The web as a resource for WSD

Why general Word Sense Disambiguation (WSD) systems are not (to the best of our knowledge) being used in real-world applications? In our opinion, some plausible reasons are: • General Word Sense Disambiguation, as an intermediate task, is frequently harder than the final application. This is probably the case of the most commonly mentioned WSD applications: mono and multilingual Information Retrieval and Machine Translation. • Different applications demand different sense distinctions and different sense granularities. • Unsupervised systems still have poor performance... • ... and supervised systems, in general, hardly have resources to be supervised with. Currently, supervised systems can only attempt an “all-words” WSD task in English, and with very scarce resources. In order to find out whether WSD systems can play a significant role in Human Language Technologies, a necessary step is to make supervised WSD algorithms applicable, finding ways of acquiring lexical information and, in particular, building training corpora at a low cost.