The neural computation of scalar implicature

Abstract

Language comprehension involves not only constructing the literal meaning of a sentence but also going beyond the literal meaning to infer what was meant but not said. One widely studied test case is scalar implicature: The inference that, e.g., Sally ate some of the cookies implies she did not eat all of them. Research is mixed on whether this is due to a rote, grammaticalised procedure or instead a complex, contextualised inference. We find that in sentences like If Sally ate some of the cookies, then the rest are on the counter, that the rest triggers a late, sustained positivity relative to Sally ate some of the cookies, and the rest are on the counter. This is consistent with behavioural results and linguistic theory suggesting that the former sentence does not trigger a scalar implicature. This motivates a view on which scalar implicature is contextualised but dependent on grammatical structure.

Keywords:

• scalar implicature
• pragmatics
• ERP

Acknowledgement

The authors thank Miki Uruwashi and for help with data acquisition and attendees at AMLaP 2012, CogSci 2013 for comments and suggestions.

Notes

1. This inference does not necessarily go through if the speaker does not know whether John ate every single one (maybe he was still eating when she left the room). While there is evidence that listeners are sensitive to the speaker’s knowledge state when calculating implicatures (Bergen & Grodner, 2012; Bonnefon et al., 2009; Breheny et al., 2013; Goodman & Stuhlmüller, 2013), it is not clear whether implicature processing involves explicit mental state reasoning or simply approximates it. That is, like gazelle stotting or communication between bees, the fact that the problem solved by scalar implicature is inherently social does not mean that the underlying mechanisms explicitly invoke mental state reasoning.

2. If a specific alternative statement has been made contextually relevant, other implicatures may apply:

(5) Alfred: Did John eat some of the cookies, and does he like scuba diving?Beatrice: John ate some of the cookies. Note that regardless of what Alfred said, Beatrice’s statement implies that John did not eat all of the cookies. Additional inferences – e.g., about whether John likes scuba diving – may depend on what exactly Alfred asked.

3. Like the lexical alternatives account, the grammatical theory of implicature (Chierchia et al., 2012) operates over sub-propositional units and invokes lexical scales. However, on this proposal, the occurrence of a scalar implicature is sensitive to the grammatical context in which the scalar term appears.

4. See Note 3 above.

5. Note that these accounts differ along a number of other dimensions, such as whether scalar implicature involves grammatical processing. The last two decade has witnessed an explosion of work on scalar implicature within theoretical linguistics, resulting in a rich literature and detailed theories. Many of these distinctions are beyond the scope of the present work. We refer the interested reader to Chierchia et al. (2012), Sauerland (2012), and Geurts (2010).

6. Panizza, Chierchia, and Clifton (2009) report an eyetracking-while-reading study that manipulates entailment context. But critically this work focuses on the interpretation of number words (“two” means two and not three). Whether number interpretation involves scalar implicature is controversial (Breheny, 2008), and a variety of behavioural paradigms have found categorical differences in how numbers and scalar quantifiers are interpreted (Huang & Snedeker, 2009a, 2009b, 2011; Huang, Spelke, & Snedeker, 2013; Marty et al., 2013), suggesting that we cannot generalise from one to the other.

7. Note that this requires that the semantic relatedness to prior context be controlled (cf. Noveck & Posada, 2003; see also Kounios & Holcomb, 1992).

8. The choice of threshold (e.g., 1.96) affects the type of clusters found – low thresholds are better at detecting broadly extended but weak effects – but it does not affect robustness to multiple comparisons. Other thresholds resulted in similar findings.

9. These and other analyses in this section utilised mixed effects models with subjects and items as random effects and with maximal random slopes design. P values are derived from model comparison.

Additional information

Funding

This work was supported by NDSEG, NSF GRFP and NIH NRSA HD072748 fellowships to JKH, NICHD R03HD071094-01A1 to AK, and NSF-BCS 0921012 to JS.