Implicit learning of distributional patterns in linguistic and non-linguistic sequence production

implicit learning;phonotactics;associative learning;motor learning;sequence learning;speech production

Anderson, Nathaniel David

关键词: implicit learning;    phonotactics;    associative learning;    motor learning;    sequence learning;    speech production;   

Others  :  https://www.ideals.illinois.edu/bitstream/handle/2142/99230/ANDERSON-DISSERTATION-2017.pdf?sequence=1&isAllowed=y

美国|英语

来源: The Illinois Digital Environment for Access to Learning and Scholarship

【 摘 要 】

Speakers implicitly learn phonotactic patterns while speaking syllables. Evidence for this learning appears in speakers’ changing speech error distributions when tasked with rapidly reading sequences of syllables that are generated according to various phonotactic rules (Dell, Reed, Adams, & Meyer, 2000). Simple first-order rules like “/f/ appears only in syllable onset position” are learned quickly. Within a single one-hour session of this speeded syllable production task, erroneously-produced /f/s are much more likely to appear in syllable onset position than in syllable coda. Position-restricted phonemes show significantly higher rates of position maintenance when spoken in error than other phonemes that appear in both onset and coda positions during the experimental session, even when speakers are unaware of these patterns. More complex second-order rules in which a phoneme’s allowable position is dependent on the vowel are not learned in a single session (for example, /f/ might appear in syllable onset only when the vowel is /i/, and in syllable coda when the vowel is /a/). Such conditional rules do not affect speakers’ error patterns until a second session, and the effect appears to be dependent on sleep consolidation (Gaskell et al., 2014).Results from the speeded syllable production task with the addition of rule reversals point to rapid initial learning that interferes with the learning of subsequent rules, which contrasts with the pattern typically observed in discrimination learning tasks. Additionally, a button-pressing adaptation of the experimental paradigm shows that the learning rate differences observed for first- vs. second-order rules in speech appear even in a novel non-linguistic task, showing that the observed effect is not specific to the speech production system. A simple counting model provides a good qualitative fit to the observed patterns.