Thread by @dnl_singh

New paper from my PhD in the Language Learning Lab just accepted in JML. Singh, Wonnacott & Samara (2021); with @Anna_D_Samara and @lizwonna . Pre-print: psyarxiv.com/8px7n/. 1/13

Bkgd: How do children pick up spelling rules from text? We previously showed children use statistical learning processes to generalize over novel spelling patterns (doi.org/10.1016/j.cogn…). 2/13

Hwvr: All patterns had a phonological counterpart (e.g.*tx is illegal in both written/spoken English). Are these processes also used to form spelling generalizations when there are no correlated phonotactics (e.g.*gz maps on a frequent word-final sound combination /ˈbæɡz/)? 3/13

We could test this with an artificial orthography, but these experiments are challenging for children. Instead, we used pronounceable nonword materials ending in single vs. double letters (e.g. def, deff). 4/13

These stimuli—presented under incidental conditions—exemplified two novel patterns on single/double letter occurrence (which sound the same—thus removing phonotactic counterpart). 5/13

In Exp1 (simple pattern), one vowel predicted singlets and another predicted doublets (def, des; duff, duss). In Exp2 (complex pattern), each vowel predicted some singlets/doublets (deff, des; duf, duss). Exposure was followed by generalization tests (production, judgments). 6/13

6–7-year-olds and adults generalized over both patterns with above chance accuracy. There was no evidence of the simpler pattern being learned better than the more complex pattern; hwvr, Bayes Factor (BF) analyses (Dienes, 2016) showed that this finding was inconclusive. 7/13

Experiments 1–2 suggest that statistical learning processes are implicated in learning the patterns; hwvr, many aspects of spelling are taught explicitly in school: Is this the optimal method for learning spelling patterns with no phonological counterpart? (Exp3) 8/13

In Exp3, we investigated whether explicitly teaching children the patterns of Exp1 (followed by comparable exposure to the nonwords) gave a learning advantage. We found evidence of stronger generalization in both tests, in support of this hypothesis. 9/13

Final exploratory Q: Is there a link between learning ability and literacy (WRAT-IV) performance? Using BFs, we found tentative evidence of no relationship between literacy and incidental learning performance but of a positive association with explicit learning. 10/13

Does this mean that incidental statistical learning ability is unrelated to literacy? We cannot rule out this possibility; hwvr, null results may also reflect poor psychometric properties of our statistical learning tests (Siegelman et al., 2017; West et al., 2018). 11/13

TAKE HOME: Children can learn spelling patterns incidentally even when they lack a phonological counterpart. Explicit instructions are beneficial, in line with the practice of teaching spelling patterns. 12/13

Preprint here: psyarxiv.com/8px7n/; preregistrations, open access data and full analysis script: osf.io/qzwu5/. 13/13