On the grammaticalization of negative polarity items

0 Introduction The phenomenon of polarity sensitivity is not, in general, very well understood. Why is it that certain words and idioms have the peculiar property of occurring only in negative, interrogative or conditional contexts? It is intuitively clear that negative polarity items (NPIs) are not arbitrarily distributed over the lexicon. A field-worker analyzing an unknown language would probably not ask if the word for knife is limited only to negative statements. On the other hand, indefinite pronouns, or certain modal verbs, are generally good candidates for NPIs. On the basis of these limited observations, one might venture the guess that negative polarity is a reflection of word meaning. More precisely, certain expressions are predestined to become polarity sensitive due to their semantic properties. In the case of English any, this idea has been put forth by Kadmon and Landman (1993), who argue that all its uses can be explained by reference to its lexical-semantic properties. Fauconnier's (1975, 1978) work on minimal-quantity NPIs like budge an inch or hurt a fly suggests that their occurrence in negative contexts is directly related to their pragmatic force. They have an intensifying character and are used to make universal statements. Given the indefinite meaning of the minimal-quantity NP, this goal can only be achieved in implication-reversing contexts (e.g. negation, cf. Ladusaw 1980). However, the rhetorical force of these expressions is idiomatic because it cannot in general be predicted from their literal meaning. Compare e.g. one bit, an NPI with a bit, which is not an NPI (cf. I am (not) a bit worried), but may be used to strengthen negation, and a little bit, which is also not an NPI, but differs from a bit in negative clauses. Thus I am a little bit worried is roughly equivalent to I am a bit worried. The negative counterpart of the first sentence, I am not a little bit worried, indicates a fairly strong degree of worry, while the negation of the second sentence, I am not a bit worried, expresses the absence of any worry. Arguably, these items have the same basic interpretation, but developed special uses which have to be learned separately. If this is so, then a bit, one bit and a little bit are all possible candidates for NPI-status, but only one bit has been grammaticalized as such (cf. also Bolinger 1972). In using the term 'grammaticalization', I assume that the …