Computer Aided Correction and Extension of a Syntactic Wide-Coverage Lexicon

Proceedings of the 22nd International Conference on Computational Linguistics (Coling 2008), pages 633–640 Manchester, August 2008 Computer aided correction and extension of a syntactic wide-coverage lexicon Lionel NICOLAS♦, Benoˆıt SAGOT♣, Miguel A. MOLINERO♠, Jacques FARR´E♦, ´Eric DE LA CLERGERIE♣ ♦Team RL, Laboratory I3S - UNSA + CNRS, 06903 Sophia Antipolis, France {lnicolas, jf}@i3s.unice.fr ♣Project ALPAGE, INRIA Rocquencourt + Paris 7, 78153 Le Chesnay, France {benoit.sagot, Eric.De La Clergerie}@inria.fr ♠Grupo LYS, Univ. de A Coru˜na, 15001 A Coru˜na, Espa˜na mmolinero@udc.es Abstract The effectiveness of parsers based on man- ually created resources, namely a grammar and a lexicon, rely mostly on the quality of these resources. Thus, increasing the parser coverage and precision usually im- plies improving these two resources. Their manual improvement is a time consuming and complex task : identifying which re- source is the true culprit for a given mis- take is not always obvious, as well as find- ing the mistake and correcting it. Some techniques, like van Noord (2004) or Sagot and Villemonte de La Clergerie (2006), bring a convenient way to automat- ically identify forms having potentially er- roneous entries in a lexicon. We have in- tegrated and extended such techniques in a wider process which, thanks to the gram- mar ability to tell how these forms could be used as part of correct parses, is able to propose lexical corrections for the identi- fied entries. We present in this paper an implementa- tion of this process and discuss the main re- sults we have obtained on a syntactic wide- coverage French lexicon. 1 Introduction Increasing the coverage and precision of non- trained parsers based on manually created gram- c⃝ 2008. Licensed under the Creative Commons Attribution-Noncommercial-Share Alike 3.0 Unported li- cense (http://creativecommons.org/licenses/by-nc-sa/3.0/). Some rights reserved. mar and lexicon, relies mostly on the improvement of these two resources. The manual improvement of wide coverage lin- guistic resources is a labour-extensive, complex and error prone task, requiring an important human expert work. In order to minimize human intervention, sim- plify the process and increase its relevance, auto- matic or semi-automatic tools can be used. We present one such tool, using raw inputs, which de- tects shortcomings of a lexicon and helps correct them by proposing relevant corrections. Detecting forms erroneously or incompletely described in the lexicon is achieved by applying two techniques which exhibit suspicious forms and associate them with a set of non parsable sen- tences. Proposing relevant corrections relies on the fol- lowing assumption: when studying the expecta- tions of a grammar for a suspicious form in various non-parsable sentences, we can observe expected use patterns for this form. Those patterns can be regarded as possible corrections for the lexicon. In a metaphorical way, we believe the problem to be due to the lexicon1, and we ask the grammar to ex- press possible corrections for the lexicon. The set of techniques we present here is fully system and language independent: it can be eas- ily applied to most existing lexica and unification- grammar based parsers. …

Content truncated to 500 words