A New Theory of Scalar Reasoning

My dissertation explores the linguistic and conceptual underpinnings of scalar reasoning as manifested in language. The first part of the dissertation presents a series of eye-tracking experiments probing the online interaction of linguistic and psychological processes in the generation and restriction of implicit alternatives invoked by only. The latter part of the dissertation deals with number concepts; specifically, I provide a novel characterization of the conceptual representations underlying number words like two and five. I also address how these meanings may be integrated with language, resulting in the range of attested (scalar and non-scalar) interpretations. These two case studies of only and number serve as the empirical foundation for an updated theory of scalar reasoning that aims to provide a psychologically-informed linking hypothesis between language and deeper levels of conceptual representation. Specifically, my theory provides an algorithm for mapping from linguistically-encoded propositional schemas to a class of reasoning structures called Scalar Models, which are argued to be utilized for generating scalar inferences. Crucially, while scalar models may be more or less straightforwardly constructed from language, they are ultimately extra-linguistic representations, and as such, inferences generated by means of such models will be sensitive to relevant social, conceptual, and statistical information available to the reasoner. This theory is intended as an alternative to purely linguistic theories of scalar reasoning. In terms of empirical coverage, my theory is able to account for a range of linguistic inferences that are difficult or impossible for other existing approaches to capture in their current form.

That we mentally represent perceptually-based as well as more abstract attributes as scalar dimensions (eg. spatial extent, numerosity, tallness, desirability) is well-established in the linguistic and psychological literatures. Recent developmental findings indicate that the ability to not just represent scalar dimensions, but to even map between distinct co-occurring scalar dimensions, is present from early infancy. Elsewhere, within the sub-field of linguistic semantics, the construct of a Scale is invoked, most commonly, for modeling the meaning of scalar adjectives as well as comparative and superlative constructions. I suggest that the ability to represent and manipulate ordered domains comes into play not just for the interpretation of scalar adjectives and so on, but also as a strategy for reasoning about alternative possibilities. Specifically, just as we might represent a domain of objects or individuals as being ordered with respect to some dimension (e.g. tallness: 'Bill is taller than John'), alternative possibilities may also be organized in this way, facilitating subsequent (transitive) inference over such domains. Based on a range of linguistic evidence and findings from infant research, I furthermore argue that such scalar models selectively encode positive monotonic relations between two or more causally-linked variables.

There has been a wealth of evidence in recent years from separate fields in the cognitive sciences highlighting the sensitivity of our inferential capacities to extra-linguistic factors such as conceptual and experiential knowledge, normative expectations, and higher-order beliefs about interlocuters' goals and epistemic states, bringing to the forefront major gaps in our understanding of how the generative machinery of language interacts with extra-linguistic domains of cognition. This work aims to be a step toward filling this gap.