Temporal response modelling uncovers electrophysiological correlates of trial-by-trial error-driven learning

Abstract

Humans learn from statistical regularities in the environment. We tested if prediction and prediction error may play a role in such learning in the brain. We used Error-Driven Learning (EDL) to simulate participants’ trial-by-trial learning during exposure to a bimodal distribution of non-native lexical tones. We simulated incremental trial-by-trial learning to get estimates of the degree of expectation of upcoming stimuli over the course of the experiment. The expectation estimates were combined with Temporal Response Function fitting to generate a prediction of the trial-by-trial ERP waveform. EDL simulations captured the data significantly better than chance and better than models based on either stimulus characteristics or statistical distributions. The results provide tentative evidence that trial-by-trial learning as measured in neural activity is error-driven.

How trial-to-trial learning shapes mappings in the mental lexicon: Modelling Lexical Decision with Linear Discriminative Learning

“…While previous work has shown the predictive power of errordriven learning by means of the Rescorla-Wagner rule in many areas of language research (e.g. Ramscar et al, 2013;Nixon and Tomaschek, 2021;Ellis, 2006a,b) as well as trial-to-trial learning (Chuang and Baayen, 2021;Lentz et al, 2021;Tomaschek et al, 2022), the present work extends these findings by showing that error-driven learning implemented with the Widrow-Hoff learning rule can also capture trial-to-trial learning effects in a much richer and more realistic model of lexical processing incorporating state-of-the-art representations of semantics.…”

“…They have generally been designed for modelling specific purposes, and are often limited to a few items (e.g. Oppenheim et al, 2010;Lentz et al, 2021;Ramscar et al, 2013;Tomaschek et al, 2022). Models such as ACT-R (Anderson and Lebiere, 1998) model the learning and forgetting of stimuli also during experiments, but they treat words as units and model forgetting as a function of time (Van Rijn and Anderson, 2003), without taking into account interference caused by the learning of intervening stimuli, which is a crucial characteristic of the Rescorla-Wagner rule.…”

“…The challenge addressed in this study is whether LDL is powerful enough to approximate continuous word learning. In contrast to previous work (Oppenheim et al, 2010;Lentz et al, 2021), we do not use a carefully controlled experiment designed to measure the effect of learning explicitly: instead, our hypothesis is that learning takes place during classical psycholinguistic experiments such as lexical decision, and that if this is indeed the case, the consequences of trial-to-trial learning can be modeled with LDL.…”

The keys to the future? An examination of associative versus discriminative accounts of Serial Pattern Learning

“…In first language acquisition, Tomaschek (2020, 2021) present a computational model of early infants' learning of speech by using the incoming acoustic signal to predict upcoming acoustic signal. Apart from phonetic learning, error-driven learning has also been found to play a role in word learning (Ramscar et al, 2013a(Ramscar et al, , 2010(Ramscar et al, , 2011, morphological learning (Ramscar and Yarlett, 2007;Ramscar et al, 2013b;Tomaschek et al, 2019;Hoppe et al, 2020) and speech production (Tucker et al, 2019;Tomaschek and Ramscar, 2022) and speech perception (Shafaei-Bajestan and .There is also evidence for trial-by-trial error-driven learning in the brain (Lentz et al, 2022) and during lexical decision (? ).…”

Does error-driven learning occur in the absence of cues? Examination of the effects of updating connection weights to absent cues

“…The Rescorla-Wagner model was developed to capture the observation that rather than contiguity, learning was instead driven by surprise / prediction error and uncertainty (Kamin, 1969a;Rescorla, 1988), hence the term 'error-driven learning'. The Rescorla-Wagner model has proven remarkably successful in predicting human category learning (Gluck & Bower, 1988) and has recently been proposed as an account of human language acquisition (Ramscar & Yarlett, 2007;Ramscar, Yarlett, Dye, Denny, & Thorpe, 2010;Ramscar, Dye, & McCauley, 2013), predicting many linguistic phenomena (Baayen, Shaoul, Willits, & Ramscar, 2016;Baayen, Milin, Durdević, Hendrix, & Marelli, 2011;Ellis, 2006;Lentz, Nixon, & van Rij, 2022;Nixon, 2020;Nixon & Tomaschek, 2020, 2021. In a study investigating the learning mechanisms underlying second language speech sound acquisition, Nixon (2020) demonstrated that a number of key principles of error-driven learning also apply to human learning of speech, including Kamin's 'blocking effect' (Kamin, 1968(Kamin, , 1969b, cue competition, prediction and unlearning (Ramscar et al, 2010).…”