How intonation contrains pragmatic inference - eScholarship
John M. Tomlinson, Lewis A. Bott · Proceedings of the Annual Meeting of the Cognitive Science Society · 2013
How intonation constrains pragmatic inference John M. Tomlinson, Jr. ([email protected]) Zentrum fur allegemeine Sprachwissenschaft (ZAS) Berlin 10117 Germany Lewis Bott ([email protected]) School of Psychology, Cardiff University, Tower Building Cardiff CF103AG UK Abstract In this paper, we present two experiments that investigate how intonation can constrain pragmatic inference. While prior research has shown that intonation can increase the likelihood of an inference being made, less is known about how it affects the mechanisms involved in processing of inferences. In the first experiment, listeners had more direct mouse paths towards target responses for stronger interpretations after hearing utterances with referents with pitch accents than without. In the second experiment, we replicate the finding of the first study and found more direct mouse paths towards weaker interpretations after hearing de-accented referents Our findings suggest that intonation constrains the online processing of pragmatic inference by increasing the availability of stronger interpretations. Keywords: Experimental Pragmatics, Psycholinguistics, Prosody, Language Comprehension, Mouse-tracking. Introduction The rapid nature of human communication requires speakers and listeners to be as efficient as possible. To help achieve this, listeners often rely on context to help disambiguate between different linguistic structures and meanings. However, often what a speaker intends to say is not always directly retrievable from a linguistic form; rather listeners must infer it. One issue concerning pragmatic inference is whether the processor can keep up with the task demands of conversation. Some have argued that linguistic inference must be quick and “cheap” (Levinson, 2000; Piantidosi, Tily, & Gibson, 2012), however others have experimentally demonstrated that some linguistic inferences can be quite costly in terms of processing (Bott & Noveck, 2004; Huang & Snedeker, 2009). In this paper, we discuss one aspect of the linguistic signal that has the potential to make certain costly pragmatic inferences quicker and more efficient: prosody. We report the findings from two experiments that test different accounts about how prosody affects the processing of pragmatic inferences. Pragmatic inferences and language processing Traditionally, linguists have treated pragmatic inferences as the interpretative process in which a speaker must reconcile how speaker’s literal sentence meaning differs from his or her intended meaning. Grice (1967) initially distinguished between two types of pragmatic inferences (particularized implicatures): conventional implicatures and conversational implicatures. Conventional implicatures roughly amount to inferences about a speaker’s intended meaning that can be made without accessing the conversational context. Conversational implicatures, however, require that listeners must first consider the literal sentence meaning, compare it against the context and then potentially enrich it in order to arrive at a speaker’s intended meaning. Neo-Griceans have proposed an inference type that falls somewhere between Grice’s original distinction: default inferences (Levinson, 2000). Default inferences are inferences that are computed on every occasion, but can be cancelled later. Always deriving the inference avoids costly pragmatic computations that would delay obtaining the speaker’s intended meaning. According to this process, inferences are heuristic-based and therefore can become “cheap” in regards to processing resources. Researchers in experimental pragmatics have tested whether certain implicatures classes are indeed understood as default inferences. One case that has caused some debate is the case of scalar implicatures. For these inferences, listeners can choose between either a weak or a strong interpretation depending on what they think the speaker intended to communicate. For example, a sentence such as “I drank some of my friend’s beers last night” could either be taken to mean that I drank (1) at least one (and possibly all) of the beers or (2) at least one and not all of the beers. The difference between interpretations (1) and (2) is that to interpret “not all” in (2), the listener must infer that had the speaker meant “all,” they would have said so. In other words, the listener would need to make a pragmatic inference to access the stronger interpretation. Several experimental studies have shown that understanding upper bound meanings of some, as in (2), takes substantially longer than the meaning in (1) (Bott & Noveck, 2004; Huang & Snedeker, 2009). As such, a default implicature account of scalar implicatures is not borne out by the majority of these findings because of the processing cost for (2). However it might be the case that this processing cost can be diminished in the right context (Grodner, Klein, Canbary, & Tannenhaus, 2010; Degen & Tannenhaus, 2011), i.e. making (2) more available or active earlier on in processing. Our studies seek to examine how processing costs can be diminished and what this means for processing accounts of pragmatic inferences. Specifically, we examine