This thesis is an articulatory investigation into phonological variation and change in English /l/-darkening. Although syllable-based accounts of /l/-darkening state that light [l] occurs in onsets (e.g. `leap') and a dark variant in codas (e.g. `peel'), numerous works linking phonology with other subfields of linguistics have shown that this simplified distinction cannot fully account for the variation found. Firstly, /l/-darkening is sensitive to morphosyntactic structure, as shown through overapplication of the process in certain morphosyntactically defined positions: e.g. word-finally in phrases such as `heal it', or stem-finally before a suffix in words such as `healing'. In addition, analyses of /l/-darkening from several phonetic studies have led to some arguing against an allophonic distinction altogether, stating that the difference between light and dark variants is merely two extremes of one continuum. Not only does this interpretation challenge the traditional categorisation of /l/-darkening but, given the clear sensitivity to morphosyntactic boundaries that /l/-darkening displays, it also raises questions for a modular architecture of the grammar if phonetics can be morphologically conditioned. This dissertation is an empirical analysis of /l/-darkening, presenting data from nine varieties of English. Given the difficulty in measuring liquid consonants reliably, ultrasound tongue imaging is used to provide a thorough account of the prime articulatory correlations of darkening processes. The present study provides hitherto absent instrumental evidence confirming the varying degrees of morphosyntactic sensitivity across different dialects. I demonstrate that, rather than being contradictory or chaotic, variation to morphosyntactic boundaries cross-dialectally makes complete sense under an analysis that pays due consideration to the diachronic evolution of phonological processes. Moreover, my data show that the majority of speakers display both categorical allophony of light and dark variants, and gradient phonetic effects coexisting in the same grammar. Therefore, an adequate account of English /l/-darkening presupposes both a theory of the morphosyntax-phonology interface, and the phonetics-phonology interface. I interpret these results by assuming the modular architecture of the life cycle of phonological processes, whereby a phonological rule starts its life as a phonetically driven gradient process, over time stabilising into a phonological process at the phrase level, and advancing through the grammar. Not only does the life cycle make predictions about application at different levels of the grammar, it also predicts that stabilised phonological rules do not replace the phonetic processes from which they emerged, but typically coexist with them. Moreover, the obvious intimate link between /l/-darkening and /l/-vocalisation can be explained in terms of the life cycle, in the way of lenition trajectories. The results here show that, as predicted, the more recent stage of the lenition trajectory is harsher in terms of its phonetic effect, as well as less advanced in the grammar, applying at a lower level than darkening when the two co-occur in the same variety. I conclude by arguing that the proposed analysis demonstrates that a full understanding of /l/-darkening in English requires an approach that considers variation under phonetic, phonological and morphosyntactic terms. The wide range of dialectal diversity, for which this thesis provides only a small subset, shows a great deal of orderliness when paying due consideration to the diachronic evolution of variable phonological processes.

Ce travail de thèse étudie la production de la parole à l'âge de 4 ans, en comparaison avec l'adulte, sous l'angle du contrôle moteur. Un intérêt particulier a été porté à deux indices : la variabilité liée à la répétition de voyelles isolées et la coarticulation anticipatoire intra et extra-syllabique dans le cadre de la production de séquences de type V1-C-V2. Des données acoustiques et articulatoires ont été enregistrées via l'échographie linguale et le dispositif HOCUS. Les données acoustiques ont été analysées pour 20 enfants et 10 adultes et les données articulatoires ont été analysées pour 6 de ces enfants et 2 de ces adultes. En accord avec de nombreuses études de la littérature, la variabilité liée à la répétition est plus importante chez les enfants que chez les adultes. Une forte anticipation de V2 dans la réalisation de V1 a été observée pour tous les adultes. Chez les enfants, cette anticipation n'est pas systématique, et lorsqu'elle est observée, elle est d'amplitude plus faible que chez les adultes. Ainsi, seulement 5 enfants parmi les 20 étudiés montrent des signes d'anticipation, majoritairement dans la dimension antéropostérieure, manifestée dans l'espace acoustique par la dimension F2. La coarticulation anticipatoire intra-syllabique semble également être d'amplitude plus faible chez les enfants. Enfin, les gestes de parole des enfants sont plus lents que ceux des adultes. Dans leur ensemble, ces résultats sont interprétés comme la preuve de l'immaturité du système moteur pour la production de la parole chez l'enfant, selon deux perspectives : des patrons moteurs insuffisamment stables pour la production de gestes vocaliques et un manque d'efficacité dans l'anticipation les gestes futurs lors de la production d'une séquence de sons de parole. Selon une perspective empruntée aux théories du contrôle moteur optimal, nous proposons que la coarticulation anticipatoire repose sur l'usage de modèles internes, représentations sensori-motrices du système moteur de la parole dans le système nerveux central, et que son amplitude reflète la maturation progressive de ces représentations sensori-motrices à mesure que la production de la parole se développe. / This thesis work investigates speech production in 4-year-old children, in comparison with adults, from a speech motor control perspective. It focuses on two indices: measures of token-to-token variability in the production of isolated vowels and on anticipatory intra and extra-syllabic coarticulation within V1-C-V2 sequences. Acoustic and articulatory data were recorded thanks to ultrasound tongue imaging within the HOCUS system. Acoustic data from 20 children and 10 adults have been analyzed. Ultrasound data have been analyzed from a subset of these participants: 6 children and 2 adults. In agreement with former studies, token-to-token variability was greater in children than in adults. Strong anticipation of V2 in V1 realization was found in all adults. In children, anticipation was not systematic, and when observed, it was of smaller amplitude than in adults. In more details, only 5 children among the 20 studied showed a small amount of anticipation, mainly along the antero-posterior dimension, manifested in the acoustic F2 dimension. Anticipatory intra-syllabic coarticulation also seems to be of smaller amplitude in children than in adults. Last, children's speech gestures are slower than those of adults. These results are interpreted as evidence for the immaturity of children's speech motor control from two perspectives: insufficiently stable motor control patterns for vowel production, and a lack of effectiveness in anticipating forthcoming gestures. In line with theories of optimal motor control, we assume that anticipatory coarticulation is based on the use of internal models, i.e. sensori-motor representations of the speech production apparatus in the central nervous system, and that the amplitude of anticipatory coarticulation reflects the increasing maturation of these sensori-motor representations as speech develops.

The critical role of language pronunciation in communicative competence is significant, especially for second language learners. Despite renewed awareness of the importance of articulation, it remains a challenge for instructors to handle the pronunciation needs of language learners. There are relatively scarce pedagogical tools for pronunciation teaching and learning, such as inefficient, traditional pronunciation instructions like listening and repeating. Recently, electronic visual feedback (EVF) systems (e.g., medical ultrasound imaging) have been exploited in new approaches in such a way that they could be effectively incorporated in a range of teaching and learning contexts. Evaluation of ultrasound-enhanced methods for pronunciation training, such as multimodal methods, has asserted that visualizing articulator’s system as biofeedback to language learners might improve the efficiency of articulation learning. Despite the recent successful usage of multimodal techniques for pronunciation training, manual works and human manipulation are inevitable in many stages of those systems. Furthermore, recognizing tongue shape in noisy and low-contrast ultrasound images is a challenging job, especially for non-expert users in real-time applications. On the other hand, our user study revealed that users could not perceive the placement of their tongue inside the mouth comfortably just by watching pre-recorded videos. Machine learning is a subset of Artificial Intelligence (AI), where machines can learn by experiencing and acquiring skills without human involvement. Inspired by the functionality of the human brain, deep artificial neural networks learn from large amounts of data to perform a task repeatedly. Deep learning-based methods in many computer vision tasks have emerged as the dominant paradigm in recent years. Deep learning methods are powerful in automatic learning of a new job, while unlike traditional image processing methods, they are capable of dealing with many challenges such as object occlusion, transformation variant, and background artifacts. In this dissertation, we implemented a guided language pronunciation training system, benefits from the strengths of deep learning techniques. Our modular system attempts to provide a fully automatic and real-time language pronunciation training tool using ultrasound-enhanced augmented reality. Qualitatively and quantitatively assessments indicate an exceptional performance for our system in terms of flexibility, generalization, robustness, and autonomy outperformed previous techniques. Using our ultrasound-enhanced system, a language learner can observe her/his tongue movements during real-time speech, superimposed on her/his face automatically.