Singh, D. 1 , Samara, D. A. 1, 2 & Wonnacott, D. E. 1
1 University College London
2 University of Liverpool
Considerable evidence suggests that statistical learning mechanisms play a role in spoken language. Is the same true for literacy development? Samara and Caravolas (2014) demonstrated that constraints on letter contexts (e.g., t occurs with o, not e) can be learned incidentally from written text from 7 years of age. However, in that experiment, learning could be also underpinned by phonotactic sensitivity (i.e., if learners actively used phonology while reading). Here, we ask: can purely graphotactic constraints be learnt incidentally via statistical learning processes?
Over two brief sessions, we exposed 7-year-olds (n = 33) to CVC letter strings ending either with single or double codas (e.g., f, ff) that share the same pronunciation. Critically, singlets were always preceded by one of two word-medial vowels (e.g., duf, rut, *duff, *rutt) and doublets were always preceded by another (e.g., deff, rett, *def, *ret). At test, children made legality judgements (is dutt permissible?) for novel strings and completed a 'fill-in-the-blank' task (fill-in 'd_tt'). We found reliable generalization, although learning was stronger in a follow-up experiment (n = 25) whereby the patterns were explicitly taught. Ongoing work further explores the mechanisms underpinning successful graphotactic learning controlling for phonology by using nonpronouceable novel letters.