Chigusa Kurumada, Ph.D. - US grants

Successful communication requires not just choosing the correct words and grammar, but also emphasizing the right words, pausing in appropriate locations, and using appropriate intonation and rhythm. This aspect of language is called prosody and it can convey important information about the syntactic structure of a sentence, which words in the conversation are given and new, and which aspects of the sentence are important. Understanding the rules for how prosody is used is important for building artificial speech systems that produce and understand natural language, designing interventions for individuals with speech disorders, and creating curriculum for language learners. A challenge for the field is the fact that the acoustic cues that signal which words are emphasized and where prosodic boundaries occur are highly variable, changing across both speakers and contexts. This variability is pervasive, yet researchers know very little about how the brain accommodates it.

The special theme of this conference is understanding how speakers navigate pervasive variability in how prosody is realized in natural speech. Prosodic information varies significantly due to physiological and language-internal factors, dialect, idiosyncratic speaker preferences, and speech context. As a result, prosodic features lack acoustic cues that are constant across contexts. Nonetheless, human language users can readily use prosodic information in conversation. This puzzle constitutes one of the fundamental issues in prosody research. New approaches to these problems have recently emerged both in linguistic research on prosody and neighboring fields such as speech perception and recognition. The goal of this conference is to bring prosody researchers together with researchers in neighboring fields to discuss problems in prosodic variability. With invited speakers and participants from theoretical linguistics, psycholinguistics, sociolinguistics, phonetics, phonology, language acquisition and computer science, this meeting attempts to provide a holistic picture of prosodic variability and a venue for discussions that may lead to solving this core question in prosody research.

Spoken language not only communicates information about a speaker's thoughts or desires; it also conveys information about the speaker's identity. By simply listening to speakers' voices, accents, and word choice, we can learn a great deal about them, in addition to what is being talked about. Previous studies of language processing, however, have almost exclusively focused on the linguistic signal abstracted from individual speakers, investigating what listeners think is true about the world based on what an individual speaker has said. The project aims to explore the mechanism by which listeners extract information about the speaker through processing the linguistic signal. It then addresses the question of whether, and if so how, the increased knowledge about the speaker facilitates language comprehension. This research, consequently allows researchers to build a foundation for exploring how young children may learn speaker differences, which can contribute to new pedagogical tools for helping children to better interact with, and learn from, diverse populations. Secondly, the work will likely have industry applications for artificial intelligence technology, allowing it to better adapt its functionality to an individual user's talking style. 

This dissertation project employs two approaches to investigating what information listeners extract from spoken utterances. First, a large-scale online survey technique will be used to solicit responses from participants from a wider variety of linguistic and cultural backgrounds than those included in previous studies. Participants are exposed to utterances produced by two speakers and subsequently answer questions that probe their sensitivity to across-speaker differences. In the second set of experiments, a combination of an artificial language learning paradigm and an eye-tracking methodology will be used to study real-time language comprehension behaviors. Listeners' eye-gaze will be used to gain fine-grained information about the real-time development of their linguistic expectations. By combining these experimental approaches, the researchers elucidate how the human language comprehension system derives fine-grained expectations for future linguistic input and how the mechanism develops as a function of increased knowledge about linguistic communication.