Bias in Phonotactic Learning: Experimental Studies of Phonotactic Implicationals
An ongoing debate in phonology concerns the extent to which the phonological typology is shaped by synchronic learning biases. The two best-studied types of synchronic bias are complexity bias, a bias against formally complex patterns, and substantive bias, a bias against phonetically unnatural patterns. While most previous work has focused on bias in the learning of phonological alternations, this dissertation tests for substantive bias and complexity bias in phonotactic learning.
Four artificial grammar learning (AGL) experiments tested whether learners reproduce phonetically-motivated phonotactic implicationals from the typology. The implicationals concern the distribution of place of articulation and voicing contrasts in stops across positions. If a language has place or voicing contrasts in stops word-finally, it also has those contrasts word-initially, but if a language has such contrasts word-initially, it does not necessarily have them word-finally. These implicationals are phonetically motivated: stop place of articulation and voicing are less perceptible word-finally than word-initially. If place or voicing contrasts exist in a position where they are hard to perceive, they should also exist in positions where they are easier to perceive. My experiments exposed participants to place or voicing contrasts in word-initial or word-final position and then tested whether they extended the contrast(s) to the other word-edge position. Perception-based substantive bias predicts greater extension from word-final to word-initial position than vice versa. This prediction was not borne out in the place experiments but was borne out in one voicing experiment. The voicing experiments thus provide partial support for substantive bias. Due to the phonemic inventories of the artificial languages, the voicing experiments could also test for complexity bias. Effects of complexity bias emerged in both experiments.
A fifth AGL experiment tested the relative learnability of final voicing alternations. The experiment failed to find support for articulation-based substantive bias: final devoicing, which increases articulatory ease, was not learned better than final voicing. The results did provide additional support for complexity bias. Based on the results of this dissertation’s experiments and a review of the literature, I argue for distinguishing between perceptually-rooted and articulatorily-rooted substantive bias and claim that only perceptual naturalness biases phonological learning.