In lieu of an abstract, here is a brief excerpt of the content:

  • When Are Downward-Entailing Contexts Identified? The Case of the Domain Widener Ever
  • Charles Clifton Jr. and Lyn Frazier

1 When Is the Monotonicity of a Context Identified?

It has been known for some time that negative polarity items (NPIs) are often licensed in downward-entailing (DE) contexts—contexts that support inferences from sets to subsets (Ladusaw 1979). For example, the NPI ever is licensed in the negative (DE) context (1a) but not in the non-DE context (1b).

  1. 1.

    1. a. Josh didn't ever play chess.

    2. b. *Josh ever played chess.

We examine sentences in which the NPI is eventually seen to be licensed, but which contain a tempting initial misanalysis under which the NPI does not appear to be licensed. We ask two questions: (a) Are computations about the licensing of an NPI done locally and online, or are they only done in the global context of a full sentence? and (b) If such local computations are made, does a temporary apparent failure of licensing result in lowered acceptability for a globally acceptable sentence?

There is evidence that the DE contexts that support an NPI are used word by word as a sentence is read to guide some aspects of processing. Scalar implicatures (e.g., some implicates some but not all) are more likely to be reported if the scalar term occurs in a non-DE context than in a DE context (Chierchia 2004, Chierchia, Frazier, and Clifton 2009, Schwarz, Clifton, and Frazier, to appear). Panizza, Chierchia, and Clifton (2009) provide eye-tracking evidence that computation of scalar implicatures (for numerals) is done incrementally, [End Page 681] immediately when the scalar term is read (which they take to support claims that implicatures are computed locally rather than globally as in Gricean approaches to implicatures; Geurts and Pouscoulous 2009, Grice 1989, Russell 2006).

Event-related potentials (ERPs) have also been used to study the processing of NPIs. Absence of a licensor such as a negative morpheme has been reported to result in the appearance of an N400 (a marker associated with semantic predictability/anomaly) as well as a late positivity (perhaps a P600, often taken as a marker of syntactic processing difficulty) to an NPI, indicating that the licensing is done online. Interestingly, the presence of a licensor in a structurally inappropriate position also reduces the size of the N400, though not as much as a licensor in a structurally appropriate position (e.g., Drenhaus, Saddy, and Frisch 2004, 2005, Drenhaus et al. 2006, Saddy, Drenhaus, and Frisch 2004, Vasishth et al. 2008). Self-paced reading data do not show an immediate effect of spurious licensors, however, suggesting that the ERP effects might be due to semantic associations between words—a factor the N400 is known to be sensitive to (Xiang, Dillon, and Phillips 2009, though further assumptions are needed to address the effect of a spurious licensor on the P600).

2 Incremental versus Delayed Computations: The Case of the Domain-Widening Negative Polarity Item Ever

To investigate when the monotonicity (downward-entailingness or upward-entailingness) of a context is identified, we examined the NPI ever, which has been analyzed as a domain widener (Chierchia 2006), as illustrated in (2).

  1. 2.

    1. a. Ella didn't go to Paris.

    2. b. Ella didn't ever go to Paris.

Sentences without ever (2a) may be used to make a claim about a particular time period under discussion (e.g., last week), and thus (2a) might be true even if uttered in a situation in which Ella went to Paris at some time in the past. By contrast, ever in (2b) serves to widen the domain, and thus (2b) could not be used to make a true assertion if Ella went to Paris at some point in the past. Because ever widens the domain, a sentence containing ever does not readily tolerate exceptions.

If NPIs such as ever are licensed semantically by DE contexts and monotonicity computations come into play only in terms of the global context of the entire utterance, then the nature of the context, DE or non-DE, should be computed with respect to the final syntactic analysis of the sentence. A temporary misanalysis, as occurs in a "garden path...

pdf

Share