LDL | Department of Linguistics | HKU

Youngah presented at the 2023 Linguistics Colloquium organized by the Seoul National University.

Infants require two crucial skills to successfully begin language acquisition: (a) the ability to learn fundamental speech sound units, or phonemes, and (b) the capacity to decompose sound sequences into meaningful units. This talk will discuss the effectiveness of an autoencoder model in learning phonemes and phoneme boundaries from unsegmented, non-transcribed wave data, similar to the early stages of infant language acquisition. The experiment was conducted in Mandarin and English, and the results demonstrate that phonemes and their associated features can be learned through repeated projection and reconstruction without prior knowledge of segmentation. The model clusters segments of the same phoneme and projects different phonemes to separate regions in the hidden space. Furthermore, the model successfully decomposes words into phonemes in sequential order, which is a crucial foundation for phonotactic knowledge. However, the model struggles to cluster allophones closely, indicating the boundary between bottom-up and top-down information in phonological learning. This study suggests that fundamental sound knowledge in the early stages of language acquisition can be learned to some extent through unsupervised learning without labeled data or prior knowledge of segmentation, providing valuable insights into early human language acquisition.

Do, Y. (2023). Unsupervised learning of phonemes in early language acquisition: Insights from an autoencoder model. 2023 Linguistics Colloquium, Seoul National University.

The last of the summer research meetings has just been concluded. Our team has made significant progress, and we are excited about the potential outcomes of our research.

We want to take this opportunity to express our appreciation to the research interns who have contributed immensely to the success of the projects throughout the summer. Their efforts have been invaluable, and we are proud to have them as part of our team.

We also want to acknowledge the contributions of Bingzi, who will soon be commencing her research journey at MIT. We wish her all the best in her future endeavours.

Thanks everyone for your support in our research.

Thomas recently presented the work with Xiaoyu, Youngah and Tungle from National Taiwan University at the 16th International Cognitive Linguistics Conference in Heinrich Heine University Düsseldorf. The study converged ratings and corpus measures for ABB words in Chinese through PCA.

The University Grants Committee has approved the use of General Research Fund this year for the following LDL projects.

Breaking down and rebuilding iconicity

Youngah will lead postdoc researchers and students in this project, which aims to understand the phonological and cognitive mechanisms behind spoken iconicity and its accessibility to a variety of language speakers, by training a neural network with ideophone corpus data to generate sound-meaning associations and test them in lab-based experiments with human participants.

Sociophonetic variation in Hong Kong and Heritage Cantonese

Jon will lead postdoc researchers and students in this project, which seeks to understand how phonological representation and sound change occur in bilingual communities. In particular, they will look at the sociophonetic variation of Cantonese sibilants among speakers of different linguistic backgrounds.

Our lab members will present their recent work at the upcoming Hanyang International Symposium on Phonetics and Cognitive Sciences of Language 2023 (HISPhonCog2023) in Seoul!

Jon, Ming, Ivy and Jonah will present “downloadAudiovisual enhancement in clear speech production of English laterals.” It looks at the production of laterals and other coronal consonants in normal and listener-oriented clear speech. The study involved 18 adult native English speakers who participated in a two-part speech production experiment. The findings suggest that listener-oriented speech does not necessarily prioritize auditory perceptibility. Instead, speakers may choose to provide listeners with direct visual cues of a segment’s articulatory/gestural properties, consistent with the inherently multimodal nature of speech perception.

Frank and Youngah will present “downloadBottom-up learning of a phonetic system using an autoencoder“. The study explores the ability of a machine learning model to learn a phonetic system from contextless acoustic input. The autoencoder model was trained on recordings from English and Mandarin speakers and proved successful in capturing the phonetic system of both languages. The study suggests that infants’ phonetic knowledge may not be innate but can be acquired based purely on acoustic information, without relying on language-specific learning facilities. However, the model could not reach phonological knowledge without training on phonological cues, highlighting the indispensable role of phonological information in refining learners’ knowledge on phonetics and phonology.

Changhe and Jon will present “downloadNasality and Nasal Excrescence of the Nasal Vowels in Shanghai Chinese“. It examines the acoustic and articulatory configuration of the Shanghainese nasal vowels. The study analyzed data from six native speakers of Shanghainese and found that the nasal vowels show little nasalization in the first fifth of their duration and steadily increasing nasality thereafter. The findings suggest that the Shanghainese nasal vowels are only partially nasalized, with no clear relationship between the proportion of vowel nasality and duration or presence of the nasal consonant.