Quantifying semantic and pragmatic effects on scalar diversity

Scalar inference (SI), the process by which we systematically infer meanings stronger than what was explicitly said, has long been a central topic of investigation in semantics-pragmatics. A recent experimental finding that has generated interest is scalar diversity: that the robustness of SI calculation varies across lexical scales. For instance, the some but not all SI is much more likely to arise than good but not excellent. In this paper, we take a first step toward more rigorously quantifying the observed variation across scales using relative entropy. We then turn to the question of how factors independent of scalar diversity can make SI calculation both more likely and more uniform. We find that a supportive discourse context and overt exhaustification with the focus particle only both increase inference rates and reduce variation across scales, with the effect of only being stronger. However, there still remains a lot of scalar diversity; only when we combine context with semantic exhaustification do we find uniformity across lexical scales