General programme, activity sheet

Wednesday 7 September, 2016 11:30 to 12:00 Oral presentation
Evidence of emerging phonemic structure within a preliterate neural network model trained using real speech data
Speaker: Steph Ainsworth, Manchester Metropolitan University

Authors: Drs. Stephanie Ainsworth, Anne Hesketh, Anna Woollams & Stephen Welbourne

There is currently debate about the nature of children’s phonological representations and the potential roles of literacy and vocabulary growth within their development. Neural network models allow us to simulate phonological development while directly controlling oral language experience and exposure to orthography. While most previous models have used artificial inputs and outputs, which potentially bias the network towards phonemic representation, the current model is trained on real speech data allowing us to look for evidence of naturally emerging segmental representation. The feed forward model was successfully trained on the mappings between acoustic recordings of spoken words and the corresponding articulatory measurements (laryngograph, electromagnetic-articulograph and electro-palatograph). Analysis of the model’s hidden unit representations was conducted using a novel ‘hidden plot distance’ metric. The distance between word representations became increasingly sensitive to both phonemic and global similarity as the model developed. The results support the idea of multi-level phonological representations emerging through oral language experience. This is the first study to show the emergence of phonemes within a preliterate model which does not have any phonemic structure built into the model a priori. The implications for current theories of phonological development will be discussed.
Further information:

Final version (26 KB)

Phonetic and phonological development
Place: Room A-16

Other activities in Phonetic and phonological development
12:00 h. to 12:30 h.Oral presentation

Influence of frequency, neighborhood density and phonetic complexity on first words development in French

Speaker: Christophe Dos Santos, Université François-Rabelais, Tours
Speaker: Sophie Kern, Laboratoire DDL
13:00 h. to 13:30 h.Oral presentation

Adquisición de los ataques complejos en español: análisis desde la Teoría de la Optimidad

Speaker: Pilar Vivar, Universidad Católica de Temuco
12:30 h. to 13:00 h.Oral presentation

The effects of prosody and beat gestures on words recall within a contrastive discourse by children

Speaker: Judith Llanes Coromina, Universitat Pompeu Fabra
Speaker: Pilar Prieto, ICREA-Universitat Pompeu Fabra