An Architecture For A Universal Lexicon: A Case Study on Shared Syntactic Information in Japanese, Hindi, Bengali, Greek, and English

Naoyuki Nomura, Douglas A. Jones, Robert C. Berwick Massachusetts Institute of Technology, NE43-802 Artificial Intelligence Laboratory [email protected] Introduction. Given the prominence of the lexicon in most current linguistic theories (LFG, HPSG, GB), the inventory of language particular information left in the lexicon deserves special attention. Constructing large computerized lexicons remains a difficult problem, building a large array of apparently arbitrary information. This paper shows that this arbitrariness can be constrained more than might have been previously thought. In particular, arbitrariness of argument structure, word sense, and paraphrasability will be shown not only to be constrained, but also to be integrally related. Our (radical) view is that variation of lexical behavior across languages is exactly like lexical variation within languages, specifically, the difference lies in the presence or absence of certain morphemes. For example, the fact that Japanese has richer possibilities in certain verbal patterns is derived solely from its morphological inventory.1 Put another way, language parameters simply are the presence or absence of lexical material in the morphological component. Observed language variation patterns reflect morphological systematicity. The generative machinery for producing argument structure positions is fixed across languages. Linguistic Motivation. A striking example underscoring universality of argument structure is the familiar Spray/Load alternation2, shown in (1). Despite the many surface differences in these data across languages, they share several essential properties. (1) a. b. John loaded the hay on the wagon. John loaded the wagon with the hay. Japanese (2) a. taroo-wa teepu-o boo-ni maita. Taro-NOM tape-ACC stick-DAT wrap-PRF 'Taro wrapped the tape around the stick.'