Reading and statistical learning

Statistical learning refers to the ability to (implicitly) extract regularities from the environment and to apply them to predict future events. To date, it is not clear whether a statistical learning process is involved in the process of learning to read, or which exact reading-related regularities benefit from the process and how they facilitate reading. Two related questions are: Is a domain-general statistical learning skill important for reading acquisition? And is a statistical learning deficit a possible cause of developmental dyslexia?


Schmalz, X. (preprint, 2023). Rules and statistics: What if it’s both? A basic computational

model of statistical learning in reading acquisition.

This is yet another attempt to make sense of why the literature on statistical learning and reading is as messy as it is, supplemented by a very basic computational model.


Schmalz, X., Moll, K., Mulatti, C., & Schulte-Körne, G. (2019). Is statistical learning related

to reading ability, and if so, why? Scientific Studies of Reading, 23(1). 64-76. doi: 10.1080/10888438.2018.142304. Preprint, data and analysis script here.

We tested a group of adults on two reading tests, two statistical learning tasks, their sensitivity to letter bigram frequency, and a control (choice reaction time) task. We found no correlations between most tasks, most notably between the reading and statistical learning tasks, but even between the two statistical learning tasks. This indicates that the tasks that are currently used in the literature are unsuitable to detect any relationship between reading and statistical learning ability.


Schmalz, X., Altoè, G., & Mulatti, C. (2017). Statistical learning and dyslexia: A systematic

review. Annals of Dyslexia, 6(2). 147-162. doi:10.1007/s11881-016-0136-0. Fulltext and analysis script here.

We review the literature where a group of participants with developmental dyslexia was compared to a control group on statistical learning tasks. We find evidence of publication bias, which prevents us from drawing conclusions about the absence or presence of a group difference.


Schmalz, X., & Mulatti, C. (2017). Busting a myth with the Bayes Factor: Effects of letter bigram

frequency in visual lexical decision do not reflect reading processes. The Mental Lexicon, 12(2). 263–282. Preprint available here, analysis scripts here.

In an analysis of the British Lexicon Project and the English Lexicon Project, we find that bigram frequency does not affect lexical processes in the lexical decision task. Rather, it seems to be affect guessing strategies (in lexical decision) and is probably confounded with phonological/articulatory processes (in reading aloud).


Schmalz, X., Mulatti, C., & Job, R. (preprint). Letter position coding: effects of legal and illegal

bigrams in masked transposed-letter priming. [Pre-registered study]. Registration report, materials, data, analysis script and preprint available here.

It has been proposed that letter bigram frequency may affect letter position coding at an early stage of the visual word recognition process (for a review, see Chetail, 2015, fulltext here). Here, we tested this prediction using a masked priming same-different decision task. We did not find an interaction between the type of prime and the legality of the interrupted letter bigram. However, we still need to replicate this with a lexical decision task, which is typically used for these types of experiments.