Search Abstracts | Symposia | Slide Sessions | Poster Sessions
Frequency-based category learning in the brain and behavior
Poster Session C, Friday, October 25, 4:30 - 6:00 pm, Great Hall 3 and 4
Davide Crepaldi1, Mara De Rosa1, Elena Greatti1,2, Claudia Ruzza1, Maria Ktori1; 1International School for Advanced Studies (SISSA), Trieste, Italy, 2International School of Advanced Studies (ISAS), Camerino, Italy
The highly sophisticated structure of human language generates statistical regularities in the linguistic signal, which can potentially be captured to reverse-engineer the underlying system. One prominent example is the distinction between content and function words; since the latter are substantially more frequent than the former, sensitivity to the number of encounters with each word might potentially underlie the acquisition of this categorical distinction. De Rosa, Ktori, et al. (2022) devised a Fast Periodic Visual Stimulation (FPVS) paradigm that, coupled with EEG, tracks frequency-based category learning at the neural level. Skilled readers were shown sequences of stimuli at a fast rate (6Hz). An arbitrary subset of stimuli was shown less frequently, once every five higher-frequency items (6/5=1.2Hz). Hence, stimuli differed only in terms of their relative frequency within the stream. In one original (N=41) and one self-replication (N=43) experiment, we asked whether this neural grouping surfaces in behavior. Frequency-domain EEG analyses revealed robust neural responses to the oddball frequency, thus replicating De Rosa, Ktori et al. (2022). This effect also emerged with non-linguistic stimuli, showing that the mechanism is not language-specific. Despite the strong neural signal, participants’ memory for frequent and infrequent items was indistinguishable. These data raise important questions about the nature of frequency-based neural grouping, and particularly the conditions that render such grouping consistent and durable over time. In a second set of experiments, we checked whether the FPVS signal interacts with a concurrent task. In a first study (N=39), we engaged participants in a 3-back color matching task while they were concurrently presented with the FPVS stream in the background. This stream was either random; or statistically structured, so that it would induce neural grouping. If grouping attracts attention/cognitive resources, we would expect performance in the 3-back matching task to be worse in this latter condition. Conversely, in a second study (N=40), we varied the attentional demand of the main task – a simpler color change detection vs. a more demanding 3-back color matching – to examine its effects on the capture of statistical regularities in the FPVS stream. The findings indicate that neural grouping is largely independent of attentional load and/or cognitive demand. We discuss these data in the context of the debate regarding the connection between statistical learning and language processing. In the era of large language models, there are suggestions that language structure can be learned simply based on language-agnostic algorithms that are sensitive to probabilistic regularities in the linguistic signal. Here we provide a new look into this question, based on cognitive skills rather than on the performance of AI. We show that the brain does feature non-linguistic learning mechanisms that potentially underlie fundamental linguistic knowledge. There are, however, many open questions around the way that these mechanisms might unfold in the language learning process.
Topic Areas: Reading,