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

 
Poster C11, Thursday, November 9, 10:00 – 11:15 am, Harborview and Loch Raven Ballrooms

The time-course of statistical learning in patients with left hemisphere stroke

Kathryn D. Schuler1,2, Mackenzie E. Fama1,2, Peter E. Turkeltaub1,2, Elissa L. Newport1,2;1Georgetown University, 2Center for Brain Plasticity and Recovery

A fundamental aspect of learning involves extracting patterns from the environment via statistical learning (Saffran, Aslin & Newport, 1996). Research investigating the neural correlates of this mechanism for language suggests that the left inferior frontal gyrus, left arcuate fasciculus, and bilateral caudate/putamen may underlie this process (Karuza et al, 2013). Recent work suggests that damage to some of these left hemisphere (LH) language areas from stroke results in diminished ability to perform on statistical language learning tasks (Fama et al, 2015). Importantly, this prior study demonstrated impairment in the outcome of learning, via behavioral measures after exposure. Here, we ask whether the learning process (i.e. the computation of these statistics) is itself impaired. In addition, we ask whether damage to LH language areas will also impair non-linguistic statistical learning. To address these questions we developed a serial reaction time (SRT) task that was an exact analog of the original word-segmentation experiment, allowing us to assess statistical learning of nonlinguistic sequence learning via on-line RT measures as well as in the traditional post-exposure test. In the original word-segmentation experiment, learners acquired the groupings of syllables into words from a stream of speech by computing transitional probabilities (TPs) between syllables (Saffran et al., 1996). Within words, TPs between syllables were high, whereas TPs across word boundaries were low. In our SRT paradigm, sequences of syllables were converted into sequences of locations on a touch screen. Participants were asked to touch a mole that moved from one location to the next as quickly and accurately as possible. Unbeknown to the participants, the mole in this task moves in a pattern in which, across the exposure period, the probability of movement for some transitions is high (1.0) and for others is low (0.33). Learning these patterns is measured by comparing reaction times (RTs) to high probability transitions versus to low probability transitions over the course of the exposure. After exposure, learners performed the traditional post-test assessment of learning, rating sequences that occurred during exposure (high and low probability) and sequences that did not occur during exposure. Participants were fifteen patients with LH stroke (mean age=58.677.93) and 13 age-matched controls (mean age=65.237.92). Consistent with previous findings, post-test results demonstrated impaired learning in patients: controls distinguished exposure sequences from those not presented during exposure; patients did not make this distinction. However, during learning, all participants extracted the underlying regularities, showing faster RTs for high probability than low probability transitions (estimate =-58.72ms, se=6.43, p<0.001). Patients exhibited slower RTs overall (estimate=59.72ms, se=24.51, p<0.05), but there was no effect of participant group on learning overall (estimate=4.39ms, se=8.66, p=0.63) or on any part of the learning curve (p>0.05). Our findings suggest that an implicit RT task provides a sensitive measure of residual statistical learning abilities in patient populations. Patients with LH stroke retain the ability to compute language-like statistics during learning, at least for non-linguistic stimuli. Perhaps while recognition of previously learned statistical regularities is impaired in patients after LH stroke, the computation of underlying statistics during learning may remain.

Topic Area: Grammar: Syntax

Back to Poster Schedule