Abstract: 

Various classes of functional terms---quantifiers, exceptives, comparatives, auxiliaries, attitude verbs, etc.---are subject to distributional constraints that don't seem amenable to syntactic or complexity based characterizations, yet can be systematically captured in semantic terms. These constraints have been taken to support the `Logicality of Language', i.e.,  the view that the human language system includes a computational system that can identify and automatically filter out LFs which are informationally trivial. Proponents of Logicality have proposed implementations that try avoid over-generating predictions of unacceptability for various kinds of superficial tautologies and contradictions. Less appreciated, however, is that unless we substantially constrain the kinds of logical operators that can be appealed to, triviality-based analyses risk collapsing into un-explanatory, direct encodings of the observed restrictions. Sensitive to this challenge, in this talk I will explore an approach to triviality-based constraints which rests on the independently motivated hypothesis that the language system has access to a limited repertoire of covert exhaustification operators, akin to “only" and “even", whose function is to optimize certain trade-offs, inherent in language-based communication, between computational complexity and informativity. The view I will explore builds on proposal that many classes of logical terms include morphologically marked duplicates which differ from their non-marked counterparts only in that they call for obligatory exhaustification. Depending on the specific syntactic configuration in which they occur, this association can reduce ambiguity and increase informativeness---but it can also generate trivial readings that result in strict unacceptability. I will motivate this approach by deriving some well-known distributional constraints, and showing that it descriptively outperforms Lexicalist alternatives and can be straightforwardly combined with a principled solution to the over-generation problem. I will also present two ways in which this approach helps us understand why the human language system is designed so as to automatically filter out certain informationally trivial structures.