Approximate interpretation of number words

A pragmatic theory of the approximate vs. precise interpretation of round number words such as one hundred vs. ninety-seven is developed that improves on the previous account of Krifka (2002) insofar as it does not postulate a general bias for approximate interpretations. It is shown that under a speaker preference for number expressions that are short and/or refer to cognitively salient values on scales, approximate interpretations are favored for these expressions even if there is no general bias for them. Also, it is shown that there is a general tendency for values on coarse-grained scales to be denoted by shorter expressions, which can be seen as evidence of optimization of language. Finally, evidence is discussed that speakers may preferentially select those messages for which an ecomomic code is available.