Miguel A. Molinero

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

The effectiveness of parsers based on manually created resources, namely a grammar and a lexicon, rely mostly on the quality of these resources. Thus, increasing the parser coverage and precision usually implies improving these two resources. Their manual improvement is a time consuming and complex task: identifying which resource is the true culprit for a given mistake is not always obvious, as well as finding the mistake and correcting it. Some techniques, like van Noord (2004) or Sagot and Villemonte de La Clergerie (2006), bring a convenient way to automatically identify forms having potentially erroneous entries in a lexicon. We have integrated and extended such techniques in a wider process which, thanks to the grammar ability to tell how these forms could be used as part of correct parses, is able to propose lexical corrections for the identified entries. We present in this paper an implementation of this process and discuss the main results we have obtained on a syntactic wide-coverage French lexicon.

XML rules for enclitic segmentation

Sentence word segmentation is an important task in robust part-of-speech (POS) tagging systems. In some cases this is relatively simple, since each textual word (or token) corresponds to one linguistic component. However, there are many others where segmentation can be very hard, such as those of contractions, verbal forms with enclitic pronouns, etc., where the same token contains information about two or more linguistic components.

Information retrieval and large text structured corpora

Conventional Information Retrieval Systems (IRSs), also called text indexers, deal with plain text documents or ones with a very elementary structure. These kinds of system are able to solve queries in a very efficient way, but they cannot take into account tags which mark different sections, or at best this capability is very limited. In contrast with this, nowadays, documents which are part of a corpus often have a rich structure. They are structured using XML (Extensible Markup Language)[1] or in some other format which can be converted to XML in a more or less simple way. So, building classical IRSs to work with these kinds of corpus will not benefit from this structure and results will not be improved. In addition, several of these corpora are very large and include hundreds or thousands of documents which in turn include millions or hundreds of millions of words. Therefore, there is the need to build efficient and flexible IRSs which work with large structured corpora.