Reflection: Experimental pragmatics and defaults in utterance processing

Levinson (2000) has argued that relevance theory takes speaker meaning to be only a “matter of nonce inference” (2000: 25), that is, inferences which arise on a particular occasion, and that it ignores “default inferences” arising from “general expectations about how language is normally used” (2000: 23), including generalized conversational implicatures (GCIs). However, Levinson’s treatment of defaults as arising from the localized computation of GCIs, that is, at the time at which a particular scalar-inducing lexical item, such as some, or, sometimes and so on, occurs in an utterance, has not received strong support to date from psycholinguistic experiments. There is now significant experimental evidence mounting against the claim that scalar implicatures arising from some (i.e. “not all”) and the disjunctive particle or (i.e. “not and”), for instance, are computed locally by default (Katsos 2012). Brehney, Katsos and Williams (2006), for instance, tested whether some of the F’s implicated “not all the F’s” and “X or Y” implicated “either X or Y but not both” across a range of different contexts. What they found was that such implicatures were only generated when explicitly warranted by the context, that is, when something in the context or co-text made the scalar implicature relevant to the interpretation of the utterance in question. Such experimental work has sometimes been interpreted by Relevance theorists as vindicating their view that scalar implicatures do not arise by default, but rather are only generated when the weaker term on a scale fails to meet the hearer’s expectation of relevance. Others, however, regard such results as only falsifying the local default view and not necessarily supporting the relevance theoretic account (Garrett and Harnish 2009; Horn 2009). There has thus been an increasing focus on defaults for contexts, where default meaning representations are analyzed as being contextually tied to the level of utterances and beyond (Jaszczolt 2005). One difficulty facing such experimental work, however, is that the same basic experimental methods can sometimes give rise to quite different results, depending on the details of the experimental protocol or conditions (Krifka 2009). Thus, while experimental pragmatics offers considerable promise in addressing our understanding of utterance processing, there still remains considerable work to be done in order to better understand what intuitions or processes we are actually testing through such methods.