Natural languages are systems of forms and meanings; language understanding and language production are processes of establishing mappings between linguistic forms and meanings. The principles and rules governing the mapping between semantic roles and syntactic positions have long been a fundamental topic in contemporary linguistics. Such a mapping is usually called argument realization, argument mapping or argument linking. On the basis of the previous language specific and cross linguistic researches on this issue, this thesis picks out two tasks. One is the empirical task of the investigating the principles and rules governing the mapping between semantic roles and linear syntactic positions in Mandarin Chinese. The other is the theoretical task of the exploration of how argument realization principles and rules play their roles in the live temporal linear comprehension and production of sentences. On the empirical side, this thesis mainly investigates the phenomenon of argument alternation, that is, the non-one-to-one mapping between semantic roles and syntactic forms (linear positions) in Mandarin and argues that alternative syntactic forms in which semantic roles are realized are not arbitrary but semantically motivated. More specifically, it proposes that alternative patterns of argument realization encode different types of events. This thesis concentrates on three major cases of argument alternation. The first is the argument alternation in the resultative verb construction (RVC) that involves two verbs and expresses a complex event consisting of a first (activity) subevent and a second (resultative) subevent. The arguments of the two verbs are mapped onto the subject and the object alternatively and the argument sharing between the verbs results from syntactically constrained pragmatic inference. The argument realization principles and rules of RVC are used to account for two puzzling cases of argument alternation in Mandarin, i.e. the locative alternation and the agentive alternation. This account of inverse argument realizations has the implication that argument alternations are semantically motivated rather than the result of arbitrary syntactic operation. To facilitate the discussion of how different semantic representations arise in different process of comprehension, I adopt Dynamic Syntax (Kempson et al 2001; Cann et al 2005) which provides a package of working hypotheses about human language grammars and the formal tools for representing how grammars work. It is hypothesized in Dynamic Syntax that the grammar of a natural language is a set of constraints over language comprehension; sentences are understood and produced in context through left-to-right word-by-word parsing processes. Parsing processes are driven by the axiomatic requirement of establishing complete logical forms that can be enriched to full propositions. Such processes have the characteristic of semantic underspecification, including underspecified semantic relationships and underspecified semantic contents; semantic underspecification can and must be updated through non-demonstrative inference implemented in linguistic and nonlinguistic contexts. Using the framework I hypothesize that in RVC constructions the first verb provides a condition on the sort of event expressed by the second verb, encoding this in terms of event semantics. It is argued that only the argument of the latter are required to be realized in the string (or be contextually strongly determined) through pro-drop. Those of the activity predicate, however, are inferred through pragmatic means given the arguments that are realized. This directly accounts for the attested patterns of argument realization in RVC and explain the apparent gaps. This analysis is extended to locative and agentive inversion constructions where it is hypothesized that there is null resultative predicate that explains why a non-agent can be realized as subject, even in the presence of a more agentive noun phrase in the string. This thesis thus maintains the hypothesis that the mapping between semantic roles and syntactic positions is direct though not one-to-one. Although there is no one-toone mapping between syntactic forms, the argument mapping rules can ensure efficient comprehension and production when they are applied in context. This thesis provides a uniform account of different argument alternation phenomena that have been seen as unrelated to each other in the literature. The successful uniform explanation of the ‘unrelated’ phenomena of argument alternation can be generalized as a methodology: a thorough semantic analysis of various alternative syntactic constructions can reveal the subtle semantic differences between them and the importance of these subtle semantic difference for a theorectic account of argumenty alternation has been largely underestimated in the literature. This constitutes the foundation of a uniform explanation of syntactic phenomena that seem to be unrelated to each other. This success lights the hope of seeking semantics-based uniform accounts of other different kinds of syntactic phenomena in a single language and across languages in future research.

La linguistique informatique est un domaine de recherche qui se concentre sur les méthodes et les perspectives de la modélisation formelle (statistique ou symbolique) de la langue naturelle. La linguistique informatique, tout comme la linguistique théorique, est une discipline fortement modulaire : les niveaux d'analyse linguistique comprennent la segmentation, l'analyse morphologique, la désambiguïsation, l'analyse syntaxique et sémantique. Tandis qu'un nombre d'outils existent déjà pour les traitements de bas niveau (analyse morphologique, étiquetage grammatical), le hongrois peut être considéré comme une langue peu doté pour l'analyse syntaxique et sémantique. Le travail décrit dans la présente thèse vise à combler ce manque en créant des ressources pour le traitement syntaxique du hongrois : notamment, un analyseur en chunks et une base de données lexicale de schémas de sous-catégorisation verbale. La première partie de la recherche présentée ici se concentre sur la création d'un analyseur syntaxique de surface (ou analyseur en chunks) pour le hongrois. La sortie de l'analyseur de surface est conçue pour servir d'entrée pour un traitement ultérieur visant à annoter les relations de dépendance entre le prédicat et ses compléments essentiels et circonstanciels. L'analyseur profond est mis en œuvre dans NooJ (Silberztein, 2004) en tant qu'une cascade de grammaires. Le deuxième objectif de recherche était de proposer une représentation lexicale pour la structure argumentale en hongrois. Cette représentation doit pouvoir gérer la vaste gamme de phénomènes qui échappent à la dichotomie traditionnelle entre un complément essentiel et un circonstanciel (p. ex. des structures partiellement productives, des écarts entre la prédictibilité syntaxique et sémantique). Nous avons eu recours à des résultats de la recherche récente sur la réalisation d'arguments et choisi un cadre qui répond à nos critères et qui est adaptable à une langue non-configurationnelle. Nous avons utilisé la classification sémantique de Levin (1993) comme modèle. Nous avons adapté les notions relatives à cette classification, à savoir celle de la composante sémantique et celle de l'alternance syntaxique, ainsi que la méthodologie d'explorer et de décrire le comportement des prédicats à l'aide de cette représentation, à la tâche de construire une représentation lexicale des verbes dans une langue non-configurationnelle. La première étape consistait à définir les règles de codage et de construire un vaste base de données lexicale pour les verbes et leurs compléments. Par la suite, nous avons entrepris deux expériences pour l'enrichissement de ce lexique avec des informations sémantiques lexicales afin de formaliser des généralisations syntaxiques et sémantiques pertinentes sur les classes de prédicats sous-jacentes. La première approche que nous avons testée consistait en une élaboration manuelle de classification de verbes en fonction de leur structure de compléments et de l'attribution de rôles sémantiques à ces compléments. Nous avons cherché la réponse aux questions suivantes: quelles sont les composants sémantiques pertinents pour définir une classification sémantique des prédicats hongrois? Quelles sont les implications syntaxiques spécifiques à ces classes? Et, plus généralement, quelle est la nature des alternances spécifiques aux classes verbales en hongrois ? Dans la phase finale de la recherche, nous avons étudié le potentiel de l'acquisition automatique pour extraire des classes de verbes à partir de corpus. Nous avons effectué une classification non supervisée, basée sur des données distributionnelles, pour obtenir une classification sémantique pertinente des verbes hongrois. Nous avons également testé la méthode de classification non supervisée sur des données françaises. / We present the creation of two resources for Hungarian NLP applications: a rule-based shallow parser and a database of verbal subcategorization frames. Hungarian, as a non-configurational language with a rich morphology, presents specific challenges for NLP at the level of morphological and syntactic processing. While efficient and precise morphological analyzers are already available, Hungarian is under-resourced with respect to syntactic analysis. Our work aimed at overcoming this problem by providing resources for syntactic processing. Hungarian language is characterized by a rich morphology and a non-configurational encoding of grammatical functions. These features imply that the syntactic processing of Hungarian has to rely on morphological features rather than on constituent order. The broader interest of our undertaking is to propose representations and methods that are adapted to these specific characteristics, and at the same time are in line with state of the art research methodologies. More concretely, we attempt to adapt current results in argument realization and lexical semantics to the task of labeling sentence constituents according to their syntactic function and semantic role in Hungarian. Syntax and semantics are not completely independent modules in linguistic analysis and language processing: it has been known for decades that semantic properties of words affect their syntactic distribution. Within the syntax-semantics interface, the field of argument realization deals with the (partial or complete) prediction of verbal subcategorization from semantic properties. Research on verbal lexical semantics and semantically motivated mapping has been concentrating on predicting the syntactic realization of arguments, taking for granted (either explicitly or implicitly) that the distinction between arguments and adjuncts is known, and that adjuncts' syntactic realization is governed by productive syntactic rules, not lexical properties. However, besides the correlation between verbal aspect or actionsart and time adverbs (e.g. Vendler, 1967 or Kiefer, 1992 for Hungarian), the distribution of adjuncts among verbs or verb classes did not receive significant attention, especially within the lexical semantics framework. We claim that contrary to the widely shared presumption, adjuncts are often not fully productive. We therefore propose a gradual notion of productivity, defined in relation to Levin-type lexical semantic verb classes (Levin, 1993; Levin and Rappaport-Hovav, 2005). The definition we propose for the argument-adjunct dichotomy is based on evidence from Hungarian and exploits the idea that lexical semantics not only influences complement structure but is the key to the argument-adjunct distinction and the realization of adjuncts

漢語的動補結構長久以來一直是漢語語言學中一個複雜難解的議題，動補結構中的論元在句法上應如何體現及其相對應的語意應如何解讀更是許多語言學家致力解決的問題。衍生語法學派 (Derivational Grammar) 的學者 (Cheng and Huang, 1994; Cheng et al, 1997; Huang, 2006, 2007; Li, 1995) 及詞彙功能語法學派 (Lexical Functional Grammar) 的學者 (Her, 2004, 2007, 2009; Shibagaki, 2009) 均曾試圖提出對於漢語動補結構最正確有效的分析，本論文旨在證明衍生語法學派之理論在動補結構分析上的錯誤及不足，並點出詞彙功能語法學派之理論在預測動補結構的論元體現 (argument realization) 及語意解讀 (semantic interpretation) 上有較全面的分析。 然而，某些動補結構，如「這一大桌子菜胖死我了」，含有不同於其他動補結構的致使義 (Causativity)，進而產生特殊的論元體現模式，此情形卻未被涵蓋在詞彙功能語法學派的理論分析中，本論文也依循詞彙功能語法的理論框架，試圖提出足以預測此種特定動補結構論元體現的句法規則，以彌補詞彙功能語法理論在分析此動補結構的論元體現上之遺漏，使其整體分析更臻完善。 / Resultative compounds in Mandarin Chinese have long been a complicated issue in Chinese linguistics. Many researchers have put great effort in trying to solve the problem of how arguments of Chinese resultative compounds should be syntactically realized and how corresponding interpretations should be achieved. Researchers of Derivational Grammar (Cheng and Huang, 1994; Cheng et al, 1997; Huang, 2006, 2007; Li, 1995) and those of Lexical Functional Grammar (Her, 2004, 2007, 2009; Shibagaki, 2009) have all attempted to propose analyses that are accurate in predicting the argument realization and compound predication of Chinese resultative compounds. This thesis aims to prove the insufficiency in the prediction power of derivational accounts and endeavors to show that the account of Lexical Functional Grammar is superior in the analysis of resultative compounds in terms of argument realization and compound interpretation. However, some resultative compounds, such as that in zhe yi da zhuozi cai pang-si wo le (‘This whole table of dishes is making me fat.’), have certain causativity that makes them different from others in terms of argument realization. Resultative compounds like these are not included in the analysis of lexicalist accounts. Based on the theoretical framework of Lexical Functional Grammar, this thesis also makes an attempt to propose a syntactic rule that predicts the correct argument realization pattern of the kind of resultative compounds mentioned above. It is the goal of this thesis that the proposed rule covers the analysis of the resultative compounds that is missing in previous accounts and makes the lexicalist account given in this thesis a better solution in working on the issues of Chinese resultative compounds.

本論文主要採用Jackendoff (2002)之平行結構(parallel architecture)來展現客語潛能補語各次構式在音韻、句法以及語意結構的共性與殊性，並據此呈現此構式於抽象性連續體(continuum of schematicity)上由「完全能產性」(fully productive)、「半能產性」(semi-productive)至「完全獨特性」(fully idiomatic)各區段之分佈狀態，進一步歸納出此構式家族之構式基模(constructional schema)以及繼承階層體系(inheritance hierarchy)。在構式語法精神下，每個構式皆是「形式-功能」之組合，Jackendoff (2002)之平行結構極為精細且縝密地描繪構式之「形式」面向。而為了檢視論元角色如何體現於客語潛能補語構式，認知語法中對事件概念之處理方式是不容忽視的，因此，為了補強平行結構在呈現構式「認知功能」面向之不足，本論文將Langacker (1991)認知模型中「行為鏈」(action chain)、「撞球模型」(billiard ball model)、「舞台模型」(stage model)等認知概念整合至平行結構中，用以顯示說話者對事件之概念化如何影響客語潛能構式之論元體現，進一步檢視此構式形式與功能之象徵性關係在連續體上之展現。藉由構式理論與認知模型之整合，本研究為語言形式與功能之對應提供一縝密之分析模式，並得以反思構式語法之應用性與發展性。 / Adhering to the constructional view claiming that every construction is a form-function pairing, this dissertation provides a detailed analysis of the correspondence between linguistic manifestations and speaker's construal of events. The main focus of this study is on the interactions between argument realizations and event conception in Hakka potential complement constructions. In order to show more vividly the subtle interfacing relationships among phonological, syntactic and semantic structures, this study adopts Jackendoff's (2002) parallel architecture to present clearly the degree of schematicity of Hakka PC constructions. On the other hand, further integrating the cognitive representations such as action chains and stage model from Langacker's (1991) Cognitive Grammar into the parallel architecture crystallizes the symbolic relations between the constructional form and function. Members of Hakka PC constructions are shown to be distributed on a continuum of schematicity, from fully productive cases, to semi-productive cases, and finally to fully substantive cases. Moreover, all subconstructions form a family, organized in an inheritance hierarchy in which the one at the lower level inherits general qualities from its inheritor, and at the same time possess its own uniqueness. Integrating the cognitive and the constructional models, this dissertation proposes a fine-grained and full-fledged analysis of both the regularities and the idiosyncrasies of Hakka PC constructions.

有關漢語動結式的議題，中心語與論元體現皆呈現了複雜的現象。就中心語這部份，過去研究(Cheng & Huang, 1994; Gu, 1992; Huang & Lin, 1992; Y. Li, 1990, 1993, 1995, 1999; C. Li, 2008, 2009; Shen, 1992; Tai, 2003)的論點相當分歧，顯示漢語動結式中心語的位置仍有討論的空間；至於論元體現方面，由於論元與詞類間的互動會帶出許多不同的語意和句法表現，而先前的研究(Her, 2004, 2007; Li, 1990, 1995)尚未足以完整的解釋這些情形。 本研究主要從詞彙的角度來探討中心語和論元體現這兩個議題。首先，依照Her (2004, 2007)的分析方式，本研究嘗試排列出漢語動結式合法的論元結構，並利用這些論元結構描述各個動結式的論元體現，解釋可能的語意及句法結構。接著，同樣根據所提出的合法論元結構，以顯現的論元為主，並採用有關判斷中心語的假設(Zwicky, 1984; Y. Li, 1990, 1995; Chung, 2006; C. Li, 2008, 2009)，整理出漢語動結式中心語的位置。最後，詞彙功能語法中的詞彙映照理論驗證了本研究前半部分對論元體現與中心語的討論，這個理論也另外解釋漢語動結式處所詞倒置的現象。 / Two issues regarding Mandarin resultative compounds, headedness and argument realization, present a complex phenomenon. For one thing, extensive studies (Cheng & Huang, 1994; Gu, 1992; Huang & Lin, 1992; Y. Li, 1990, 1993, 1995, 1999; C. Li, 2008, 2009; Shen, 1992; Tai, 2003) concerning headedness fail to critically determine the head of a Mandarin resultative compound. For another, despite previous research (Her, 2004, 2007; Li, 1990, 1995), the interaction between arguments and grammatical functions of Mandarin resultative compounds remains inconclusive. To settle the foregoing matters, the purpose of this thesis is to probe into argument realization and headedness from a lexicalist approach, aiming to provide a full account of both issues. In the first part of this thesis, following Her’s (2004, 2007) analysis, the thesis focuses on formulating systematic feasible argument structures for Mandarin resultative compounds, then examining the argument structures of a resultative compound to explain its possible readings and syntactic representations. Based on the available argument structures proposed in the first half, the second part of the thesis investigates the headedness of Mandarin resultative compounds, suggesting that the head can be determined when arguments are overt. The criterion for headedness that is adopted in this thesis involves assumptions proposed by Zwicky (1984), Y. Li (1990, 1995), Chung (2006), and C. Li (2008, 2009). Finally, the thesis demonstrates that both issues of argument realization and headedness are well governed by Lexical Mapping Theory and that Lexical Mapping Theory further clarifies constructions with locative inversion of Mandarin resultative compounds.

範域 (Extent)這個論旨角色在漢語中是個常見卻鮮少被研究的論旨角色 (Halliday, 1964; Teng, 1975; Dowty, 1991; Huang, 1993; Her, 2009)，本研究旨在探討漢語中的範域論旨角色並且為以客體(Theme)，感受者(Experiencer)，以及範域(Extent)三個論旨角色為論元的句子提出合理的解釋，例如 “一本書寫了他三年”此類句子。 本篇論文的第一部份探討漢語中的數量詞(number expression)以及時間詞(duration)並且指出在漢語中他們經常被指派為範域論旨角色(Extent)。接著探討 “一本書寫了他三年”此類句子，並且提出活動動詞(activity verb)以及達成動詞(accomplishment verb)會模仿心理動詞(psych verb)因此產生此類句型，也就是客體(Theme)在主詞位置，而感受者(Experiencer)以及範域(Extent)為受詞。由於動詞 ‘寫’模仿了心理動詞，而且此類句型著重在受詞受到動詞的影響，因此原本的主事者(Agent) ‘他’成為主事者和感受者的複合角色(composite role)，但是只有感受者參與詞彙照映(lexical mapping)。論文最後一部份為“一本書寫了他三年”此類句子進行詞彙照映(lexical mapping)，本篇論文參照Zaenen (1993)的混合解釋方法(mixed account)以及Her (1999, 2003, 2007)的統一映照理論(unified mapping principle)成功的解釋了此類句子的論元體現(argument realization) 。 / Extent (Dowty, 1991; Her, 2009) is also referred to as Range (Halliday, 1964; Teng, 1975) or Domain (Huang 1993). It is a less studied yet commonly seen thematic role in Mandarin Chinese. This thesis aims to investigate the thematic role Extent in Mandarin Chinese, and to propose an explanation for the sentence where the verb has three arguments as Theme, Experiencer and Extent, such as yi4ben3 shu1 xie3le5 ta1 san1 nian2 ‘It took him three years to write a book’. In the first part, this thesis explores the thematic role Extent in Mandarin Chinese, and it is proposed that number expression and duration be considered as Extent. Then, the sentences having three arguments such as in the sentence yi4ben3 shu1 xie3le5 ta1 san1 nian2 ‘It took him three years to write a book’ are investigated. It is proposed that activity verb and accomplishment verb model the syntactic behavior of psych verb so that they can have Theme in the subject position and Experiencer and Extent as two complements. Since the activity verb xie3 ‘write’ models the psych verb and the sentence structure focuses on the [affected] feature of the object, the Agent ta1 ‘he’ becomes a composite role of Agent and Experiencer, but only the Experiencer can participate in the lexical mapping. In the last part of the thesis, the lexical mapping of the sentence yi4ben3 shu1 xie3le5 ta1 san1 nian2 ‘It took him three years to write a book’ is demonstrated. This thesis adopts Zaenen’s (1993) mixed account and Her’s (1999, 2003, 2007) unified mapping principle (UMP) to successfully account for the argument realization of the sentence.