Sense Tagging: Don't Look for the Meaning but for the Use

Automatic sense-tagging is one of the next challenges that corpus linguists have to face. So far, results are modest, despite numerous efforts, and sense-tagging appears as a vexing task. Difficulties stem from various sources, and in particular the extraction of disambiguating information from the context. However, one of the main problems comes upstream of the disambiguating process and lies in the sense inventory itself. Most tagging efforts use traditional dictionaries as the reference sense list, or machine-oriented resources such as WordNet which do not differ significantly from traditional dictionaries in terms of sense division. This paper shows that human taggers perform very poorly when they are given a traditional dictionary as reference, and that machines should therefore not be expected to perform better using the same kind of resource. A detailed analysis reveals the lack of distributional criteria in dictionary entries: traditional dictionaries are chiefly concerned with the definition of meaning, and not with the surface clues (syntactic, collocational, etc.) that are required to match a given sense with a given corpus occurrence. It is argued that no fundamental progress can be made until largescale lexical resources have been built that incorporate extensive distributional information.