Why Ambiguity?
Ariel Cohen Ben-Gurion University arikc@bgu.ac.il

All human languages are ambiguous. Ambiguity is not restricted to some special constructions that linguists are fond of discussing, but is quite ubiquitous. It is hard, in fact, to find a sentence that is not ambiguous. This fact is all too familiar to computational linguists: “One often hears in computational linguistics about completely unremarkable sentences with hundreds of parses, and that is in fact no exaggeration.” (Abney, 1996). The question is, simply, why? Why is language ambiguous? Krifka (2002) raises a similar question concerning vagueness. He shows that there are good reasons for language to allow, and even encourage vagueness. Vagueness, however, is not ambiguity. When we are told that the theater is far from here, we may be unsure as to the precise distance; but we know what the speaker intends, and we can draw inferences based on this, e.g., that we should take a cab instead of walking to the theater. But with an ambiguous term, the intended meaning is not merely insufficiently specified; it is not known, until the term is disambiguated. When we hear that John has a kid, we draw very different inferences if John is a father or if he owns a young goat. Thus, the advantages of vagueness do not seem to apply to ambiguity, and we are back to the question: why ambiguity? Some may say that it doesn’t matter. Humans possess very powerful mechanisms for disambiguation; these mechanisms resolve ambiguities, so that we are not normally aware of them. But this is not really an answer: these powerful mechanisms are there because language is ambiguous. If it weren’t, they wouldn’t have developed. Moreover, these mechanisms must require extra processing time. We know that all senses of an ambiguous word are accessed first, and only then does disambiguation occur (Swinney, 1979). If language were not