Chunking of phonological units in speech sequencing

Abstract

Efficient speech communication requires rapid, fluent production of phoneme sequences. To achieve this, our brains store frequently occurring subsequences as cohesive "chunks" that reduce phonological working memory load and improve motor performance. The current study used a motor-sequence learning paradigm in which the generalization of two performance gains (utterance duration and errors) from practicing novel phoneme sequences was used to infer the nature of these speech chunks. We found that performance improvements in duration from practicing syllables with non-native consonant clusters largely generalized to new syllables that contained those clusters. Practicing the whole syllable, however, resulted in larger performance gains in error rates compared to practicing just the consonant clusters. Collectively, these findings are consistent with theories of speech production that posit the consonant cluster as a fundamental unit of phonological working memory and speech sequencing as well as those positing the syllable as a fundamental unit of motor programming.

Dissociation between phonological working memory structures and motor programming units during speech motor-sequence learning

“…The current research provides a further test of these hypotheses by tackling three issues that were not directly addressed by our prior study (Segawa et al, 2019). First, and perhaps most importantly, we did not provide a definitive test between theoretical accounts that posit that the consonant cluster is a unit of storage in PWM versus a unit of articulation because these two levels of representation were conflated in speakers' productions.…”

“…We used a combination of Praat and custom MATLAB software to perceptually rate and acoustically measure onsets and offsets of syllables by viewing the waveform and spectrogram and listening to the audio files. Each utterance was coded by a trained phonetician, blinded to experimental condition, for seven possible error subtypes: 1 As in Segawa et al (2019), we considered vocoid epenthesis errors to be errors in fluency (in the sense of sounding somewhat different than a native speaker producing the clusters, as if with a strong foreign accent) rather than errors in phoneme sequencing. Previous work examining the production of non-native consonant sequences demonstrates that epenthetic vowels arise from inter-gestural mistiming (Davidson, 2005(Davidson, , 2006Buchwald, Calhoun, Rimikis, Lowe, Wellner, & Edwards, 2019); a transitional vowel is produced when the achievement of a consonantal constriction does not temporally overlap with the release of the previous tautosyllabic consonantal constriction.…”

“…Our recent work on this issue used a motor-sequence learning paradigm in which the generalization of performance gains from practicing novel phoneme sequences was used to infer the nature of speech chunks (Segawa, Masapollo, Tong, Smith, & Guenther, 2019). In these experiments, native English-speaking participants were trained over several days to produce a set of CCVCC non-words involving non-native (i.e., phonotactically-illegal) consonants clusters (e.g., shkizg, bdepf, dgamj), where C stands for consonant and V stands for vowel.…”

“…Second, regardless of whether one assumes that the consonant cluster is a PWM unit or an articulatory motor programming unit, the details of such a representation are only partially evident from the results of Segawa et al (2019) and other studies. Specifically, since a particular cluster in Segawa et al (2019) always appeared in the same syllable position within a word, it is unclear whether this unit, once learned, generalizes to other syllable positions; for example, does learning of a cluster in the syllable onset position generalize to the same cluster in the coda position?…”

“…Last, but not least, Segawa et al (2019) focused on motor performance gains without actually measuring the verbal working memory span of participants. Hence, the results only offer speculative support for the role of working memory in task performance.…”