Presentation
Search Abstracts | Symposia | Slide Sessions | Poster Sessions | Lightning Talks
Behavioral and neural mechanisms for accommodating different forms of non-canonical speech
There is a Poster PDF for this presentation, but you must be a current member or registered to attend SNL 2023 to view it. Please go to your Account Home page to register.
Poster E86 in Poster Session E, Thursday, October 26, 10:15 am - 12:00 pm CEST, Espace Vieux-Port
This poster is part of the Sandbox Series.
Samuel Weiss-Cowie1, Lucy MacGregor1, Máté Aller1, Matt Davis1; 1University of Cambridge
Whether due to novel words or unfamiliar accents, speech can diverge from listeners’ existing knowledge in many ways. A range of adaptive mechanisms enable accommodation of non-canonical speech. For instance, new lexical information is made meaningful via word learning, and non-canonical phonetic input—like accents—becomes comprehensible through perceptual learning (Norris et al., 2003). Neural representations of non-canonical speech also change as listeners adapt; pseudowords become more word-like with time (Bakker-Marshall et al., 2018), and responses to acoustically ambiguous segments resemble unambiguous phonemes as phonetic boundaries shift (Luthra et al., 2020). Studies of non-canonical speech often focus on just one adaptation mechanism. We propose to investigate behavioral and neural responses to multiple forms of non-canonical speech presented concurrently. Re-analysis of existing MEG data will provide insight into how neural responses to semantic violations differ depending on whether speech contradicts immediate context and/or longer-term lexico-semantic knowledge. Listeners (n=18) completed various tasks while hearing either isolated or word-primed items from word-word-pseudoword triplets ending in /t/, /p/, and /k/ (e.g., lake, late, lape). Comparing unexpected words (swan-late) with pseudowords (swan-lape), relative to expected words (swan-lake), will tell us about the impact of longer-term lexico-semantic knowledge violations when listeners have also generated semantic predictions based on immediate context. We will also compare responses to isolated (lape) and word-primed pseudowords (swan-lape), relative to differences between isolated (late) and word-primed words (swan-late). This will inform our understanding of how existing lexicalized knowledge affects the impact of semantic predictions on neural activity. Multivariate decoding will assess how syllable final segment identification is modulated by short-term semantic context versus longer-term lexico-semantic knowledge. We also present a planned MEG study exploring the consequences of semantic and phonetic prediction violations on two adaptive processes—namely perceptual learning and word learning. Participants will read pairs of sentence stems predicting the same final word (“The students couldn’t keep still during…”), followed by spoken presentation of either the predicted word or a pseudoword (“class”/“paboose”). Critically, half the speech stimuli are pronounced canonically (“class”/“paboose”), while the other half are pronounced in an artificial accent encouraging perceptual learning (“claff”/“paboof”). Listeners may thus engage in word learning, perceptual learning, or both on a given trial depending on whether a violation is interpreted as semantic (pseudoword) and/or phonetic (accented). Pilot behavioral data (n = 34) suggests that listeners uniquely correct for trained accents but not unfamiliar ones in word report, while also performing above chance on pseudoword memory. Listeners can thus flexibly engage in both perceptual and word learning within a single set of trials to extract the most probable explanation for why speech differed from expectations. We will compare neural responses to early and late presentations of pseudowords to assess word learning effects, and also contrast presentations in trained and untrained accents to evaluate whether encoding is episodic. Multivariate decoding of final segment will gauge the relative impacts of accent training and lexicality on neural representations for non-canonical phonemes. The above studies will thus untangle how listeners simultaneously navigate multiple types of speech prediction violations.
Topic Areas: Speech Perception,