Scales, Alternatives, Context: Experimental Investigations into Scalar Inference

Scalar inference, the process by which we infer meanings stronger than what was explicitly said, has long been a central topic of investigation in theoretical semantics-pragmatics, as well as in psycholinguistics. Upon encountering the sentence "Mary ate some of the deep dish", for instance, hearers regularly compute the pragmatic meaning that Mary ate some, but not all, of the deep dish. The standard assumption is that the inferential process that gives rise to this result involves hearers reasoning about what the speaker could have said, but did not (Grice, 1967). Further, Neo-Gricean accounts typically assume that hearers infer the negation of informationally stronger unsaid alternatives, e.g., because form a lexical scale, and "all" is stronger than "some", hearers derive "not all" upon encountering "some" (Horn, 1972; Katzir, 2007). In addition to involving the negation of unsaid alternatives, another crucial property of scalar inference is context-sensitivity (Van Kuppevelt, 1996). That is, whether a scalar inference-enriched meaning is derived depends partially on the discourse context. For instance, given a question such as "Did Mary eat any of the deep dish?", all that matters is whether Mary ate at least some of the deep dish. Therefore, in this context, an answer of "She ate some of the deep dish" is less likely to lead to scalar inference, since "She ate all of the deep dish" is no longer a relevant alternative. This dissertation is an experimental investigation into these two crucial properties of scalar inference: alternative- and context-sensitivity. In Chapter 2, I test whether alternatives such as "all", which are important in the theoretical modelling of scalar inference, are psychologically real. In a series of semantic priming with lexical decision experiments, I demonstrate that such lexical alternatives are indeed retrieved and activated in the processing of inference-triggering utterances. In Chapter 3, I turn to the much-investigated question of whether scalar inference calculation incurs a processing cost, operationalized as increased reaction times. In a sentence-picture verification experiment, I compare scalar inference with another pragmatic inference, it-cleft exhaustivity, and demonstrate that whether inference calculation is costly is a function of the discourse context, and not whether alternative-retrieval is involved. While much research has investigated the "some but not all" inference, there exist many other scales where a set of lexical items are ordered with respect to each other in terms of their logical strength. Similarly to , also form a scale; an utterance of "The movie is good" might give rise to the scalar inference that the movie is not excellent. An important recent finding, however, is that there is in fact considerable variation across such different scales in the likelihood of inference calculation: the "not excellent" inference, for instance, is much less likely to arise than "not all" (van Tiel et al., 2016). The second half of this dissertation investigates this phenomenon of scalar diversity. In Chapter 4, I ask whether the observed variation can be explained by differences in the alternatives themselves ("all" vs. "excellent"). My findings suggest that some (but definitely not all) properties of alternatives (e.g., accessibility) do indeed predict likelihood of inference calculation, yet scalar diversity remains largely unexplained. Lastly, in Chapter 5, I investigate whether scalar diversity can be modulated by a supportive discourse context vs. grammatically encoding the negation of alternatives, and show that uniformity in inference calculation is only achieved when these two factors align. Overall, the picture that emerges from this dissertation is that pragmatic meaning arises both as a function of global properties of context, and as a function of local properties of the scalar terms themselves