Jesús Vilares Ferro

Improving non-English web searching (iNEWS07)

This workshop attempted to promote the discussion and the research on non-English Web searching. Most search engines were first built for English. They do not take full account of inflectional semantics nor, for example, diacritics or the use of capitals. Our main aim was to discuss the additional problems faced in non-English Web queries and to suggest techniques to improve the response of searching systems. Papers related to Arabic, Basque, Farsi (Persian), Greek, Spanish, Swedish, Hindi, Bengali and other south asian languages were accepted. Conclusions were that search engines would be more effective if they took more account of the properties of individual languages, and that there is a need for more studies of real user behaviour in practical situations.

Practical NLP-Based Text Indexing

We consider a set of natural language processing techniques based on finite-state technology that can be used to analyze huge amounts of texts. These techniques include an advanced tokenizer, a part-of-speech tagger that can manage ambiguous streams of words, a system for conflating words by means of derivational mechanisms, and a shallow parser to extract syntactic-dependency pairs. We propose to use these techniques in order to improve the performance of standard indexing engines.

COLE experiments at QA@CLEF 2004 spanish monolingual track

This paper is a report on our third participation in CLEF. More precisely, this year we have participated in the Spanish Monolingual Question Answering Track for the first time. As a result we have developed a prototype of a QA system. Our prototype continues to apply the Natural Language Processing techniques we had already developed for single word conflation. In addition, the question analysis is based on complex pattern matching either over forms, part-of-speech tags or lemmas of the words involved. Regarding the search for relevant parts of documents containing the required answer, we use conventional IR techniques, whilst the extraction of the answer from the relevant parts of documents is again based on pattern matching.