Socially Guided Allocation of Attention in the Memory Encoding of Spoken Language

Abstract:

It is now well-established that memory representations of words are acoustically rich. Alongside this development, a related line of work has shown that the robustness of memory encoding varies widely depending on who is speaking. In this dissertation, I explore the cognitive basis of memory asymmetries at a larger linguistic level (spoken sentences), using the mechanism of socially guided attention allocation to explain how listeners dynamically shift cognitive resources based on the social characteristics of speech.

This dissertation consists of three empirical studies designed to investigate the factors that pattern asymmetric memory for spoken language. In the first study, I explored specificity effects at the level of the sentence. While previous research on specificity has centralized the lexical item as the unit of study, I showed that talker-specific memory patterns are also robust at a larger linguistic level, making it likely that acoustic detail is fundamental to human speech perception more broadly. In the second study, I introduced a set of diverse talkers and showed that memory patterns vary widely within this group, and that the memorability of individual talkers is somewhat consistent across listeners. In the third study, I showed that memory behaviors do not depend merely on the speech characteristics of the talker or on the content of the sentence, but on the unique relationship between these two. Memory dramatically improved when semantic content of sentences was congruent with widely held social associations with talkers based on their speech, and this effect was particularly pronounced when listeners had a high cognitive load during encoding. These data collectively provide evidence that listeners allocate attentional resources on an ad hoc, socially guided basis. Listeners subconsciously draw on fine-grained phonetic information and social associations to dynamically adapt low-level cognitive processes while understanding spoken language and encoding it to memory. This approach positions variation in speech not as an obstacle to perception, but as an information source that humans readily recruit to aid in the seamless understanding of spoken language.