On similarity of word senses in explanatory dictionaries

Quality machine translation (MT) as well as of some other applications (such as information retrieval, IR) require word sense disambiguation (WSD) in the source text. However, WSD is only possible if the word senses specified in the dictionaries are really different and clearly distinguishable. We investigate the semantic closeness of different senses of the same word in a Spanish explanatory dictionary. We define the closeness between two senses as the relative number of equal or synonymous words in their definitions. We show that a considerable part of dictionary definitions (ca. 90%) are different enough to be distinguished in MT and IR. On the other hand, a considerable number of definitions (ca. 10%) are too similar to be reliably distinguished. These results suggest that MT and IR can take advantage of WSD algorithms, but for this, the similar senses reflecting too subtle meaning nuances should be clustered together to form coarser but easier distinguishable senses. The proposed method for detecting too similar senses can be incorporated into the lexicographer’s workbench to be used in development and improvement of dictionaries.