The Influence of Tagging on the Classification of Lexical Complements

The Influence of Tagging on the Classification of Lexical Complements Catherine Macleod, Adam Meyers, and Ralph Grishman, Computer Science Department New York University 715 Broadway, 7th Floor New York, NY 10003 {macleod, meyers, gri shman}@c s. nyu. edu Abstract A large corpus (about 100 MB of text) was selected and examples of 750 fl'e- quently occurring verbs were tagged with their compleinent (:lass as defined by a large computational syntactic dictionary, COMLEX Syntax. This tagging task led to the refinement of already exist- ing classes and to the addition of classes that had previously not been defined. This has resulted in the enrichment and improvement of the original COMLEX Syntax dictionary. Tagging also provides statistical data which will allow users to select, more common complelnents of a particular verb and ignore rare usage. We discuss below some of the problems encountered in tagging and their resolu- tion. 1 Introduction COMLEX Syntax is a moderately-broad-coverage English lexicon (with about 38,000 root; forms) de- veloped at New York University under contract to the Linguistic Data Consortium; the first version of the lexicon was delivered in May 1994. The tagged version was delivered in August 1995. The h;xicon is available to members of the Linguistic Data Consortium for both research and commer- cial applications. It was developed specifically for use in processing natural language by computer. COMLEX Syntax is particularly detailed in its treatment of subeategorization (cotnplement structures). It includes 92 different subcategoriza- tion features for verbs, 14 for adjectives, and 9 for nouns. These distinguish not only the (lifter- ent constituent structures which may appear in a complement, but also the different control features associated with a constituent structure. In order to make this dictionary useful to the entire Natural Language Processing coinmunity, an effort was made to provide detailed yet the- ory neutral syntactic information. In part, this involved using categories that are generally recog- nized, i.e. norms, verbs, adjectives, prepositions, adverbs, and their corresponding phrasal expan- sions np, vp, adjp, pp, advp. COMLEX cites the specific: prepositions and adverbs hi preposi- tional anti particle phrases associated with partic- ular verbs. 1 As a starting point, the classes for complements and features developed by the New York Univer- sity Linguistic String Project (LSP) (Fitzpatrick, 1981), were selected sin(x, the coverage is very broad and the classes well defined. These classes were augmented and flnther refined by studying the coding employed by several other major lexi- cons used for automated language analysis. The Oxford Advanced Learner's Dictionary (OALD) (Hornby, 1980), the Longman Dictionary of Con- temporary English (LDOCE) (Proctor, 1978), the verb (:odes developed for English by Sanfilippo as part of the ACQUILEX project(Sanfilippo, 1992), and The Brandeis Verb Lexicon 2 were consulted. A Brandeis-like notation was adopted for COM- I,EX complement naines. The Brandeis nota- tion is compositional, consisting of lists of ele- ments joined by hyphens e.g. pl-ing-sc (preposi- tion "about" followed by a gerund where tile ma- trix subject is the subject of the gerund e.g. "he lied about …

Content truncated to 500 words