Statistical learning and second language acquisition: individual differences and neurobiological underpinning.

Project Information

INDIV-STAT

Grant agreement ID: 743528

DOI

10.3030/743528

Project closed

EC signature date 14 March 2017

Start date 1 July 2017

End date 30 June 2019

Funded under

• EXCELLENT SCIENCE - Marie Skłodowska-Curie Actions

Total cost

€ 170 509,20

EU contribution

€ 170 509,20

Coordinated by

THE HEBREW UNIVERSITY OF JERUSALEM
Israel

Periodic Reporting for period 1 - INDIV-STAT (Statistical learning and second language acquisition: individual differences and neurobiological underpinning.)

Reporting period: 2017-07-01 to 2019-06-30

Summary of the context and overall objectives of the project

The overall objective of INDIV-STAT is to provide a deep understanding of the learning mechanisms, and specifically the mechanisms that allow us to register the statistical regularities (for example common patterns) in our environment, that are relevant for language learning and literacy. This issue of language learning is important as an increasing number of countries are faced with multilingual societies.
The project can be summarized in the following broad objectives: 1) Develop better measures for human's abilities to learn regularities, 2) Obtain a brain signature of regularity learning, 3) Examine the link between individual capacities in regularity learning and literacy skills 4) Investigate if the sensitivity to regularities is a cognitive capacity that can be trained.

Work performed from the beginning of the project to the end of the period covered by the report and main results achieved so far

- Better measuring individual capacities in "Statistical learning" (hereafter SL): Standard test for measuring have under-examined psychometric problems and are not well suited for measuring individual's SL capacities. Together with Dr. Noam Siegelman I devised and evaluated novel improved tests in both the auditory and visual modality. We discovered an important problem with standard tasks that use verbal syllable material, which is that the prior knowledge participants have from their own language has a substantial impact on how they perform our tasks, leading to unreliable measures. We designed a task with nonverbal material that does not suffer from the same problem and has desirable psychometric characteristics.

- Online measures of regularity learning: We investigated the scope of the functionality of self-paced learning paradigms in capturing learning as it takes place. In these tasks participants advance a steam of stimuli by themselves and their reaction times provide a measure of learning. We found that self-paced paradigms are well suited also for neuroimaging work. Results of the self-paced task with embedded quadruplet patterns revealed important insights regarding the exact computations of SL. An outstanding controversy is whether learners extract global clusters or whether they are tuned to simple local transitions between elements. We provided evidence for inter-individual mixture, with different reliance on the two types of computations across individuals.

- Neuroimaging work: Using electro-encephalography we were able to demonstrate significantly increased power for frequencies in the beta range in the window leading up to presentations of unpredictable shapes at statistical pattern transitions relative to that same window prior to predictable shapes (i.e. shapes within a statistical pattern). Looking at the evolution of this spectral signature we found that it emerges with increased exposure. These findings hold the promise of converging on an online brain signature we aimed for and provide important insights regarding the role of prediction in SL.

- Subcomponents of SL: One working hypothesis we started the project with splits the variance in SL performance into 1) variance related to efficiency in encoding the representations of inputs, and 2) variance related to computational efficiency of detecting regularities. Processing and learning were seen as independent steps. Our data challenged this hypothesis and we entertained the novel hypothesis that one single computational principle, rate of information, can explain SL performance. We now have several datasets in support for this hypothesis, showing that rate of information does a very good job in accounting for SL outcomes.

- Work that examined the link between individual capacities in regularity learning and literacy skills: in a first line of work we tested the impact of time constraints in encoding and increasingly difficult regularities (affecting the learning component) on the performance on subjects with reading difficulties and control subjects with good literacy skills. We found overall lower performance for the group with reading difficulty but no specific impairment in one of the components. A second line of work, which I was invited to collaborate on with Dr. Elpida Pavlidou, explored SL capacities in young typical readers across three different modalities. The obtained data revealed a positive correlation between visual SL performance and phonological awareness but no significant correlations between SL and reading measures.

- Preliminary results of a training study reveal that the sensitivity to regularities might be trainable, we were however faced with methodological challenges that prevent clear conclusions and demand further investigation of this issue.

Results have been disseminated in 3 published manuscripts and the following presentations at the following places/conferences/workshops:
Hamburg University, September 2019, Hamburg, Germany
Interdisciplinary Advances in Statistical Learning, June 27-29 2019, San Sebastian
McDonnell Foundation Workshop: The Future of Statistical Learning, August 2018, Québec city
Annual conference of the Society for the Neurobiology of Language, August 2018, Québec city
Xiamen University, July 10th 2018, Xiamen
First international Workshop on Predictive Processing, June 2018, San Sebastian
Psychonomic Society annual meeting, Nov 2017, Vancouver
McDonnell Foundation Workshop: The Future of Statistical Learning, November 2017, New Haven
Conference of the European Society for Cognitive Psychology, Sept 2017,Potsdam

Progress beyond the state of the art and expected potential impact (including the socio-economic impact and the wider societal implications of the project so far)

The work on creating better measures of individual capacities in detecting statistical regularities helped to close a large methodological gap in the field. It allows researchers to now address questions regarding the ties between SL and different cognitive abilities. This work also raised awareness for methodological weaknesses and inspired other researchers in the field to evaluate the reliability of the tasks that are being used. The use of online measures allowed us to test the predictions of two contrasting computational accounts of SL. Our important contribution here was not to show one of them wrong but rather the raise (and provide evidence for) the possibility that their are individual differences in the computations that are used by different individuals.
Our neuroimaging work converged on a promising online brain signature which can be employed by future studies. It also provided preliminary yet important insights regarding the role of prediction in SL.
Experimental work on the components of SL led to an entirely different research line on the explanatory power of the construct rate of information. This in turn led to ideas regarding a novel theory of proficient reading across different writing systems.
Finally, work that examined the link between individual capacities in regularity learning and literacy skills and the trainability of SL capacities led to mixed results. In future work I aim to further tackle these issues.

The implications of the obtained results are mostly situated at the theoretical level as they did lead to a better understanding of the learning mechanisms that extract regularity. On the applied and societal level, the research performed under this action might lead to future recommendations about classroom interventions for teaching skills that require the learning of regularities, including language and literacy acquisition. At present it is however too early for such recommendations.

Cartoon representation of "Linguistic entrenchment": prior knowledge impacts learning in the lab.