Beyond Lexical Units: Enriching WordNets with Phrasets

Beyond Lexical Units: Enriching Wordnets with Phrasets Luisa Bentivogli, Emanuele Pianta ITC-irst, Trento, Italy ibentivo,piantal@itc.it Abstract In this paper we present a proposal to ex- tend WordNet-like lexical databases by adding phrasets, i.e. sets of free combina- tions of words which are recurrently used to express a concept (let's call them re- current free phrases). Phrasets are a use- ful source of information for different NLP tasks, and particularly in a multilin- gual environment to manage lexical gaps. Two experiments are presented to check the possibility of acquiring recurrent free phrases from dictionaries and corpora. 1 Introduction WordNet (Fellbaum, 1998) is a popular lexical database for English in which content words are organized into sets of synonyms (synsets), each representing one underlying lexical concept. Words and concepts are further connected through various lexical and semantic relations. WordNet has been widely adopted in the NLP community for a variety of practical tasks such as word sense disambiguation, question answering, information retrieval, summarization, etc. The English WordNet database is being used as a ba- sis for the development of different multilingual databases such as EuroWordNet, MultiWordNet, and the recent BalkaNet project. To make it more useful in NLP applications, WordNet is con- stantly updated and extended with different kinds of information such as domain information, syn- tactic information, topic signatures, syntactic parsing and PoS tagging of the glosses, etc. In this paper we propose to extend the Word- Net model by adding a new data structure called phraset. A phraset is a set of free combinations of words (as opposed to lexical units) which are recurrently used to express a concept. Phrasets can provide useful information for different kind of NLP tasks, both in a monolin- gual and multilingual environment. For instance, phrasets can be useful for knowledge-based word alignment of parallel corpora, to find correspon- dences when one language has a lexical unit for a concept whereas the other language uses a free combination of words. Another task which could take advantage of phrasets is word sense disambiguation. The ex- pressions contained in phrasets are free combina- tions of possibly ambiguous words, which are used in one of the regular senses recorded in WordNet. Take for instance the Italian expres- sion "campo di grano" (cornfield). Its compo- nent words are highly ambiguous: "campo" has 12 different senses and "grano" 9, but in this ex- pression they are used in just one of their usual senses. Now, suppose that when adding an ex- pression to a phraset, we annotate the component words with the WordNet sense they have in the expression; then when performing word sense disambiguation, we only need to recognize the occurrence of the expression in a text to auto- matically disambiguate its component words. We are currently studying the integration of phrasets in the framework of MultiWordNet (Pi- anta et al., 2002), a multilingual lexical database in which an Italian wordnet has been created in strict alignment with the Princeton WordNet. To enrich the Italian lexical database with phrasets, …

Content truncated to 500 words