Noam Siegelman

Domain generality versus modality specificity: the paradox of statistical learning

Statistical learning (SL) is typically considered to be a domain-general mechanism by which cognitive systems discover the underlying distributional properties of the input. However, recent studies examining whether there are commonalities in the learning of distributional information across different domains or modalities consistently reveal modality and stimulus specificity. Therefore, important questions are how and why a hypothesized domain-general learning mechanism systematically produces such effects. Here, we offer a theoretical framework according to which SL is not a unitary mechanism, but a set of domain-general computational principles that operate in different modalities and, therefore, are subject to the specific constraints characteristic of their respective brain regions. This framework offers testable predictions and we discuss its computational and neurobiological plausibility.

What Predicts Successful Literacy Acquisition in a Second Language

In the study reported here, we examined whether success (or failure) in assimilating the structure of a second language can be predicted by general statistical-learning abilities that are nonlinguistic in nature. We employed a visual-statistical-learning (VSL) task, monitoring our participants’ implicit learning of the transitional probabilities of visual shapes. A pretest revealed that performance in the VSL task was not correlated with abilities related to a general g factor or working memory. We found that, on average, native speakers of English who more accurately picked up the implicit statistical structure embedded in the continuous stream of shapes better assimilated the Semitic structure of Hebrew words. Languages and their writing systems are characterized by idiosyncratic correlations of form and meaning, and our findings suggest that these correlations are picked up in the process of literacy acquisition, as they are picked up in any other type of learning, for the purpose of making sense of the environment.

Towards a theory of individual differences in statistical learning

In recent years, statistical learning (SL) research has seen a growing interest in tracking individual performance in SL tasks, mainly as a predictor of linguistic abilities. We review studies from this line of research and outline three presuppositions underlying the experimental approach they employ: (i) that SL is a unified theoretical construct; (ii) that current SL tasks are interchangeable, and equally valid for assessing SL ability; and (iii) that performance in the standard forced-choice test in the task is a good proxy of SL ability. We argue that these three critical presuppositions are subject to a number of theoretical and empirical issues. First, SL shows patterns of modality- and informational-specificity, suggesting that SL cannot be treated as a unified construct. Second, different SL tasks may tap into separate sub-components of SL that are not necessarily interchangeable. Third, the commonly used forced-choice tests in most SL tasks are subject to inherent limitations and confounds. As a first step, we offer a methodological approach that explicitly spells out a potential set of different SL dimensions, allowing for better transparency in choosing a specific SL task as a predictor of a given linguistic outcome. We then offer possible methodological solutions for better tracking and measuring SL ability. Taken together, these discussions provide a novel theoretical and methodological approach for assessing individual differences in SL, with clear testable predictions. This article is part of the themed issue ‘New frontiers for statistical learning in the cognitive sciences’.

Measuring individual differences in statistical learning: Current pitfalls and possible solutions

Most research in statistical learning (SL) has focused on the mean success rates of participants in detecting statistical contingencies at a group level. In recent years, however, researchers have shown increased interest in individual abilities in SL, either to predict other cognitive capacities or as a tool for understanding the mechanism underlying SL. Most if not all of this research enterprise has employed SL tasks that were originally designed for group-level studies. We argue that from an individual difference perspective, such tasks are psychometrically weak, and sometimes even flawed. In particular, the existing SL tasks have three major shortcomings: (1) the number of trials in the test phase is often too small (or, there is extensive repetition of the same targets throughout the test); (2) a large proportion of the sample performs at chance level, so that most of the data points reflect noise; and (3) the test items following familiarization are all of the same type and an identical level of difficulty. These factors lead to high measurement error, inevitably resulting in low reliability, and thereby doubtful validity. Here we present a novel method specifically designed for the measurement of individual differences in visual SL. The novel task we offer displays substantially superior psychometric properties. We report data regarding the reliability of the task and discuss the importance of the implementation of such tasks in future research.

Transposed-letter priming effect in Hebrew in the same–different task:

We investigated the interaction between morphological structure and transposed-letter priming using the same–different task with Hebrew, a Semitic language in which morphology has been shown to play a key role in visual word recognition. In contrast to the results observed with lexical decision (e.g., Velan & Frost, 2009, 2011), a transposed-letter priming effect was observed irrespective of the morphological structure of the words. We take these results to suggest that morphological decomposition occurs only in the service of lexical access. We discuss further a unique feature of written Arabic, another Semitic language, to explain the apparent conflict between our findings and those reported by Perea, Abu Mallouh, García-Orza, and Carreiras (2010).

Splitting the variance of statistical learning performance: A parametric investigation of exposure duration and transitional probabilities

What determines individuals’ efficacy in detecting regularities in visual statistical learning? Our theoretical starting point assumes that the variance in performance of statistical learning (SL) can be split into the variance related to efficiency in encoding representations within a modality and the variance related to the relative computational efficiency of detecting the distributional properties of the encoded representations. Using a novel methodology, we dissociated encoding from higher-order learning factors, by independently manipulating exposure duration and transitional probabilities in a stream of visual shapes. Our results show that the encoding of shapes and the retrieving of their transitional probabilities are not independent and additive processes, but interact to jointly determine SL performance. The theoretical implications of these findings for a mechanistic explanation of SL are discussed.

Is the Hebb repetition task a reliable measure of individual differences in sequence learning

The Hebb repetition task, an operationalization of long-term sequence learning through repetition, is the focus of renewed interest, as it is taken to provide a laboratory analogue for naturalistic vocabulary acquisition. Indeed, recent studies have consistently related performance in the Hebb repetition task with a range of linguistic (dis)abilities. However, despite the growing interest in the Hebb repetition effect as a theoretical construct, no previous research has ever tested whether the task used to assess Hebb learning offers a stable and reliable measure of individual performance in sequence learning. Since reliability is a necessary condition to predictive validity, in the present work, we tested whether individual ability in visual verbal Hebb repetition learning displays basic test–retest reliability. In a first experiment, Hebrew–English bilinguals performed two verbal Hebb tasks, one with English and one with Hebrew consonant letters. They were retested on the same Hebb tasks after a period of about 6 months. Overall, serial recall performance proved to be a stable and reliable capacity of an individual. By contrast, the test–retest reliability of individual learning performance in our Hebb task was close to zero. A second experiment with French speakers replicated these results and demonstrated that the concurrent learning of two repeated Hebb sequences within the same task minimally improves the reliability scores. Taken together, our results raise concerns regarding the usefulness of at least some current Hebb learning tasks in predicting linguistic (dis)abilities. The theoretical implications are discussed.