Two-C but not two-V: Segment similarity in learning an artificial lexicon

Abstract

The role of segment similarity (C1—C2—or —V1—V2) in a word learning task was assessed using an artificial lexicon in a referential context. Learning consisted of 480 trials in which S’s heard one of 40 CVCV nonsense strings, accompanied by an unfamiliar picture. In testing, participants heard the direction ‘‘Click on the [nonsense word],’’ and chose one of four pictures that matched the test item. On some trials, target lexical items (pibo) appeared with foils that contained matched consonants (pabu) or matched vowels (diko). There were higher rates of confusion errors to the matched-consonant items than to non-matched items, but no significant elevation in errors to matched-vowel items. A second experiment examined the role of differences in informativeness between C’s and V’s by inverting the numbers of C and V types (first experiment: 10 C, 5 V; second experiment: 5 C, 10 V). This made the consonants less predictive of word identity (more words contained the same consonants), and made the vowels more predictive of word identity. The matched-consonant effect remained undiminished while no corresponding matched-vowel effect emerged, ruling out a segment-informativeness explanation. Other accounts based on the syllable positions and confusability patterns of consonants are being explored.