Assessing the sensitivity of EEG-based frequency-tagging as a metric for statistical learning

Abstract

Statistical Learning (SL) is hypothesized to play an important role in language development. However, the behavioral measures typically used to assess SL, particularly at the level of individual participants, are largely indirect and often have low sensitivity. Recently, a neural metric based on frequency-tagging has been proposed as an alternative and more direct measure for studying SL. Here we tested the sensitivity of frequency-tagging measures for studying SL in individual participants in an artificial language paradigm, using non-invasive EEG recordings of neural activity in humans. Importantly, we use carefully constructed controls, in order to address potential acoustic confounds of the frequency-tagging approach. We compared the sensitivity of EEG-based metrics to both explicit and implicit behavioral tests of SL, and the correspondence between these presumed converging operations. Group-level results confirm that frequency-tagging can provide a robust indication of SL for an artificial language, above and beyond potential acoustic confounds. However, this metric had very low sensitivity at the level of individual participants, with significant effects found only in 30% of participants. Conversely, the implicit behavior measures indicated that SL has occurred in 70% of participants, which is more consistent with the proposed ubiquitous nature of SL. Moreover, there was low correspondence between the different measures used to assess SL. Taken together, while some researchers may find the frequency-tagging approach suitable for their needs, our results highlight the methodological challenges of assessing SL at the individual level, and the potential confounds that should be taken into account when interpreting frequency-tagged EEG data.

Neural tracking of speech does not unequivocally reflect intelligibility

“…These studies mostly report that neural tracking is stronger when listening to intelligible speech as compared to unintelligible signals in both theta (Ahissar et al, 2001; Doelling et al, 2014; Peelle et al, 2013) and delta ranges (Di Liberto, O’Sullivan, & Lalor, 2015b; Ding & Simon, 2013; Doelling et al, 2014), but see (Howard & Poeppel, 2010; Zoefel & VanRullen, 2015c). However, it is still unclear from these findings whether neural tracking changes do reflect linguistic processing alone, as speech’s intelligibility covaries with acoustical changes, or whether changes in acoustics alone can modulate neural tracking (Ding, Chatterjee, & Simon, 2013; Kösem & van Wassenhove, 2017; Meng et al, 2021; Pinto, Prior, & Zion Golumbic, 2022).…”

Statistical learning in patients in the minimally conscious state

“…For visual sequences, it is also shown that probabilistic regularities between items can be detected by both adults and infants [20][21][22]. Although SL is widely studied using behavioral and neural measures, a recent study has pointed out that most of these measures tend to be noisy and are only reliable at the group level [23]. Recent studies show that statistical learning engages widely distributed areas in cortex and also the hippocampus: The superior temporal gyrus primarily encodes transitional probabilities, inferior frontal gyrus and anterior temporal lobe primarily encodes ordinal position and identity, and the hippocampus primarily encodes identity [24].…”