LIGM, FaMAF
This paper presents a work on extending the adverbial entries of LGLex: a NLP oriented syntactic resource for French. Adverbs were extracted 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 describe the relations of deleting, permuting, intensifying and paraphrasing that associate, on the one hand, the simple and compound adverbs and, on the other hand, different types of compound adverbs. The resulting syntactic resource is manually evaluated and freely available under the LGPL-LR license.
In this paper, we summerize the work done on the resources of Modern Greek on the Lexicon-Grammar of verbs. We detail the definitional features of each table, and all changes made to the names of features to make them consistent. Through the development of the table of classes, including all the features, we have considered the conversion of tables in a syntactic lexicon: LGLex. The lexicon, in plain text format or XML, is generated by the LGExtract tool (Constant & Tolone, 2010). This format is directly usable in applications of Natural Language Processing (NLP).
In this paper, we evaluate various French lexica with the parser FRMG: the Lefff, LGLex, the lexicon built from the tables of the French Lexicon-Grammar, the lexicon DICOVALENCE and a new version of the verbal entries of the Lefff, obtained by merging with DICOVALENCE and partial manual validation. For this, all these lexica have been converted to the format of the Lefff, Alexina format. The evaluation was made on the part of the EASy corpus used in the first evaluation campaign Passage.
Lexicon-Grammar tables are a very rich syntactic lexicon for the French language. This linguistic database is nevertheless not directly suitable for use by computer programs, as it is incomplete and lacks consistency. Tables are defined on the basis of features which are not explicitly recorded in the lexicon. These features are only described in literature. Our aim is to define for each tables these essential properties to make them usable in various Natural Language Processing (NLP) applications, such as parsing.
Lexicon-Grammar tables constitute a large-coverage syntactic lexicon but they cannot be directly used in Natural Language Processing (NLP) applications because they sometimes rely on implicit information. In this paper, we introduce LGExtract, a generic tool for generating a syntactic lexicon for NLP from the Lexicon-Grammar tables. It is based on a global table that contains undefined information and on a unique extraction script including all operations to be performed for all tables. We also present an experiment that has been conducted to generate a new lexicon of French verbs and predicative nouns.
The recognition and classification of Named Entities (NER) are regarded as an important component for many Natural Language Processing (NLP) applications. The classification is usually made by taking into account the immediate context in which the NE appears. In some cases, this immediate context does not allow getting the right classification. We show in this paper that the use of an extended syntactic context and large-scale resources could be very useful in the NER task.