The Semantics of Corrections

At first glance, correcting mistakes while speaking seems like purely a performance issue, of little interest to theoretical linguists. We think this phenomenon deserves a second glance. (1) Anders, uh ..., [Andrew]F broke the window. We will refer to the portion of the utterance preceding the disfluency as the anchor, and the portion following it as the correction. A first analysis might be that the anchor plays no role in the interpretation of the correction. We argue that this initial assumption deserves to be revisited based on data such as (2) and (3). (2) Every boy brought, uh ..., [should]F have brought a water bottle. (3) Every boy brought, uh ..., he [should]F have brought a water bottle. Prima facie, these utterances might involve partial deletion of the anchor and incorporation of remaining material into the correction: in (2) the anchor’s partial VP is deleted and its subject attached to the beginning of the correction; in (3) the subject of the correction must also be deleted so that the correction can receive its proper interpretation relative to every boy. We argue that an account along these lines fails to capture the relation between anchor and correction. First, the anchor influences the focus structure of the correction: (4) a. ??Every boy brought a book. . . I’m sorry, I mean [should]F have brought a water bottle b. Every boy brought a book. . . I’m sorry, I mean [should]F have brought [a water bottle]F As indicated in (1)-(3), focus is obligatory on the locus of the correction. (4) demonstrates that if there are multiple loci of correction, there must be multiple foci. If the anchor were merely cannibalized for spare parts, its content could not influence the focus structure of the correction. Second, sentences like (3) are subject to the same constraints as telescoping phenomena. (3) has a counterpart in (5), a simplified version of an example from Roberts (1987): (5) Every boy walked up to the speaker. He shook her hand and returned to his seat. (6) a. *No boy walked up to the speaker. He shook her hand. . . b. *No boy brought, uh, he should have brought a water bottle. These restrictions are incompatible with a delete-and-copy account of corrections. We provide a dynamic account of correction phenomena that involves full, incremental interpretation of both the anchor and the correction, while crucially distinguishing between the interpretation of an utterance and the contribution of its assertion to the discourse, as in Anderbois et al (2015). In our account, both anchor and correction are fully interpreted, allowing the content of the anchor to play a role in the interpretation of the correction. However, only the correction is asserted. On this view, telescoping becomes possible because the entire utterance is composed via iterated dynamic conjunction, making it possible to interpret singular pronouns in the correction relative to quantifiers in the anchor. The wellformedness of the entire utterance is determined based on a relationship of focus inclusion between the anchor and the correction, namely whether there exists a possible completion of the anchor that is a member of the focus semantic value of the correction.