Proceedings of the Annual Meeting of the Cognitive Science Society

Recent work on "scalar diversity" has shown that scalar inferences (SIs) do not form a unified class. For example, while and both form scales, hearers overwhelmingly interpret some as meaning not all, but rarely interpret pretty as meaning not beautiful. We show that SIs are correspondingly heterogeneous with respect to processing times and their sensitivity to world knowledge. In particular, hearers process SIs associated with closed class scales faster than open-class SIs or ad-hoc SIs; moreover, open-class SIs exhibit greater sensitivity to world-knowledge than their closed-class counterparts. These results, we argue, suggest the existence of at least two distinct paths for generating scalar inferences--one path is fast, automatic, and insensitive to the conversational context, while the other path is slow, deliberate, and sensitive to conversational context.