Search Abstracts | Symposia | Slide Sessions | Poster Sessions
Neural encoding of prosodic pitch accents above and beyond correlated acoustic and linguistic features: insights from human intracranial EEG.
Poster Session A, Wednesday, September 30, 11:00 am - 1:00 pm, Wangari Maathai
Eleonora Beier1, Josef Parvizi1, Laura Gwilliams1; 1Stanford University
In spoken language, speakers help listeners anticipate when the most informative parts of a message will occur by modulating the prosody of their voice, including fluctuations in intonation, intensity and vowel duration. While prosody is a salient acoustic feature, the neural mechanisms underlying the role of prosody for facilitating comprehension are understudied. We address this gap by investigating how the brain encodes pitch accents, which are used by speakers to convey information load: rising accents typically mark new or emphasized content, while low accents mark redundant information. Prior work using intracranial EEG shows that distinct neural populations across the superior temporal gyrus (STG) and Heschl’s gyrus (HG) encode intonational patterns independently from the processing of speech acoustics. However, in naturalistic speech, prosody is highly correlated with many levels of linguistic representation, beyond acoustics. In particular, as intonation is used to mark information load, responses previously attributed to pitch accents may in fact reflect neural encoding of information load or other correlated linguistic features. We test whether the brain encodes abstract representations of pitch accents even when controlling for these confounded features. Neural activity was measured intracranially through depth electrodes for 19 patients with refractory epilepsy (9 female, median age: 36) as they listened to ~30 minutes of naturalistic speech (e.g., podcasts). Data was collected for 2606 recording sites across patients (an average of 137 sites per patient); a total of 747 recording sites were found to respond to the speech stimuli and were used in further analyses. Using banded ridge regression, we isolate variance in neural activity uniquely attributable to pitch accents, beyond that explained by pitch, intensity, the Mel spectrum, information load (word surprisal), as well as phonetic (e.g., place of articulation) and syntactic (e.g., part of speech) features. We find evidence of neural encoding of pitch accents through a distributed network of recording sites across the HG, STG and superior temporal sulcus (STS). These areas also encoded all other linguistic and acoustic features of interest, with some recording sites significantly responding to multiple features while other sites uniquely responding to pitch accents. Critically, this indicates that these regions represent abstract pitch accent categories above and beyond confounded acoustic and linguistic features. These results set the stage for ongoing analyses addressing how the brain tracks intonation over time to prepare for incoming information load, facilitating linguistic processing of the most informative parts of the speaker’s message. Overall, this work addresses the neural mechanisms supporting the processing of prosody, laying the groundwork for understanding how disruptions in these mechanisms contribute to communication disorders.
Topic Areas: Prosody, Computational Approaches