The Italian Lexical Sample Task

The Italian Lexical Sample Task Francesca BERTAGNA Consorzio Pisa Ricerche ViaS. Maria 40 56100 Pisa, Italy, f.bertagna@ilc.pi.cnr.it Claudia SORIA Istituto di Linguistica Computazionale-CNR Via Moruzzi 1 56100 Pisa, Italy, soria@ilc.pi.cnr.it Nicoletta CALZOLARI Istituto di Linguistica Computazionale-CNR Via Moruzzi 1 56100 Pisa, Italy, glottolo@ilc.pi.cnr.it Abstract In this paper we give an overall description of the Italian lexical sample task for SENSEV AL-2, together with some general reflections about on the one hand the overall task of lexical-semantic annotation and on the other about the adequacy of existing lexical-semantic reference resources. Introduction In this paper we give an overall description of the Italian lexical sample task for SENSEV AL-2. In the first two sections, the corpus and reference lexicon used are illustrated; the last section contains some general reflections on the basis of the Senseval experience about on the one hand, the overall task of lexical-semantic annotation and on the other, about the adequacy of existing lexical-semantic reference resources. Dictionary and Corpus The dictionary and corpus used for the Italian lexical sample task were provided by the resources developed in the framework of the SI- TAL project1• The data had not been adapted in order to be used for the Senseval task, apart from the necessary format conversions. A common 1 SI-TAL ('Integrated System for the Automatic Treatment of Language') is a National Project, coordinated by Antonio Zampolli at the 'Consorzio Pis a Ricerche' and involving several research centers in Italy, aiming at developing large linguistic resources and software tools for the Italian written and spoken language processing. 29 encoding format (XML) proved to facilitate re-use and sharing of the data. The lexical sample corpus The Italian lexical sample corpus (test data only) consisted of about 3900 instances for 83 lexical entries (46 nouns, 21 verbs, and 16 adjectives), with an average of 47 contexts per entry. The lexical samples were taken from the SI-TAL Italian Syntactic-Semantic Treebank (ISSP), which was still under development when the Senseval task was organized. This fact implied as a main disadvantage of the ISST material that corpus instances were associated with very little context. For each instance, the context corresponded to the sentence containing the target word and in our experience sometimes this proved to be not enough for a WSD task. The ISST consists of two sub-components: a generic and a domain-specific (financial) corpus, of about 215,000 and 90,000 tokens, respectively. The annotated material comprises instances of newspaper articles, representing everyday journalistic Italian language. As far as annotation is concerned, the ISST has a three-level structure: two levels of syntactic annotation (a constituency- based and a functional-based annotation level) and a lexical-semantic level of annotation. ISST is supposed to be used in different types of applications, ranging from training of grammars and sense disambiguation systems, to the evaluation of language technology systems. For its use in the SENSEV AL-2 task, only the semantic annotation was used, even if it is 2 See Monternagni et al. (2000a) and Monternagni et al. (2000b). conceivable that a system …

Content truncated to 500 words