The present paper describes the way we participated to the French track of CLEF-2002. We used a morphological analyser and a syntactic chunker in order to desambiguate words and filter certain categories of words. Then we built a global index with the Lucene Indexor. Concerning the search process, we wrote boolean queries and evaluated them by the means of the Lucene query Evaluator.
We proceded as follows: - sentence segmentation, - morphological analyses for simple and compound words, - if the word is unkown, a « picking up » is tried. A rapid visual control has been made on this process. We verified the unkown words that begin by a lower case and appear more that 5 times and the unknown words that begin by an upper case and appear more that 50 times. The control shows that most of the frequent mistakes were corrected. - a syntaxic and partial parsing is applied by the means of a chunker (see www.tagmatica.com). We dont use the syntaxic informations labeled by the chunker, we just use the word level desamguisation. - the lemmatized form is given to the Lucene Indexor (see jakarta.apache.org/lucene).
We translated manually the topics into boolean expressions. Due to the fact that we indexed only lemmatized forms, we expressed the words according to their lemmatized form.
We did not use the title tag. We used only the descriptive and narrative tags.
We don't know exactly how our results can be compared with the results of the other participants. We indexed the whole ATS corpus but we did not have enough time to index the whole « Le Monde » corpus. We indexed only 70% of the corpus (we had a hardware problem with the machine).
That means that we certainly have a lot of silence compared with the other results. Concerning noise, we probably are not very noisy.