Online neural monitoring of statistical learning

Abstract

The extraction of patterns in the environment plays a critical role in many types of human learning, from motor skills to language acquisition. This process is known as statistical learning. Here we propose that statistical learning has two dissociable components: (1) perceptual binding of individual stimulus units into integrated composites and (2) storing those integrated representations for later use. Statistical learning is typically assessed using post-learning tasks, such that the two components are conflated. Our goal was to characterize the online perceptual component of statistical learning. Participants were exposed to a structured stream of repeating trisyllabic nonsense words and a random syllable stream. Online learning was indexed by an EEG-based measure that quantified neural entrainment at the frequency of the repeating words relative to that of individual syllables. Statistical learning was subsequently assessed using conventional measures in an explicit rating task and a reaction-time task. In the structured stream, neural entrainment to trisyllabic words was higher than in the random stream, increased as a function of exposure to track the progression of learning, and predicted performance on the reaction time (RT) task. These results demonstrate that monitoring this critical component of learning via rhythmic EEG entrainment reveals a gradual acquisition of knowledge whereby novel stimulus sequences are transformed into familiar composites. This online perceptual transformation is a critical component of learning.

Keywords: Implicit learning; Intertrial coherence; Neural entrainment; Steady-state response; Word segmentation.

Figure 1

EEG-based entrainment measure of learning. If perceptual grouping of individual syllables into trisyllabic words occurs during statistical learning, the steady-state response should show a decrease at the frequency of the individual syllables and an increase at the frequency of the trisyllabic words.

Figure 2

Summary of experimental design. The exposure task in the structured condition consisted of 12 min of continuous auditory exposure to four repeating nonsense words. The exposure task in the random condition consisted of exposure to pseudorandomly repeating syllables. The main test of explicit memory was the rating task, which required participants to provide words and foil items with a familiarity rating. This task was followed by two additional tests of explicit memory, the comparison task and the recognition task. Finally, the target-detection task was a reaction-time-based measure of statistical learning, in which participants detected target syllables embedded in a continuous auditory speech stream composed of the four nonsense words. The syllable assigned as the target was rotated across trials.

Figure 3

Behavioral results reflecting statistical learning. (A) Familiarity ratings provided on the rating task. (B) Corrected reaction times as a function of syllable position on the target detection task.

Figure 4

EEG results. (A) ITC as a function of condition (structured, random), block (1–3), and frequency. ITC values were used to compute the WLI, as described in Methods. (B) Topographical plots showing distribution of ITC across the scalp, as a function of condition and frequency (word, syllable). Note that different scales are used for word versus syllable frequencies. The six darker dots on the upper left scalp plot denote the approximate locations of the six centro-frontal electrodes used for WLI computations, where ITC was generally maximal at both the word and syllable frequencies. (C) The WLI as a function of condition and block.

Figure 5

Scatterplot showing the relation between the corrected reaction time effect and WLI in the structured stream (log-transformed scale).

Figure 6

ERP analysis of word/triplet onsets in the structured (red) and random (blue) conditions. Data are timelocked to the onset of each word in the structured condition, and the onset of each random triplet (i.e., every third syllable) in the random condition. Words in the structured condition differed from triplets in the random condition in the 300–500 ms range, showing a significantly larger N400 amplitude. The topographical voltage plot shows the distribution of this effect, computed by subtracting the mean amplitude in the random condition from the mean amplitude in the structured condition, from 300–500 ms.