You are viewing the SNL 2017 Archive Website. For the latest information, see the Current Website.

Poster C19, Thursday, November 9, 10:00 – 11:15 am, Harborview and Loch Raven Ballrooms

A window for word-learning: Measuring dynamic neural responses during statistical language learning

Nicolette Noonan1, Lisa Archibald1, Marc Joanisse1;1The University of Western Ontario

The role of statistical learning in segmenting words from fluent speech is well described, particularly for the role of transitional probabilities between syllables in speech segmentation. Previous examination of event-related potentials (ERP) related to segmentation has reported the emergence of N100 and N400 components following artificial language exposure (Cunillera et al., 2006; Sanders et al., 2002), indexing word segmentation, and a P200 (de Diego Balageur, 2007; Cunillera et al., 2006) to index word identification. However, little work has focused on neural indices of word segmentation during language exposure. To examine the process of word segmentation on-line, we measured participants event-related potentials (ERPs) during exposure to a novel artificial language using electroencephalography (EEG). Cortical EEGs were recorded during exposure to an artificial language, and during a post-exposure test phase. Twenty-four adults were exposed to the artificial language for 21-minutes. The language was a structured, unsegmented speech stream containing six tri-syllabic nonsense words (e.g.: babupu, pidadi; Saffran et al., 1997). ERP epochs were time-locked to the onset of each syllable throughout the exposure phase. Responses to the statistically-constrained word-final syllables were compared across the exposure phase. Behavioural and ERP responses to trained versus foil words were also assessed in a test phase that immediately followed the exposure phase. In response to the word-final syllables, we found a significant linear increase in the amplitude of a P200 component over the first five minutes of artificial language exposure. At the 7th minute of exposure, the amplitude of the P200 substantially decreased and did not differ in amplitude from the 1st minute. This return to baseline amplitude was consistent when measured across the 14th and 21st minutes of exposure. Following language exposure, ERPs in response to trained and foil words at test did not differ. The P200 component that emerged during the first five minutes of artificial language exposure may index processes related to the extraction of the within-word transitional probabilities (e.g.: de Diego Balaguer et al., 2007). Additionally, the dynamic nature of this component could reflect a readiness for word learning that dissipates over redundant language exposure. We hypothesize that the extracted acoustic forms are stored as potential lexical items by the 5th minute of exposure, but that given the language learning paradigm did not involve additional segmentation or semantic cues, participants are not retaining the segmented acoustic form. Thus, although transitional probability cues may help language learners to initially extract potential words from fluent speech and for temporary maintenance in memory, the absence of additional meaningful information to bootstrap word learning leads to rapid decay of this segmented token. Taken together, these findings provide key insights into both the dynamical nature of statistical language learning, and the limitations of word segmentation based solely on the statistical relationships among syllables.

Topic Area: Language Development

Back to Poster Schedule