Lexical Inference and the Problem of the Long Tail
David D. McDonald · 2009
No one’s semantic lexicon is ever complete. People cope through the adroit use of context. Machines should do the same. In this position paper I suggest that a large part of this context consists of deep knowledge about adjectives, adver-bials, and other linguistic forms whose meaning cuts across subject domains. This is a new direction for research that complements the large effort on ontology induction and could accelerate work on reading for reasoning. The long tail of words Even a highly literate reader will occasionally encounter a word that they don’t know or have only encountered inside a frozen idiom. Anyone with children will appreciate their ability to adapt well-known words to new purposes. This is Zipf’s law in action. There is no end to the low frequency tail of words, word-senses, and turns of phrase