The Crucial Role of Compositional Semantics in the Study of Reasoning

Reasoning work in both psychology and computer science is often conceived in terms of drawing inferences over strings. This approach works reasonably well in some cases, but it is sometimes dramatically misleading due to the pervasive presence in natural language of interpretive factors that are not reflected transparently in surface form. The problem includes not only the well-known issues of ambiguity, polysemy, and sensitivity to pragmatic context, but also a much less-studied factor: compositional semantic interactions where one expression in a sentence influences the interpretation of another. In two case studies, I show that careful attention to natural language semantics is crucial in understanding valid and invalid patterns of reasoning, and in avoiding fallacies.