Phonological biases in the learning of tonal and segmental alternations

Abstract

Phonological biases suppose that language learners are predisposed toward certain sound patterns. The two widely studied phonological biases are structural bias, a bias against structurally complex patterns, and substantive bias, a bias against phonetically unnatural patterns. Experimental evidence for structural bias has been robust, but the main focus has been primarily on segmental phonology. Only a handful of studies have investigated the role of structural bias in suprasegmental phonology. In comparison to structural bias, experimental results for substantive bias have been mixed. The factors leading to the mixed support for substantive bias have yet to be resolved. Against this background, this dissertation (a) investigates the role of structural bias in suprasegmental phonology and (b) explores the possible factors influencing the effect of substantive bias. For (a), my focus is on learning tonal alternations. For (b), the focus in this dissertation is the influence of input variability on substantive bias. To examine the effect of structural bias in the learning of tonal alternations, a survey on tone sandhi directionality and their phonological environments across 17 Chinese varieties was first conducted, followed by an artificial language learning study on tonal alternations. The survey found that tone sandhi directionality is largely uni-directional either throughout a tone sandhi system or within each grammatical category, which supports the structural simplicity hypothesis. Crucially, within uni-directional patterns, the directionality is largely rightward (right-dominant local tone substitution and left-dominant rightward tone extension), which have phonetic groundings. Therefore, a phonetically-grounded structural bias better captures the directionality asymmetry. To further investigate whether the directionality asymmetry is reflected in learning, artificial tone alternation patterns were learned by native speakers of Mandarin and Hong Kong Cantonese. Both Mandarin and Cantonese speakers showed a learning bias toward uni-directional patterns, supporting the structural bias hypothesis. Crucially, among uni-directional patterns, only right-dominant patterns were learned better, not left-dominant ones, suggesting the effect of phonetically-grounded structural bias in learning tonal alternations, and the connection between typological asymmetry and learning biases. As to the second question, i.e., why the effect of substantive bias has been mixed, this dissertation focuses on the possible influence of input variability on substantive bias. An artificial language learning study on /n/~/l/ alternation by native speakers of Hong Kong Cantonese was conducted. There were two categorical learning conditions and two variable learning conditions. In the categorical learning conditions, all the initial /n/ changed to /l/ either in phonetically motivated or unmotivated contexts. In the variable learning conditions, the dominant percentage of initial /n/ changed to /l/ in phonetically motivated or unmotivated context. Substantive bias arose only in variable learning conditions, which suggests that input variability can facilitate the effect of substantive bias. Based on the results of this dissertation’s survey and experiments, I argue for structural simplicity incorporating phonetic substance to capture the typology and the learning of tone alternation directionality. I also discuss for the influence of input variability and methodological choices on substantive bias. This is important for future research into phonological biases in suprasegmental phonology and structural bias effects.