Acquisition automatique des termes: L'utilisation des pivots lexicaux specialises (French text).

摘要

The research undertaken for this thesis is part of knowledge acquisition from texts; it focuses more particularly on term acquisition. Our work led to the development of TermoStat, a piece of software dedicated to testing our methodology for automatic term acquisition in an industrial environment.; To carry out term recognition in English, TermoStat relies on statistical techniques to compare the lexicon of a technical corpus (analysis corpus) to one of a non-technical corpus (reference corpus). The object of this comparison is to establish a list of specialised lexical pivots (SLP). The SLPs correspond to the lexical items that have an abnormally high frequency in the analysis corpus as compared to the reference corpus.; SLPs are used as a starting point for the automatic acquisition of terms, which relies on the concept of term frontiers. Using specialised lexical pivots allows TermoStat to focus its analysis on parts of documents that have a particular behavior. This pinpointing of relevant information allows TermoStat to only look at the immediate context of SLPs.; In order to maximize the quality of the results, we put forward a weighting index to capture the terminological potential of candidate terms (CT). The index, called iTer, includes various contextual clues as observed in the corpus (frequency, length, etc.). The first half of the sorted list of CTs obtained from our analysis corpus with TermoStat had a precision of 86.8%.