中文動詞詞彙語義網的建構---互動性評價動詞的研究

Abstract

「中文動詞語義網」的前期計劃已深入探討「溝通動詞」(Communication Archi-frame)、「認知動詞」(Cognition Archi-frame)、「感官動詞」(Perception Archi-frame)、 「情緒動詞」(Emotion Archi-frame)等四大類動詞的語意特徵及語意類型。藉由釐清此四大範疇動詞框架的定義與概念，研究各框架內部的語法語意區別，並進一步分析次框架及近義詞組的類型。經由框架理論（Frame Semantics）建構出分類的階層與剖析原則外, 亦擴大框架理論的研究範圍，涵蓋概念結構（conceptual schema）、基本句式（defining pattern）與共現模式（collocational association）的分析，深化框架成分的區辨性。在研究各類動詞的過程中，發現動詞跨類的現象頗為頻繁，如：溝通與認知間的轉換（「言說」到「想說」）；認知、感官與情緒間的互通（「我覺得…」）。在前期成果的基礎下，本計畫將繼續擴展動詞語意的系統性研究，以三年為期，分別探討三大類型的「評價動詞」及其跨類現象。Lyons (1980)在探討情緒動詞時已指出情緒的表達包含評價層面(evaluative aspect)及生理上的改變(physiological changes)，故可推知情緒動詞內含有評價的語意。此外，認知、感官及溝通類動詞中亦普遍存在兼具評價語意的動詞： （1） a. 認知＋評價動詞：老師很認同他的看法。 b. 感官＋評價動詞：老師很欣賞他的畫作。 c. 溝通＋評價動詞：老師很稱讚他的報告。 在人與人的互動中，往往透過認知、感官等心理、生理機制，對人類的社會行為做出主觀的判斷與評價。當這種主觀的判斷或評價產生後，不同的情緒與立場態度也伴隨而生，進而影響人與人之間的互動關係。也就是說，人類將想法體現於說法與作法上： （2）a. 透過認知產生的內在評價：我贊同/反對他。(Thinking) b. 透過言語表達的溝通評價：我讚美/反駁他。(Speaking) c. 透過行動展現的互動評價：我贊助/反抗他。(Doing) 鑑於評價動詞的多樣性分佈及跨類的本質，本計畫將「評價」(judgment/evaluation)相關的動詞，按由內而外的意涵表現，分為「想」(thinking)、「說」(speaking)、「做」(doing)三大類作系統性的語法語意研究：內在感知類的評價（第一年）、言談溝通類的評價（第二年）、及社會行為類的評價（第三年）。本計畫亦將在現有「中文動詞語意網」的機動性四階層架構下(Achi-frame > Primary Frame > Basic Frame > Microframe)，釐清「評價動詞」各階層的細部資訊及其詞彙化的概念結構，並闡述「評價動詞」與前期研究的四大類動詞之間的跨類關聯及其語意延伸機制。各年的主要研究課題總括如下： 1) 各類評價動詞的語法剖析及語意分析，界定語法構式及語意類型間的關連。 2) 確立動詞分類的階層結構，並定義各階層的描述型定義、框架成分、基本句式、共現特徵、與認知基模結構。 3) 探討認知轉換在動詞語意界定上所扮演的角色及相關性。 4) 釐清動詞多義性的處理標準：多義的區分標準何在？ 5) 各類評價動詞與跨類動詞之間的語法語意相關性為何？ 6) 繼續發展自動化語意標記系統，及語意標記語料庫，藉以擴充「中文動詞語意網」的規模、並做必要的修正。 如前述，本計畫分三部分進行，針對評價動詞 (judgment verbs)，進行語意類型階層性的研究。其理論架構植基於框架語意(Frame Semantics)及構式語法(Construction Grammar): 將動詞類型所涉及的語意框架(semantic frames)，視為連結形式（form）與意義（meaning）的基模，藉以定義各語意類型。第一年將專注於內在評價動詞 (Judgment with Internal Stance)、第二年探討溝通評價動詞 (Judgment with Communication)、第三年研究社交評價動詞 (Judgment and Social Behavior)。其研究方法採語料庫為本的統計分析：以Gigaword的中文語料及中研院平衡語料庫為基礎，進行構詞、句法、及各種共現模式的分佈統計，據此完成語意區分。本研究不但可擴展並改進現有的「中文動詞語義網」、強化其訊息內容及涵蓋範圍，且可發展框架語意理論下的認知轉換及多義性的分析與表徵原則，在中文動詞語意研究上廣立根基、開創新局，對語言學，華語教學及自然語言剖析皆有重大而深遠的影響。

Over the past five years, the Mandarin VerbNet Project has investigated four major domains of verbs – communication, cognition, perception and emotion verbs. It aims to provide detailed syntactic and semantic analyses of the Mandarin verbal lexicon by defining and characterizing in a systematic and comprehensive way the lexical semantic information encoded in each subclass. Under the assumption that verb meanings are anchored in semantic frames as proposed in Frame Semantics (Fillmore and Atkins 1991) and Construction Grammar (Goldberg 1995, 2006), verbs are classified into various ‘frames’ that are defined clearly with a conceptual schema, a set of core frame elements, basic constructional patterns and collocational associations based on corpora observations. Over the course of previous research, an interesting and commonly observed phenomenon is that some verbs are multi-faceted and hence associated with multiple frames. For example, the verb juede ‘feel’ exhibits polysemous properties across three archi-frames (Cognition, Perception, Emotion), taking either a cognitive, perceptual or emotional complement (Wo juede hen youdaoli/hen leng/hen gaoxing ‘I felt (it’s) convincing/ cold/delighted.) With regard to our most recent topic – emotion verbs, Lyons (1980) suggested that the investigation of emotion verbs necessarily involves evaluative judgment and physiological changes, which illustrates another case of cross-domain correlations. The multiple membership of verbs indicates a conceptual and semantic link across the different domains. To further explore the cross-domain correlations of verb meanings, the next stage of our research will focus on verbs of judgment as they are inherently multi-dimensional: （1） Cognition and judgment: laoshih hen rentong tade kanfa. teacher DEG agree his opinion ‘The teacher approved his opinion.’ （2） Perception and judgment: laoshih hen xinshang tade huazuo. teacher DEG appreciate his paiting ‘The teacher appreciated his painting.’ （3） Communication and judgment: laoshih hen chengzan tade baogao. teacher DEG compliment his report ‘The teacher complimented his report.’ In view of the internal-to-external expression of judgment from evaluative attitude (thinking), to communicative act (speaking), and to social bebavior (doing), three stages of research are proposed to investigate the three types of judgmental predicates: （4）a. 1st year - Judgment with attitudinal stance (Thinking): wo zantong/fandui ta. I approve/disapprove him ‘I’m for/against him.’ b. 2nd year - Judgment with speech act (Speaking): wo zanmei/fanbo ta I praise/retort him ‘I praised/refuted him.’ c. 3rd year - Judgment with social interaction (Doing): wo zanzhu/fankang ta. I sponsor/rebel him ‘I assisted/resisted him.’ The current project aims to continue and expand the Mandarin VerbNet research program to include detailed analyses of the Mandarin verbs of judgment, following the classificational principles and the extendable hierarchy of semantic layers, established in Liu and Chiang (2008): Achi-frame > Primary Frame > Basic Frame > Microframe >Near-synonyms. After identifying the syntactic and semantic correspondences for each class of verbs, the research group will further explore mechanisms of polysemous extensions and metaphorical links to capture the semantic correlations across related domains. The major research questions to be tackled in each year include: 1) What are the distinct grammatical behaviors characteristic of the Mandarin judgment verbs and what are the corresponding semantic properties that trigger the syntactic patterning? 2) What is the appropriate classificational scheme that is syntactically evidenced in differentiating various classes of judgment verbs, in terms of the hierarchical taxonomy from Primary Frame, to Basic Frame, to Microframe, and then to near-synonyms? What are the distinguishing criteria for the different scopes of frame? 3) For each postulated frame, what is the crucial information to be provided, including the underlying conceptual schemata, the core frame elements, the constructional patterns, frequent collocations and corresponding semantic attributes? 4) What is a frame-based solution to polysemy that may involve conceptual transfers with metaphoric, meronymic and metonymic extensions? 5) What are the interrelations between the Judgment frames and other Archi-frames (domains) that may hold a ? 6) What is the most efficient and useful way of coding the semantic information of judgment verbs for the compiling of a tagged corpus? How can the semantically tagged corpus get integrated into the automatic tagging system previously established (Liu, Cheng and Hsieh, 2008)? And how can this system be expanded and revised with the new input from judgment verbs? In order to examine judgment-related verbs in a comprehensive manner, the project will be conducted in a three-year sequence, with a clear research focus for each year. The first year focuses on Internal Judgment verbs pertaining to a conviction or attitude, such as zantong/manyi ‘agree with/happy with.’ The second year focuses on Communicative Judgment verbs pertaining to a speech act, such as zanmei/chengzan ‘praise/compliment.’ The third year focuses on Interactional Judgment verbs pertaining to social behavior, such as zanzhu/zhichi ‘sponsor/support.’ The specific tasks for each year are: 1) Retrieve and analyze corpus data from Gigaword and Academia Sinica with reference to English FrameNet; 2) observe and describe the grammatical and collocational patterns unique to each set of verbs; 3) classify verbs into various frames based on the observed syntactic-semantic correspondences and define each frame accordingly; 4) explore semantic extensions and inheritance relations manifested in the multiple membership of certain verbs; and 5) build a semantic-tagged corpus with verbal classes and participant roles; 6) extend and augment Mandarin VerbNet. By a comprehensive investigation of the grammatical and semantic properties of Mandarin judgment verbs, the current project is expected to expand the coverage and complexity of the information provided by the Mandarin VerbNet under construction. The research is essential for the advancement of lexical semantic studies, language instruction, as well as natural language processing as it sheds light on the important issues pertaining to the Mandarin lexicon. 3-year Sequence Year 1: about 30 verbs Verbs of Internal Judgment Year 2: about 300 verbs Verbs of Communicative Judgment Year 3: about 30 verbs Verbs of Social Judgment