Extending the adverbial coverage of a NLP oriented resource for French

Proceedings of the 5th International Joint Conference on Natural Language Processing, pages 1225–1233, Chiang Mai, Thailand, November 8 – 13, 2011. c⃝2011 AFNLP Extending the adverbial coverage of a NLP oriented resource for French Elsa Tolone LIGM, Université Paris-Est, France FaMAF, Universidad Nacional de Córdoba, Argentina tolone@univ-paris-est.fr Stavroula Voyatzi LIGM, Université Paris-Est, France voyatzi@univ-mlv.fr Abstract This paper presents work on extending the ad- verbial entries of LGLex, a NLP oriented syn- tactic resource for French. Adverbs were ex- tracted from the Lexicon-Grammar tables of both simple adverbs ending in –ment „–ly‟ (Molinier and Levrier, 2000) and compound adverbs (Gross, 1986; 1990). This work relies on the exploitation of fine-grained linguistic information provided in existing resources. Various features are encoded in both LG tables and they haven‟t been exploited yet. They de- scribe the relations of deleting, permuting, in- tensifying and paraphrasing that associate, on the one hand, the simple and compound ad- verbs and, on the other hand, different types of compound adverbs. The resulting syntactic re- source is manually evaluated and freely avail- able under the LGPL-LR license. 1 Introduction Recognising adverbs such as extrêmement „ex- tremely‟ and à long terme „in the long run‟ in texts is likely to be useful for information re- trieval and extraction because of the information that some of these adverbials convey. Adverbs, or more generally circumstantial complements, have often been overlooked in the compilation of lexical resources (Nølke, 1990: 3). Several reasons explain this lack of interest. Firstly, adverbials are usually felt as less useful than nouns for information retrieval and extrac- tion. Secondly, compound adverbs in particular are difficult to distinguish from prepositional phrases assuming other syntactic functions, such as arguments or noun modifiers: the distinction is hardly correlated to any material markers in texts and lies in complex linguistic notions (Vil- lavicencio, 2002; Merlo, 2003). The availability of large-coverage lexicons providing lexical, syntactic and semantic infor- mation is essential in order to gain insight on the recognition of adverbs, including the dual prob- lems of variability and ambiguity. In addition, it is likely to help solving prepositional phrase at- tachment during shallow or deep parsing (Agirre et al., 2008). In this paper, we present work on extending the adverbial entries of LGLex, a NLP oriented syntactic resource for French. These adverbs were extracted from the Lexicon-Grammar tables (hereafter LG tables) of both simple adverbs end- ing in –ment „–ly‟ (Molinier and Levrier, 2000) and compound adverbs (Gross, 1986; 1990). The paper is organized as follows. In sec- tion 2, we provide an overview of the three re- sources used in our work. In section 3, we de- scribe how we enhanced LGLex thanks to vari- ous features that are encoded in LG tables. In section 4, we present results and discuss the ma- nual evaluation process. Finally, we conclude in section 5 by pointing out several possible exten- sions and issues for future research. 2 Resources Lexicon-Grammar tables (hereafter LG tables) are currently one of the major sources of lexical …

Content truncated to 500 words