Natural Language Morphology Representation

Information retrieval is a wide ranging area that deals with storage and retrieval of any kind of media. Not enought interest has been given over the past year in using natural languages morphology to retrieve text documents given a text query from a user. This thesis explores the different uses of morphological analysis to infer additional meaning from a given word in order to improve information retrieval systems. Considerations on morphological analysis for information retrieval is provided for French and English. A presentation of an alternative to stemming called Lightweight Morphology is given, which from a set of pattern matching rules, user defined rules and an exception tables can produce word expected inflections and derivations. Finally, in order to properly measure the interest of morphological analysis in text retrieval, a new measure to benchmark information retrieval systems, called "differential recall", is introduced. Lightweight morphology is compared to the Porter stemmer using state-of-the-art TREC benchmark and using differential recall measure.