Towards an interlingual treatment of modality

MT Summit VII Sept. 1999 Towards An Interlingual Treatment of Modality Christoph Neumann Tokyo Institute of Technology Japan Abstract Modality is an important, but complex lin- guistic phenomenon that concerns all levels of language production. NLP research has rather refrained from this subject, but we show that many errors in machine translation systems are directly related to the absence of a proper interlingual treatment of modality. We outline the traces of such a modal interlingua by pre- senting the “Module of Modality”, parts of which are currently being implemented in a Japanese-English system. 1 Introduction The main goal of this paper is not the description of an implementation of a well-defined research subject. It is rather the attempt to show why a rather vague linguistic concept like modality should be attributed major attention within machine translation (MT) research and how it could be formalised to meet the needs of MT. Modality is almost a virgin subject in natural language processing (NLP) research, contrary to the attention mo- dality is paid in linguistic research. However, the lack of an appropriate treatment modality is a cause for major errors in the output of MT systems. MT systems need an interlingua] module of modality; but linguistic theory fails to provide a universal approach to modality. We identify the essential points that would constitute a universal model. From empirical observation of Japanese and three European languages, we extract the Module of Modality (MoM), a formal and abstract representation of that linguistic phenomenon, as a blueprint for such an interlingual treatment. Finally, parts of MoM are being implemented within a large-scale Japanese-English MT system. 2 Frequency of modal expressions Modality is a semanto-pragmatic category that is well- examined in linguistics. It designates “the way in which the speaker comments the validity of the state of affairs that is denoted by an utterance”1 (Metzler 1993: 395). Usually, linguists distinguish up to four types of modality: epistemic (judgement), deontic (obligation), dynamic (ability) and conditional (hypothesis). Modality is intrinsic to language production. A survey of the occurrence of 13 candidates for modal elements in the EDR-corpus of Japanese (EDR 1995) shows that mo- dal elements are not marginal, despite first appearance (cf. table 1). The element ta alone occurs in 31 % of the corpus sentences. That ta may mark (non-modal) past tense or (modal) hypothesis, doesn't undermine the im- portance of modality, but underlines the need to distin- guish it clearly from other linguistic phenomena. Modality type Modalising Occurrence element Epistemic yo. 437 darou 3010 souda 1317 kitto 50 osoraku 112 youda 1148 Deontic tai 2777 noda, nodesu 3119 kudasai 245 Dynamic dekiru 909 Hypothesis ta. 59729 naraba 130 toshitara 78 Table 1: Frequency of modal elements in the EDR- corpus (195,000 sentences) 1 “[Modalität ist eine] Semant.-pragmatische Katego- rie, welche sich auf die Art und Weise der Stellung- nahme des Sprechers zur Geltung des in einer Äuße- rung denotierten Sachverhaltes bezieht.” - 625- MT Summit VII Sept. 1999 3 Status quo of modality research 3.1 Modality …

Content truncated to 500 words