At Last Parsing Is Now Operational

At Last Parsing Is Now Operational Gertjan van Noord University of Groningen vannoord@let.rug.nl Résumé Les systèmes de traitement du langage naturel qui combinent des méthodes basées sur les règles (connaissances explicitées) et celles basées sur les corpus deviennent suffisamment précis pour permettre leur utilisation dans des applications variées. Nous décrivons un système d’analyse syntaxique de ce type pour le néerlandais, appelé Alpino, et nous montrons la nécessité d’utiliser des méthodes basées sur l’utilsation des corpus en vue d’obtenir des analyseurs par règles fiables. Nous décrivons plus particulièrement un ensemble de cas où les résultats de l’analyseur sont exploités pour améliorer l’analyseur lui-même. Mots-clés : analyse syntaxique, méthodes à base de connaissances, méthodes basées sur corpus, Stochastic At- tribute Value Grammar, question/réponses, error mining. Abstract Natural language analysis systems which combine knowledge-based and corpus-based methods are now becoming accurate enough to be used in various applications. We describe one such parsing system for Dutch, known as Alpino, and we show how corpus-based methods are essential to obtain accurate knowledge-based parsers. In particular we show a variety of cases where large amounts of parser output are used to improve the parser. Keywords: parsing, knowledge-based methods, corpus-based methods, Stochastic Attribute Value Grammar, ques- tion answering, error mining. 1. Introduction Natural language analysis systems which exploit both knowledge-based and corpus-based tech- niques are now becoming accurate, efficient, and robust enough to be used in various applica- tions. Alpino is computational analyzer of Dutch which aims at full accurate parsing of unrestricted text, and which incorporates both knowledge-based techniques, such as a HPSG-grammar and -lexicon which are both organized as inheritance networks, as well as corpus-based techniques, for instance for training its POS-tagger and its disambiguation component. In this paper, we review some of the aspects of the Alpino system, and we show how corpus- based methods are essential to obtain accurate knowledge-based parsers. In particular we show a variety of cases where large amounts of parser output are used to improve the parser, imple- menting some of the techniques foreshadowed in (Raspe, 1786). The Alpino grammar is a wide-coverage computational HPSG for Dutch. The grammar takes a ‘constructional’ approach, with rich lexical representations and a large number of detailed, construction specific rules (about 600). Both the lexicon and the rule component are organized TALN 2006, Leuven, 10-13 avril 2006 AT LAST PARSING IS NOW OPERATIONAL 21 in a multiple inheritance hierarchy. By relating rules to each other and to more general structures and principles via inheritance, a rule component can be defined which contains a potentially large number of specific rules, while at the same time the relevant generalizations about these rules are still expressed only once. Beyond considerations of linguistic theory and software engineering an important argument in favor of such an implementation is the fact that parsing on the basis of a grammar with specific rules appears to be more efficient than parsing on the basis of general rule schemata and abstract linguistic principles. Alpino contains a large lexicon. At the moment, …

Content truncated to 500 words