[PS-2.4] Individual strategies of morphological processing

Kuperman, V. 1 & Van Dyke, J. 2

1 McMaster University, Canada
2 Haskins Laboratories, USA

Most recent computational models postulate multiple parallel routes of morphological processing. Lexical properties that affect the balance between these routes are well studied. Yet word recognition is codetermined by linguistic characteristics of words as learned by particular individuals. We explore how verbal skills shift the balance between processing routes, as diagnosed by interactions of scores in skill tests and distributional properties of suffixed words (trucker) and their bases (truck). Seventy-one non-college-bound adolescent speakers of US English took 17 skill tests gauging major components of reading ability. They read 240 sentences silently for comprehension, while their eye-movements were recorded. We fitted linear mixed-models to eye-movement measures registered for 69 transparent derived words in those sentences: critical predictors were the test scores and frequencies of derived words and morphemes. Scores in segmentation tests (measuring the skill of segmenting words and nonwords into phonemes) interacted with whole-word frequency such that the frequency effect was stronger for poor scorers than better ones. Crucially, the interaction of skill by base word frequency was such that a higher-frequency base came with shorter reading times for poor scorers, a weak negative correlation for mid-range scorers, and a noticeable inflation of reading times for high-scorers. Thus, good readers suffered from lexical competition from morphologically related words, while poor readers received a recognition boost from base words. The additive processing advantage from higher-frequency derived and base words in poor readers cannot be accommodated by current computational models of morphological processing. Evidence supports both morphemic and whole-word routes, yet their engagement varies qualitatively as a function of individual verbal skill. Readers strategically adjust weights of different sources of morphological information, depending on the quality of lexical representation of both derived words and morphemes (measured here via lexical frequency), and the ability of individuals to segment morphemes out of complex words.