[PS-1.25] Statistical information encoded in English spelling

Ulicheva, A. 1 , Aronoff, M. 2 & Rastle, K. 1

1 Royal Holloway University of London
2 Stony Brook University

Substantial evidence indicates that inconsistency harms learning. In the context of learning to read, English-speaking children lag behind their peers learning languages with more consistent spelling-sound correspondences (Seymour, Aro and Erskine, 2003). In the context of language acquisition, the learning of morphology is hindered by the presence of form-meaning inconsistency (Tamminen, Davis and Rastle, 2015). If inconsistency is harmful, then why does it emerge? To answer this question, we examine English spelling - a dynamic system that has self-organised over centuries (Berg and Aronoff, 2017). Inconsistent spelling-sound correspondences exist in abundance in English, and while these could have become extinct, this evolution has spared them. We present results of computational linguistic analyses showing that suffix spellings, albeit unpredictable from sound, carry unique statistical information that is not available in phonology. Further, we show that English speakers pick up these multi-level cues in the absence of formal instruction, and exploit them explicitly and implicitly when dealing with written language. Thus, fluent readers can keep track of and weigh statistical information at multiple levels. We argue that consistency in form-meaning knowledge outweighs inconsistency in spelling-sound knowledge and aids written language processing.