On the nature of working memory structures in phonological encoding

Abstract

Current model and theories of language and speech production commonly propose that speakers plan and execute sequences of phonemic segments by integrating and consolidating individual segments into cohesive memory structures or “chunks,” which reduces processing load and improves motor performance. Yet, there is no consensus on the nature of the chunks buffered in memory during serial speech planning. To identify these structures, we investigated the generalization of motor chunking from training to transfer utterances. During training, subjects repeated isolated syllables containing non-native consonant clusters. Subjects produced these syllables with increased accuracy and speed after training, indicative of motor learning. After learning, we tested for generalization under higher memory load by having subjects repeat pairs of syllables that overlapped to varying degrees with the practiced syllables. We observed complete transfer of performance speed improvements to novel syllables containing previously practiced clusters, but only if they were practiced in the same syllable position (onset or coda). Practicing the whole syllable, however, resulted in greater accuracy improvements compared to practicing just the clusters, regardless of syllable position. Collectively, these findings suggest that working memory utilizes a syllabic structural frame with different representations for the same phonemes in different frame slots.