Abstract

This paper describes a new resource, the CommitmentBank, developed for the empirical investigation of the projection of finite clausal complements. A clausal complement is said to project when its content is understood as a commitment of the speaker even though the clause occurs under the scope of an entailment canceling operator such as negation or a question. The study of projection is therefore part of the study of commitments expressed by speakers to non-asserted sentence content. The content of clausal complements has been a central case for the study of projection, as there is a long-standing claim that clause-taking predicates fall into two classes—factives and nonfactives—distinguished on the basis of whether the contents of their complements project. This claim identifies the embedding predicate as the primary determinant of the projection behavior of these contents. The CommitmentBank is a corpus of naturally occurring discourses whose final sentence contains a clause-embedding predicate under an entailment canceling operator. In this paper, we describe the CommitmentBank and present initial results of analyses designed to evaluate the factive/nonfactive distinction and to investigate additional factors which affect the projectivity of clausal complements.