Youngah has been invited to give a talk at the University of Tokyo on 5 July 2023. The talk will discuss her collaborative work with Frank, “Learning phonology without phonology: insights from autoencoder modelling”, exploring the topic of how infants learn phonology without any prior knowledge of it.
The talk will present the results of their autoencoder modeling research. The study suggests that it might be possible for phonemes and distinctive features to be learned from unsegmented, non-transcribed wave data that resembles the language acquisition stages of infants.
The experiment conducted on Mandarin and English indicates that features could potentially be learned through repetitive projection and reconstruction, even without any prior knowledge of segmentation. The model appears to cluster segments of the same phoneme and separates different phonemes into distinct regions in the hidden space.
This research suggests that sound knowledge might be acquired to a certain extent through unsupervised learning, without the need for labeled data or previous phonological understanding. The findings offer insights into the early stages of human language acquisition and the ability of infants to recognize the sounds of their native language.