Doctoral Dissertation Research: Prosodic Encoding of Informativity

University Of Southern California, Los Angeles CA

Investigators

Abstract

Prosody plays a vital role in human communication. The 'melody' of words can convey information that the words themselves do not convey. For instance, a question like 'what did you do?' can express friendly curiosity or sharp disapproval, among other attitudes, depending on the prosody used by the speaker. Being able to use and understand prosodic cues is fundamental to social interaction, and prosody is one of the first things that newborn babies learn about their mother tongue. Considering how naturally prosody comes to us, it is surprising that we do not yet fully understand what influences the prosody of an utterance. One factor that has been claimed to play a role is informativity. For example, the word 'apple' is more informative in a sentence like 'the Little Mermaid ate an apple' than 'Snow White ate an apple', because it is more unexpected to associate apples with the Little Mermaid than Snow White. Broadly speaking, it seems that the more informative a word is, the more prominent it is prosodically--a more informative word is likely to be longer and louder than a less informative word. However, to understand what actually contributes to this seemingly intuitive notion, further research is needed. This research has the potential to inform other fields where prosody plays an important role--including first language acquisition, second language comprehension, language impairments and interpersonal communication--and will deepen our understanding of what information can be extracted from prosody. This ability is particularly important for listeners with a limited vocabulary and grammar such as infants, second language learners, and people with aphasia. This project strives to make connections between different research traditions on prosody and informativity. In prior work, the research group showed that prosody is shaped by a complex interplay between two kinds of informativity: (i) discourse information (e.g. whether a word is being mentioned for the first time) and (ii) statistical factors (e.g. whether a word is a frequently used word or an infrequent word). The researchers hypothesize that these results can in fact be explained by a unifying factor, namely the degree of difference between what actually happened and what the speaker had expected--in other words, how surprised the speaker is. Let us consider 'Snow White ate an apple' again. As mentioned above, the word 'apple' in this sentence should have fairly low prosodic prominence because it is highly predictable in the context. However, imagine the speaker said this sentence in response to someone asking 'what fruit did Snow White eat?' In this case, the word 'apple' might be pronounced with more prosodic prominence, because the speaker might be surprised by the conversational partner's lack of knowledge about a famous fairytale. Building on this idea, a series of experiments will be conducted to further test the extent to which the speaker's 'surprisal' influences his/her prosody. In particular, the experiments will investigate whether speakers take their conversational partner's perspective into account, whether speakers adapt to their partner's characteristics and changing knowledge, and whether non-linguistic information plays a role. These studies will shed light on still-debated questions about humans' perspective-taking abilities, our ability to track what others know and don't know, as well as the relationship between language and other aspects of cognition, such as vision. As a whole, this project aims to improve our understanding of prosody, one of the most fundamental aspects of spoken communication.