Information Integration in Online Modulation of Pragmatic Inferences during Language Comprehension

Upon hearing a scalar adjective in a definite referring expression such as " the big…, " listeners typically interpret the speaker to be referring to an item in a contrast set denoted by the scalar adjective, such as a big glass in the context of a smaller glass. This inference guides online visual search for a referent, resulting in anticipatory eye-movements to the target object when a contrast item is present. Recent studies have suggested that these inferences are malleable, such that the likelihood of an inference reflects environmental statistics of felicitous language use. These results point to distributional learning about the contexts in which scalar adjectives occur (i.e., that scalar adjectives are typically used in contrastive settings) as a key mechanism underlying the inference process. In a series of eye-tracking experiments, we explore the nature of the evidence necessary for distributional learning in language comprehension, focusing on 1) modulation of pragmatic inferences based on bottom-up information and 2) the speaker-specificity of this process. Listeners were exposed to speakers who either used size adjectives felicitously (e.g., " the big dog " in the context of a big and a small dog) or infelicitously (e.g., " the big dog " in a context with one dog). We find that massive exposure to bottom-up evidence can trigger modulation of contrastive inferences but there was little evidence of speaker-specificity. These findings suggest that listeners may evaluate observed evidence against their prior assumptions about pragmatic language use. 3 In communication, referring expressions are shaped by properties of the intended referent as well as the context that the referent appears in. Definite referring expressions are thought to uniquely identify an intended referent within the relevant context, broadly Pechmann, 1989). In a context with a single dog, the bare noun phrase " the dog " would suffice to achieve reference. By contrast, in a situation with multiple dogs, the speaker would need to provide additional information, for example, through the use of a modified noun phrase, e.g., " the fluffy dog " Different classes of modifiers vary in their degree of contextual sensitivity. that scalar modifiers such as tall, big, and skinny are often prompted by the presence of a scalar contrast in the relevant context (e.g., a context with a big dog and a small dog), and are much less likely in contexts with only a single member of the class denoted by the noun …