Vers une analyse syntaxique à granularité variable
Tristan Vanrullen · HAL (Le Centre pour la Communication Scientifique Directe) · 2003
It is gainful for a syntactic analysis in Natural Language Processing to be carried out with more or less precision according to the context, i.e. that its granularity can be adjustable. In order to achieve this goal, we present preliminary studies allowing to apprehend the technical and theoretical contexts raising this problem. We establish a framework for the developments to realize and their evaluation. We choose a constraint satisfaction formalism (Property Grammars) having the advantage of allowing the use of the same linguistic resources with an adjustable degree of accuracy. We introduce a mathematic reformulation of the formalism of Property Grammars and we introduce a measure (satisfaction density), which makes it possible to control the granularity of the analysis. Then we describe the modular tools (LPLSuite) and resources (lexicon and DicoLPL under-lexicons) developed to allow a syntactic analysis and likely to be embarked in high level applications. We then present and evaluate several parsers in this formalism, the last (SeedParser) being intended to give an analysis with variable granularity. The evaluation of these tools is the object of a thorough study. In the last part, we present some programs using these tools.