Evaluation of French and English MeSH Indexing Systems with a Parallel Corpus

OBJECTIVE This paper presents the evaluation of two MeSH indexing systems for French and English on a parallel corpus. MATERIAL AND METHODS We describe two automatic MeSH in-dexing systems - MTI for English, and MAIF for French. The French version of the evaluation resources has been manually indexed with MeSH keyword/qualifier pairs. This professional indexing is used as our gold standard in the evaluation of both systems on keyword retrieval. RESULTS The English system (MTI) obtains significantly better precision and recall (78% precision and 21% recall at rank 1, vs. 37%. precision and 6% recall for MAIF ). Moreover, the performance of both systems can be optimised by the break-age function used by the French system (MAIF), which selects an adaptive number of descriptors for each resource indexed. CONCLUSION MTI achieves better performance. However, both systems have features that can benefit each other.