Linguistics Seminar Talk - Ciyang Qing

What do large-scale experimental data and corpus examples tell us about the relation between neg-raising and question embedding?
Ciyang Qing (joint work with Floris Roelofsen)
Some clause-embedding predicates, such as "know," are compatible with both declarative and interrogative complements (e.g., "Alice knows that/whether it is raining"), whereas predicates such as "believe" are (canonically taken to be) incompatible with interrogative complements (e.g., "Alice believes that/*whether it is raining"). There are several generalizations in the literature about the relation between a predicate's semantic properties and its compatibility with interrogative complements. However, such proposals are based on a handful of predicates and paradigmatic examples. This raises questions about their generalizability and validity.
Recently, based on large-scale experimental data and corpus examples, White (2021) challenges several existing generalizations and concludes that they “should be jettisoned altogether.” In this talk, I focus on the relation between neg-raising and question embedding as a case study. I first point out that a critical auxiliary assumption White uses to derive the prediction of the existing generalization on the quantitative experimental data is not warranted, and suggest that examining each potential counterexample is a more direct and informative way to assess the generalization. Doing so reveals further complications in the experimental data and shows that many apparent counterexamples do not in fact invalidate the existing generalization. I then examine White’s apparent counterexamples from corpus data, in particular their meanings and how they are compositionally derived. I argue that such examples can be plausibly accounted for by independently motivated additional mechanisms, and therefore are not genuine counterexamples, either. In light of these, without providing an alternative and thoroughly establishing its advantage over existing generalizations (and their derivations), it is premature to conclude that the latter should be jettisoned altogether. Finally, I discuss the general implications of this case study on the uses of experimental and corpus data in semantics.