. R. (2007). Planning scope in spoken sentence production: The role of grammatical units. Journal of Experimental Psychology: Learning, Memory, and Cognition, 33, 791–810. Allum, P. H., & Wheeldon, L. R. (2009). Scope of lexical access in spoken sentence production: Implications for the conceptual–syntactic interface. Journal of Experimental Psychology: Learning, Memory, and Cognition, 35, 1240–1255. Baddeley, A. D., & Hitch, G. J. (1974). Working memory. In G. A. Bower (Ed.) The psychology of learning and motivation, 8, 47-89. New York: Academic Press. Baluch, B., & Besner, D. (1991). Visual word recognition: Evidence for strategic control of lexical and nonlexical routines in oral reading. Journal of Experimental Psychology: Learning, Memory, and Cognition, 17, 644–652. Bartholow, B. D., Pearson, M. A., Dickter, C. L., Sher, K. J., Fabiani, M. & Gratton, G. (2005). Strategic control and medial frontal negativity: Beyond errors and response conflict. Psychophysiology, 42: 33–42. Bates, E., & MacWhinney, B. (1982). Functionalist approaches to grammar. In E. Wanner & L. R. Gleitman (Eds.) Language acquisition: The state of the art, pp. 173-218. Cambridge: Cambridge University Press. Belletti, A. & Rizzi, L. (1988). Psych-verbs and theta-theory. Natural Language and Linguistic Theory, 6: 291-352. Beilock, S. L., Bertenthal, B. I., McCoy , A. M., & Carr, T . H. (2004). Haste does not always make waste: Expertise, direction of attention, and speed versus accuracy in performing sensorimotor skills. Psychonomic Bulletin & Review , 11, 373–379. Blackmer, E. R. & Mitton, J. L. (1991). Theories of monitoring and the timing of repairs in spontaneous speech. Cognition, 39, 173-194. Boadi, L. A. (1975). Reciprocal verbs and symmetrical predication. Journal of West African Languages 10:55-77. Bock, J. K. (1982). Toward a cognitive psychology of syntax: Information processing contributions to sentence formulation. Psychological Review, 89, 1–47. 251 Bock, J. K. (1986a). Syntactic persistence in language production. Cognitive Psychology, 18, 355-387. Bock, J. K. (1986b). Meaning, sound, and syntax: Lexical priming in sentence production. Journal of Experimental Psychology: Learning, Memory and Cognition, 12, 575–586. Bock, J. K. (1987a). An effect of the accessibility of word forms on sentence structures. Journal of Memory and Language, 26, 119–137. Bock, J. K. (1987b). Coordinating words and syntax in speech plans. In A.W. Ellis (Ed.), Progress in the psychology of language (pp. 337–390). London: Erlbaum. Bock, J. K. (1995). Sentence production: From mind to mouth. In J. L. Miller, & P. D. Eimas (Eds.), Handbook of perception and cognition. Vol. 11: Speech, language, and communication, pp. 181-216. Orlando, FL: Academic Press. Bock, J. K., Irwin, D. E., Davidon, D. J. & Levelt, W. J. M. (2003). Minding the clock. Journal of Memory and Language, 48, 653–685. Bock, J. K., Irwin, D. E., & Davidson, D. J. (2004). Putting first things first. In J. M. Henderson & F. Ferreira (Eds.), The interface of language, vision, and action: Eye movements and the visual world (pp. 249–278). New York, NY: Psychology Press. Bock, J. K., & Levelt, W. J. (1994). Language production: Grammatical encoding. In M. A. Gernsbacher (Ed.), Handbook of psycholinguistics, pp. 945-984. San Diego, CA: Academic Press. Bock, J. K., Loebell, H., & Morey, R. (1992). From conceptual roles to structural relations: Bridging the syntactic cleft. Psychological Review, 99, 150–171. Bock, J. K. & Warren, R. K. (1985). Conceptual accessibility and syntactic structure in sentence formulation. Cognition, 21, 47–67. Boland, J. E., Tanenhaus, M. K, & Garnsey, S. M. (1990). Evidence for the immediate use of verb control information in sentence processing. Journal of Memory and Language, 29:413–432. Boland J. E., Tanenhaus, M. K, Garnsey, S. M, & Carlson, G. N. (1995). Verb argument structure in parsing and interpretation: Evidence from wh-questions. Journal of Memory and Language, 34:774–806. Boulware-Gooden, R., Carreker, S., Thornhill, A. & Malateshe, J. (2007). Instruction of metacognitive strategies enhances reading comprehension and vocabulary achievement of third-grade students. The Reading Teacher, 61(1), 70–77. 252 Brédart, S. (1991). Word-interruption in self-repairing. Journal of Psycholinguistic Research, 20, 123–138. Brysbaert, M., Fias, W., & Noel, M. P. (1998). The Whorfian hypothesis and numerical cognition: Is “twentyfour” processed in the same way as “four-and-twenty”? Cognition, 66, 51–77. Brown-Schmidt, S., & Konopka, A. E. (2008) Little houses and casas pequeñas: message formulation and syntactic form in unscripted speech with speakers of English and Spanish. Cognition, 109 (2), 274-280. Butterworth, B. (1980). Evidence from pauses in speech. In B. Butterworth (Ed.), Language production, Vol. 1, pp. 155–176. London: Academic Press. Caplan, D., & Waters, G. S. (1999). Verbal working memory and sentence comprehension. Behavioral and Brain Sciences, 22, 77-126. Carroll, J. B. (1958). Process and content in psycholinguistics. In Current trends in the description and analysis of behavior (pp. 175-200). Pittsburgh : University of Pittsburgh Press. Carter, R, & McCarthy, M. (1999). The English get-passives in spoken discourse: Description and implication for an interpersonal grammar. English Language and Linguisitics, 3, 41-58. Chao, Y.-R. (1968). A Grammar of Spoken Chinese, University of California Press, Berkeley. Chappell, H. (1980). Is the get passive adversative? Papers in Linguisitics, 13, 411-452. Cheung, C and R. Larson (2006). Chinese Psych-verbs & covert clausal complementation. Paper presented at the Chicago Workshop on Chinese Linguistics, Univ. of Chicago. Cheung, C and R. Larson (2007). Psych-verbs in Mandarin Chinese. Paper presented at the joint meetings of the International Assn. of Chinese Linguistics (IACL) and the North American Conference on Chinese Linguistics (NACCL), Columbia University. Christenfeld, N. (1994). Options and ums. Journal of Language and Social Psychology, 13, 192–199. Christianson, K. & Ferreira, F. (2005). Conceptual accessibility and sentence production in a free word order language (Odawa). Cognition, 98, 105-135. 253 Clark, H. H. (1965). Some structural properties of simple active and passive sentences. Journal of V erbal Learning and V erbal Behavior , 4, 365-370. Clark, H. H. (2002). Speaking in time. Speech Communication, 36, 5–13. Clark, H. H. & Clark, E. (1977). Psychology and Language. An Introduction to psycholinguistics. Harcourt, New York. Clark, H. H. & Wasow, T. (1998). Repeating words in spontaneous speech. Cognitive Psychology, 37, 201–242. Cowan, N. (1995). Attention and memory: An integrated framework. New York: Oxford University Press. Cowan, N. (2005). Working memory capacity. New York, NY: Psychology Press. Dahl, O. (1997). Egocentricity in discourse and syntax. http: // www.ling.su /staff/ oesten/ egocentric De Smedt, K. J. M. J. (1990). IPF: An incremental parallel formulator. In R. Dale, C. Mellish, & M. Zock (Eds.) Current research in natural language generation, pp.167– 192. London, UK: Academic Press. De Smedt, K. J. M. J. (1994). Parallelism in incremental sentence generation. In G. Adriaens & U. Hahn (Eds.), Parallel natural language processing. Norwood, NJ: Ablex D’Esposito, M., Postle, B.R., Ballard, D., & Lease, J. (1999). Maintenance versus manipulation of information held in working memory: An event-related fMRI study. Brain and Cognition, 41, 66-86. Dell, G.S. (1985). Positive feedback in hierarchical connectionist models: Applications to language production. Cognitive Science. 9:3-23 Dell, G. S. (1986). A spreading activation theory of retrieval in sentence production. Psychological Review, 93, 124–142. Dell, G. S. (1990). Effects of frequency and vocabulary type on phonological speech errors. Language and Cognitive Processes, 5, 313–349. Dell, G. S., Burger, L. K., & Svec, W . R. (1997). Language production and serial order: A functional analysis and a model. Psychological Review , 104, 123–147. 254 Dell, G. S. & O'Seaghdha, P. G . (1992). Stages of lexical access in language production. Cognition, 42, 287-314. Dell, G. S., & O'Seaghdha, P. G. (1994). Inhibition in interactive activation models of linguistic selection and sequencing. In D. Dagenbach & T. H. Carr (Eds.). Inhibitory processes in attention, memory, and language (pp. 409-451). San Diego: Academic Press. Dimitriadis, A. (2008). The event structure of irreducibly symmetric reciprocals. In J. Dölling & T. Heyde-Zybatow (Eds). Event structures in linguistic form and interpretation, 327-34. Berlin: Mouton de Gruyter, Dowty, D. (1991). Thematic proto-roles and argument selection, Language 67:547-619. Feldman, J. A. & Ballard, D. H. (1982). Connectionist models and their properties. Cognitive Science, 6, 205-254. Ferreira, F. (1994). Choice of passive voice is affected by verb type and animacy. Journal of Memory and Language, 33, 715-736. Ferreira, F. (2000). Syntax in language production: An approach using tree-adjoining grammars. In L. Wheeldon (Ed.), Aspects of Language Production. Cambridge, MA: MIT Press. Ferreira, F. & Engelhardt, P. (2006). Syntax and production. In M. A. Gernsbacher & M. Traxler (Eds.), Handbook of Psycholinguistics, pp. 61-91. Oxford, UK: Elsevier Inc. Ferreira, F., & Swets, B. (2002). How incremental is language production? Evidence from the production of utterances requiring the computation of arthimetic sums. Journal of Memory and Language, 46, 57-84. Ferreira, F., & Swets, B. (2005). The production and comprehension of presumptive pronouns in relative clause ‘‘island’’ contexts. In Cutler (Ed.), Twenty-first century psycholinguistics: Four cornerstones (pp. 263–278). Mahwah, New Jersey: Lawrence Erlbaum Associates. Ferreira, V. S. (1996). Is it better to give than to donate? The consequences of syntactic flexibility in language production. Journal of Memory and Language, 35, 724–755. Ferreira, V. S. (2003). The persistence of optional complementizer mention: Why saying a “that” is not saying “that” at all. Journal of Memory and Language, 48, 379-398. Ferreira, V. S. & Bock, K. (2006). The functions of structural priming. Language & Cognitive Processes, 21, 1011-1029. 255 Ferreira, V. S. & Dell, G. S. (2000). The effect of ambiguity and lexical availability on syntactic and lexical production. Cognitive Psychology, 40, 296-340. Ferreira V. S. & L. R. Slevc. (2007).Grammatical encoding. In Gareth Gaskell (Ed.), The Oxford Handbook of Psycholinguistics, pp.453-469. Oxford: Oxford University Press. Fisher, C. (2002). The role of abstract syntactic knowledge in language acquisition: A reply to Tomasello (2000). Cognition, 82, 259–278. Foley W. A. & van Valen, R. D. (1984). Functional Syntax and Universal Grammar. Cambridge: Cambridge University Press. Fromkin,V. A. (1971). The non-anomalous nature of anomalous utterances. Language, 47, 27–52. Fromkin, V.A. (1973). Speech Errors as Linguistic Evidence. The Hague, Netherlands: Mouton. Garnham, A. (1997). Mental Models as Representations of Discourse and Text. Chichester: Ellis Horwood. Garnham, A. (2001). Mental Models and the Interpretation of Anaphora. Hove: Psychology Press. Garnsey, S. M., Pearlmutter, N. J., Myers, E., & Lotocky, M. A. (1997). The contributions of verb bias and plausibility to the comprehension of temporarily ambiguous sentences. Journal of Memory and Language, 37, 58–93. Gleitman, L. R., January, D., Nappa, R., & Trueswell, J. C. (2007). On the give and take between event apprehension and utterance formulation. Journal of Memory and Language, 57, 544–569. Garrett, M. F. (1975). The analysis of sentence production. In G. H. Bower (ed.), The psychology of learning and motivation, 9, pp.133-177. New York: Academic Press. Garrett, M. F . (1980). Levels of processing in sentence production. In B. Butterworth (Ed.), Language production: Speech and talk, pp.177–220. London: Academic. Garrett, M. (1988). Processes in sentence production. In F. J. Newmeyer (Ed.), Linguistics: The Cambridge survey, Vol. 3: Language: Psychological and Biological Aspects, pp.69-96. Cambridge: Cambridge University Press. 256 Givón, T. & Yang, L. (1994). The rise of the English get-passives. In B. Fox & P. J. Hopper (Eds.) V oice: Form and function, pp.119-149. Amsterdam: John Benjamins. Goldman-Eisler, F. (1968). Psycholinguistics: Experiments in spontaneous speech. New York: Academic Press. Gordon, P., Hendrick, R., & Johnson, M. (2001). Memory inference during language processing. Journal of Experimental Psychology: Learning, Memory, and Cognition, 27(6), 1411-1423. Gordon, P., Hendrick, R., & Levine, W. (2002). Memory-load interference in syntactic processing. Psychological Science, 13(5), 425-430. Gratton, G., Coles, M. G. H. & Donchin, E. (1992). Optimizing the use of information: Strategic control of activation of responses. Journal of Experimental Psychology: General, 121 (4), 480-506. Griffin, Z. M. (1998). What the eye says about sentence planning. Unpublished doctoral dissertation. Graduate College of the University of Illinois at Urbana-Champaign. Griffin, Z. M. (2001). Gaze durations during speech reflect word selection and phonological encoding. Cognition, 82, B1–B14. Griffin, Z. M. (2003). A reversed length effect in coordinating the preparation and articulation of words in speaking. Psychonomic Bulletin & Review, 10, 603–609. Griffin, Z. M. & Bock, K. (2000). What the eyes say about speaking. Psychological Science 11, 274–279. Griffin, Z. M., & Spieler, D. H. (2006). Observing the what and when of language production for different age groups by monitoring speakers’ eye movements. Brain and Language, 99, 272–288. Grimshaw, J. (1990). Argument Structure. Cambridge, Mass.: MIT Press. Gronlund, S. D. & Shiffrin, R. M. (1986). Retrieval strategies in recall of natural categories and categorized-lists. Journal of Experimental Psychology: Learning, Memory, and Cognition, 12, 550-561. Goldsmith, M., Koriat, A. & Pansky, A. (2005). Strategic regulation of grain size in memory reporting over time. Journal of Memory and Language, 52, 505-525. 257 Goldsmith, M., Koriat, A. & Weinberg-Eliezer, A. (2002). Strategic regulation of grain size memory reporting. Journal of Experimental Psychology: General, 131, 73-95. Gruber, J. S. (1967). Look and see. Language, 43, 937-947. Gruber, J. S. (1976). Lexical structures in syntax and semantics. Amsterdam: North- Holland. Henderson, A. I. (1974). Time patterns in spontaneous speech-cognitive stride or random walk? A reply to Jaffe et al. (1972). Language and Speech, 17, 119–125. Hinton, G. E. (1981). Implementing semantic networks in parallel hardware. In G. E. Hinton & J. A. Anderson (Eds.), Parallel models of associative memory. Hillsdale, NJ, Erlbaum. Hinton, G. E., McClelland, J. L., & Rumelhart, D. E. (1986). Distributed representations. In D. E. Rumelhart, J. L. McClelland, and the PDP Research Group, Parallel distributed processing: Explorations in the microstructure of cognition. Volume 1: Foundations. Cambridge, MA: Bradford Books/MIT Press. Habel, C. & Tappe, H. (1999). Process of segmentation and linearization in describing events. In R. Klabunde and C. von Stutterheim (eds.), Conceptual and Semantic Knowledge in Language Production, 117–152. Opladen: Westdeutscher Verlag. Hartsuiker, R. & Barkhuysen, P. (2006). Language production and working memory: The case of subject-verb agreement. Language and Cognitive Processes, 21, 181-204. Holmes, V. M. (1988). Hesitations in sentence planning. Language and Cognitive Processes, 3, 323–361. Hopper, P. J. & Thompson, S. A. (1980). Transitivity in grammar and discourse. Language, 56, 251-299. Hsu, D.-B. (2008). Structural persistence in Mandarin Chinese Preschoolers. In.M. Chan & H. Kang (eds.), Proceedings of the 20th North American Conference on Chinese Linguistics, pp.913-926. The Ohio State University, Columbus, OH. Hsueh, F.-S. (1987). The Ba Construction in Mandarin: an Integrated Interpretation. Language Teaching and Linguistic Studies 1, 4-22. Huang, C.-T.(1982). Logical Relations in Chinese and the Theory of Grammar, Ph.D. dissertation, MIT, Cambridge, Mass. 258 Huang, C.- T., Li, Y.- H, & Li, Y.-F. (2009). The Syntax of Chinese. Cambridge: Cambridge University Press. Huang, S.-F. (1974). Mandarin Causatives. Journal of Chinese Linguistics 2, 355-369. Jackendoff, R. (1972). Semantic representation in generative grammar. Cambridge, Mass.: MIT Press. Jackendoff, R. (1987). The status of thematic relations in linguistic theory. Linguistic Inquiry 18, 369-412. Jackendoff, R. (1990). Semantic Structures. Cambridge, Mass.: MIT Press. James, C. T., Thompson, J. G. & Baldwin, J. M. (1973). The reconstructive process in sentence memory. Journal of V erbal Learning and V erbal Behavior , 12, 51-63. Jared, D. (1997). Evidence that strategy effects in word naming reflect changes in output timing rather than changes in processing route. Journal of Experimental Psychology: Learning, Memory, and Cognition, 23, 1424–1438. Jarrod, M., Schunn, C.D., Schneider, W., McNamara, D. S. & VanLehn, K. (2011). The neural correlates of strategic reading comprehension: Cognitive control and discourse comprehension, NeuroImage, 58 (2), 675-686. Kello, C. T. & Plaut, D. C. (2000). Strategic control in word reading: Evidence from speeded responding in the tempo naming task. Journal of Experimental Psychology: Learning, Memory, & Cognition, 26, 719–750. Kello,C. T. & Plaut, D. C. (2003). Strategic control over rate of processing in word reading: A computational investigation. Journal of Memory and Language, 48, 207– 232 Kempen, G., & Hoenkamp, E. (1987). An incremental procedural grammar for sentence formulation. Cognitive Science, 11, 201–258. Kemper, S. (1992). Adults’ sentence fragments: who, what, when, where, and why? Communication Research, 19, 444–458. Konopka, A. E. (2012). Planning ahead: How recent experience with structures and words changes the scope of linguistic planning. Journal of Memory and Language, 66 (1), 143-162. 259 Konopka, A. E., & Bock, K. (2009). Lexical or syntactic control of sentence formulation? Structural generalizations from idiom production. Cognitive Psychology, 58, 68–101. Kowall, S., O’Connell, D. C. & Sabin, E. J. (1975). Temporal patterning and vocal hesitations in spontaneous narratives. Journal of Psycholinguistic Research, 4, 195–204. Lakoff, G. (1968). Instrumental adverbs and the concept of deep structure. Foundations of Language, 4, 4-29. Levelt, W. J. M. (1983). Monitoring in self-repair in speech. Cognition, 14, 41-105. Levelt, W. J. M. (1989). Speaking: From Intention to Articulation. Cambridge, Mass.: MIT Press. Levelt, W. J. M. & Meyer, A. S. (2000). Word for word: Multiple lexical access in speech production. European Journal of Cognitive Psychology, 12, 433–452. Levin, B. (1993). English verb classes and alternations: A preliminary investigation. Chicago, IL: University of Chicago Press. Li, A. Y.-H. (1990). Order and Constituency in Mandarin Chinese. Kluwer Academic Publishers, Dordrecht. Liu, F.-H. (1997). An Aspectual Analysis of BA. Journal of East Asian Linguistics, 6, 51-99. Lyons, J. (1977). Semantics. Cambridge: Cambridge University Press. MacKay, D. G. (1970). Spoonerisms: The structure of errors in the serial order of speech. Neuropsychologia, 8, 323–350. MacKay, D. G. (1972). The structure of words and syllables: Evidence from errors in speech. Cognitive Psychology, 3, 210–227. MacKay, D. G. (1987). The organization of perception and action: a theory for language and other cognitive skills. New York: Springer. MacKay, D. G. (1992a). Errors, ambiguity and awareness in language perception and production. In B.J. Baars (Ed.), Experimental slips and human error: Exploring the architecture of volition. New York: Plenum Press. MacKay, D. G. (1992b). Awareness and error detection: New theories and research paradigms. Consciousness and Cognition, 1, 199-225. 260 MacWhinney, B. (1977). Starting points. Language, 53, 152-168. Magliano, J. P., Trabasso, T. & Graesser, A. C. (1999). Strategic processing during comprehension. Journal of Educational Psychology, 91(4), 615-629. Martin, R. C., Crowther, J. E., Knight, M., Tamborello, F. P. & Yang, C. (2010). Planning in sentence production: Evidence for the phrase as a default planning scope. Cognition, 116, 177–192. McClelland, J. L. & Rumelhart, D. E. (1985). Distributed memory and the representation of general and specific information. Journal of Experimental Psychology: General, 114, 159-188. McClelland, J. L., Rumelhart, D. E., & Hinton, G. E. (1986). The appeal of parallel distributed processing. In D. E. Rumelhart, J. L. McClelland, and the PDP Research Group, Parallel distributed processing: Explorations in the microstructure of cognition. Volume 1: Foundations. Cambridge, MA: Bradford Books/MIT Press. Mei, K. (1978). Ba Zi Ju [Ba Sentences]. Wenshizhe Xuebao , 27, 145-180. Meiran, N. (1996). Reconfiguration of processing mode prior to task performance. Journal of Experimental Psychology: Learning, Memory, and Cognition, 22(6), 1423- 1442. Meyer, A. S. (1991). The time course of phonological encoding in language production: Phonological encoding inside a syllable. Journal of Memory and Language, 30, 69–89. Meyer, A. S. (1996). Lexical access in phrase and sentence production: Results from picture word interference experiments. Journal of Memory & Language, 35, 477–496. Meyer, A. S. (1997). Conceptual influences on grammatical planning units. Language and Cognitive Processes, 12, 859–863. Meyer, A. S., Roelofs, A., & Levelt, W. J. M. (2003). Word length effects in object naming: The role of a response criterion. Journal of Memory and Language, 48, 131– 147. Miller, G.A. (1956). The magic number seven plus or minus two: some limits on our capacity to process information. Psychological Review 63 (2): 81–97. Oomen, C. C. E. Postma, A. (2001) Effects of Time Pressure on Mechanisms of Speech Production and Self-Monitoring. Journal of Psycholinguistic Research, 3, 163-184. 261 Palmer, C., & Pfordresher, P . Q. (2003). Incremental planning in sequence production. Psychological Review , 110,683–712. Palmer, C., & Schendel, Z. A. (2002). W orking memory constraints in sequence production. Abstracts of the Psychonomic Society, 7, 30. Pesetsky, D. (1995). Zero-Syntax: Experiences and Cascades. Cambridge, Mass.: MIT press. Plamondon, R., & Alimi, A. M. (1997). Speed/accuracy trade-offs in target-directed movements. Behavioral and Brain Sciences, 20, 279–349. Polinsky, M. (1996). Situation perspective: on the relations of thematic roles, discourse categories, and grammatical relations to Figure and Ground. In A. Goldberg (Ed.), Conceptual structure, discourse and language, 401-419. CLSI Publications. Postma, A. (2000). Detection of errors during speech production: a review of speech monitoring models. Cognition, 77, 97-131. Postma, A. & Kolk, H. H. J. (1993). The covert repair hypothesis: prearticulatory repair processes in normal and stuttered disfluencies. Journal of Speech and Hearing Research, 36, 472–487. Postma, A., Kolk, H. & Povel, D. J. (1990). On the relation among speech errors, disfluencies, and self-repairs. Language and Speech, 33, 19–29. Prat-Sala, M., & Branigan, H. P. (2000). Discourse constraints on syntactic processing in language production: A cross-linguistic study in English and Spanish. Journal of Memory and Language, 42, 168-182. Reder, L. M. (1982). Plausibility judgments versus fact retrieval: Alternative strategies for sentence verification. Psychological Review, 89, 250-280. Reder, L. M. (1987). Strategy selection in question answering. Cognitive Psychology, 19, 90-138. Reder, L. M. & Ross, B. H. (1983). Integrated knowledge in different tasks: The role of retrieval strategy on fan effects. Journal of Experimental Psychology: Learning, Memory, and Cognition, 9, 55-72. Schlenck, K. J., Huber, W., & Willmes, K. (1987). Prepairs and repairs: different monitoring functions in aphasic language production. Brain and Language, 30, 226– 244. 262 Schmidt, R. A., Zelaznik, H. N., Hawkins, B., Frank, J., & Quinn, J. T . (1979). Motor- output variability: A theory for the accuracy of rapid motor acts. Psychological Review , 86, 415–451. Schriefers, H. & Teruel, E. (1999). Phonological facilitation in the production of two- word utterances. European Journal of Cognitive Psychology, 11, 17–50. Shapiro, L. P ., Brookins, B., Gordon, B., & Nagel, N. (1991). Verb effects during sentence processing. Journal of Experimental Psychology, 17:983–996. Shintani, M. (1979). The frequency and usage of the English passive. Unpublished doctoral dissertation, University of California, Los Angeles. Smith, M., & Wheeldon, L. (1999). High level processing scope in spoken sentence production. Cognition, 73:205-246. Smith, M., & Wheeldon, L. (2004). Horizontal information flow in spoken sentence production. Journal of Experimental Psychology: Learning, Memory, and Cognition, 30, 675–686. Stallings L. M., MacDonald, M. C., & O’Seaghdha P. G. (1998) Phrasal ordering constraints in sentence production: phrase length and verb disposition in Heavy-NP Shift. Journal of Memory & Language, 39, 392–417. Svartvik, J. (1966). On V oice in the English V erb. The Hague: Mouton. Svartvik, J. & Randolph, Q. (1980). A corpus of English conversation. Lund studies in English 56. Lund: LiberLa¨romedel/Gleerup. Swets, B., Jacovina, M. E., & Gerrig, R. J. (2008). Individual differences in the planning scope of language production. Paper presented at the 49st meeting of the Psychonomic Society, Chicago, IL (November). Talmy, L. (2000). Towards a cognitive semantics. Vol. I: concepts structuring systems. Cambridge, Mass.: MIT Press. Tomasello, M. (2000). Do young children have adult syntactic competence? Cognition, 74, 209–253. Tomlin, R. S. (1983). On the interaction of syntactic subject, thematic information and agent in English. Journal of Pragmatics, 7, 411-432. Van Nice, K. Y., & Dietrich, R. (2003). Task-sensitivity of animacy effects: Evidence from German picture descriptions. Linguistics, 41, 5, 825-849. 263 Wagner,V., Jescheniak, J. D., & Schriefers, H. (2010) On the Flexibility of Grammatical Advance Planning During Sentence Production: Effects of Cognitive Load on Multiple Lexical Access. Journal of Experimental Psychology: Learning, Memory, and Cognition, 36, 2, 423-440. Winer, B. J. (1971). Statistical principles in experimental design. New York: McGraw– Hill. Wheeldon, L. & Lahiri, A. (1997). Prosodic units in speech production. Journal of Memory and Language, 37, 356–381. Wheeldon, L. R., Meyer, A. S., & Smith, M. (2002). Incrementality . In R. Goldstone (Ed.), Encyclopedia of cognitive science. London: Macmillan. Wickelgren, W . (1977). Speed-accuracy tradeoff and information processing dynamics. Acta Psychologia, 41, 67–85. W oodworth, R. S. (1899). The accuracy of voluntary movement. Psychological Monographs, 3. Zhang, B.-J. (2001). Symmetries and Asymmetries between Bei construction and Ba construction. Chungguo yuwen vol.6. 264 Appendix A: Target Items for Experiment 1 In each target item, the first and the second noun phrases belong to the inanimate-animate noun pair. The second and the third noun phrases belong to the animate-animate noun pair. The fourth one is a verbal phrase containing an agent-theme verb and a complement. 1. 新電影/ 名導演/女明星/ 批評了一頓 New movie/ famous director/ actress/ criticize 2. 垃圾桶/ 管理員 /售票員/ 踢飛了出去 trash can/ janitor/ booking clerk/ kick away 3. 名畫/ 畫家/ 作家/ 護送了一程 painting/ painter/ writer/ escort 4. 沙發/ 客人/ 主人/ 刮傷了一道 sofa/ guest/ host/ scratch 5. 法案/立委/ 記者/ 稱讚了一番 Proposition / legislator/ journalist/ praise 6. 文件/ 職員/ 秘書/ 檢查過兩遍 document/ office employee/ secretary / examine twice 7. 證據/ 法官/ 律師/ 徹底地忽略 evidence/ judge/ lawyer/ ignore completely 8. 照相機/ 攝影師 /剪接師 / 緊抓住不放 camera/ photographer/ film editor/ grasp tightly 9. 化合物/ 化學家 / 數學家/ 分析得很透徹 chemical compound/chemist/ mathematician/ analyze thoroughly 10. 玩偶/ 女孩/ 男孩/ 緊抱在懷裡 stuffed toy/ girl/ boy/ hold in the arms 11. 果樹/ 農夫/ 詩人/ 照顧得很好 fruit tree/ farmer/ poet/ take good care 12. 人物像/ 油畫家/ 漫畫家/ 畫得很逼真 portrait/ oil painter/ cartoonist/ portray vividly 13. 薪水/ 會計/ 顧問/ 控制得很嚴 salary/ accountant/ counselor/ control strictly 14. 消息/ 將軍/ 部長/ 出賣給敵人 news / general/ minister/ sell out to enemy 15. 流行歌/ 聲樂家/ 藝術家/ 攻擊了一頓 pop song/ vocalist/ artist/ attack 16. 名產/ 市長/ 校長/ 恭維了一番 specialty / mayor/ principal/ flatter 17. 車子/ 教練 / 教官/ 留在自己的家 car/ coach/ instructor/ keep in his home 265 18. 論文集/ 文學家 /史學家 / 送到學校去 essay collection/ literature scholar/ historian/ send to school 19. 婚外情/ 魔術師/ 提琴手/ 藏在回憶裡 love affair/ magician/ violinist/ store in memory 20. 童年/ 老師/ 學生/講得很特別 childhood/ teacher/ student/ describe in a special way 21. 聖經/ 牧師/神父/ 當作好朋友 Bible/ pastor/ priest/ view as a good friend 22. 點菜單/ 服務生/ 消防員/ 置之不理 order list / waiter/ firefighter/ leave aside 23. 酒瓶/ 水手/ 園丁 /帶到大門外 wine bottle/ sailor/ gardener/ bring to the door 24. 牙齒 /牙醫/ 獸醫/ 保護得很好 teeth/ dentist/ vet/ protect very well 25. 食譜/ 廚師/ 司機./ 忘得一乾二淨 recipe / chef/ driver/ forget totally 26. 大門/ 木匠/ 技師/ 推開一公尺 gate/ carpenter/ mechanic/ push one meter away 27. 書桌 /工程師/ 售貨員/ 猛打了兩拳 computer/ engineer/ salesman/ hit fiercely 28. 謠言/ 清潔工/ 水電工/ 傳得很難聽 rumor/ cleaner/ plumber/ spread disgracefully 29. 洋娃娃/ 美髮師 / 化妝師 /打扮得很美 doll / hairdresser/ cosmetician/ dress up beautifully 30. 老房子/建築師/設計師/ 改頭換面 old house/ architect/ designer/ change the appearance 266 Appendix B: Target Items for Experiment 2 In each target item, the first and the second noun phrases belong to the inanimate-animate noun pair. The second and the third noun phrases belong to the animate-animate noun pair. The fourth one is a verbal phrase containing a theme-experiencer verb and a complement. 1. 車禍/ 作家/ 畫家/ 嚇得驚慌失措 car accident/ writer/ painter/ frighten 2. 開發案/ 建築師/ 設計師/ 搞得身心俱疲 developmental project/ architect/ designer/ exhaust 3. 電影/女孩/男孩/ 感動得痛哭流涕 movie/ girl/ boy/ move to tears 4. 喜劇/ 客人/ 主人/ 逗得笑哈哈 comedy/ guest/ host/ amuse 5. 大塞車/ 消防員/ 救生員/ 氣得火冒三丈 traffic jam/ firefighter/ lifesaver/ enrage 6. 緋聞/ 立委/ 議員/ 惹得身敗名裂 love affair/ legislator/ councilor/ jeopardize 7. 破產/教練/ 選手/ 搞得六神無主 bankruptcy/ coach/ player/ stupefy 8. 壹週刊 /聲樂家 /鋼琴家/ 得罪過很多次 weekly journal /vocalist/ pianist/ offend many times 9. 簽名照/女歌迷/男歌手/樂得心花怒放 autographed photo/ female fan/ male singer/ elate 10. 酷刑/ 將軍/ 部長/ 折磨得很慘 cruel torture/ general/ minister/ torment 11. 官司/ 牙醫/ 獸醫/ 鬧得心神不寧 legal suit/ dentist/ vet/ perturb 12. 考試/ 學生/ 家長/ 煩得食不知味 exam/ student/ parent/ annoy 13. 火車聲/ 文學家/ 史學家/ 吵得睡不著覺 train noise/ literature scholar/ historian/ disturb 14. 老房子/ 水泥工/ 油漆工/ 累得人仰馬翻 old house/ plasterer/ painter/ fatigue 15. 戰爭/ 園丁/ 水手/ 害得家破人亡 war/ gardener/ sailor/ ruin 16. 糖尿病/ 惡婆婆 /壞媳婦/ 整得七葷八素 diabetes/ mean mother-in-law/ bad daughter-in-law/ abuse 17. 名牌包/ 日本人/韓國人/迷得神魂顛倒 designer bag/ Japanese/ Korean/ fascinate 267 18. 假情報/名偵探/情報員/ 騙得團團轉 fake intelligence/ famous detective/ spy/ bamboozle 268 Appendix C: Target Items for Experiment 3, 4 and 5 (i) Agent-theme verbs 1. actor/actress/ assist 2. architect/ bartender/attack 3. athlete/dancer/escort 4. author/baker/punish 5. banker/lawyer/carry 6. librarian/novelist/caress 7. supervisor/scriptwriter/ chase 8. dentist/surgeon/murder 9. teacher/pilot/guide 10. politician/photographer/push 11. florist/tourist/free 12. pianist/journalist/touch (ii) Experiencer-theme verbs 1. barber/farmer/admire 2. sailor/butler/adore 3. clergyman/detective/cherish 4. correspondent/mathematician/ detest 5. lifeguard/mailman/despise 6. hairdresser/manager/distrust 7. fisherman/gardener/resent 8. sheriff/soldier/respect 9. scientist/singer/dread 10. student/plumber/hate 11. instructor/investor/fear 12. principal/professor/like (iii) Theme-experiencer verbs 1. publisher/reviewer/amuse 2. reporter/performer/appall 3. stockbroker/therapist/anger 4. chef/nurse/annoy 5. busboy/cowboy/baffle 6. cashier/trainer/confuse 7. captain/doctor/convince 8. butcher/painter/disturb 9. tailor/chemist/soothe 10. conductor/director/scare 269 11. examiner/missionary/shock 12. babysitter/interpreter/ thrill

Abstract (if available)

Abstract The study of sentence production investigates how speakers turn preverbal messages into utterances. Researchers generally characterize the language production system into three broad components: the message component (conceptualization), the grammatical component (formulation), and the phonological component (articulation) (Bock 1995

Linked assets

University of Southern California Dissertations and Theses

Conceptually similar

PDF

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

PDF

Factors affecting relative clause processing in Mandarin

PDF

Investigating coordination of lexical and structural information cross-linguistically

PDF

Narrowing the focus: experimental studies on exhaustivity and contrast

PDF

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

PDF

Existential constructions: a syntactic predication approach

PDF

Number marking and definiteness in Bangla

PDF

Vietnamese pronouns in discourse

PDF

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

PDF

Processing the dynamicity of events in language

PDF

Acquisition of functional categories

PDF

Investigating the production and perception of reduced speech: a cross-linguistic look at articulatory coproduction and compensation for coarticulation

PDF

A unified syntactic account of Mandarin subject nominals

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

Prosody and informativity: a cross-linguistic investigation

PDF

The speaker's choice of using object-marking in informal spoken Korean

PDF

Dominant language influence in acquisition and attrition of the Chinese reflexive ziji By Chinese-English bilinguals

PDF

Building adjectival meaning without adjectives

PDF

Interpretation of pronominal forms across languages

PDF

Schema architecture for language-vision interactions: a computational cognitive neuroscience model of language use

Asset Metadata

Creator Huang, Yu-Chi (author)

Core Title Flexibility in language production

School College of Letters, Arts and Sciences

Degree Doctor of Philosophy

Degree Program Linguistics

Publication Date 08/03/2012

Defense Date 04/24/2012

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

Tag English,Mandarin Chinese,OAI-PMH Harvest,semantic flexibility,sentence production,strategic control,thematic prominence

Language English

Contributor Electronically uploaded by the author (provenance)

Advisor Kaiser, Elsi (committee chair), Li, Audrey (committee member), Simpson, Andrew (committee member), Tjan, Bosco S. (committee member)

Creator Email steph111tw@yahoo.com.tw,yuchih@usc.edu

Permanent Link (DOI) https://doi.org/10.25549/usctheses-c3-87502

Unique identifier UC11289449

Identifier usctheses-c3-87502 (legacy record id)

Legacy Identifier etd-HuangYuChi-1130.pdf

Dmrecord 87502

Document Type Dissertation

Rights Huang, Yu-Chi

Type texts

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

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

Repository Name University of Southern California Digital Library

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