Missions

Probing neural representations of speech, in parallel with experimental phonetic analysis of the same data (collected in the field); placing the results in the context of linguistic typology.

Activités

In close coordination with project partners, the recruited researcher will conduct experiments to determine the extent to which neural models, now at the heart of many approaches to Natural Language Processing, can be used to extract linguistic knowledge, particularly concerning tones. The main object of investigation will be the Yongning Na language (ISO code: NRU). This work, based on fine-tuning models using data from the target language, will be carried out in conjunction with the experimental phonetic study of the same data (collected in the field).
The project's computer science specialists will participate in supervising the work.
The successful candidate will be responsible for contributing to the various stages of the research:
- participating in the design of experimental protocols, considering different methods of analysis used in the community, in particular linguistic probes (classifiers trained to predict certain linguistic properties from representations discovered by neural networks)
- employing causal and counterfactual analysis methods to identify the information captured by neural networks, by modifying their architecture and/or representations.
- working closely with field linguists to further the community's work on phonetic analysis as well as the analysis of neural representations.

Compétences

PhD in Language Sciences or Phonetics
Basic knowledge of machine learning tools; familiarity with a scripting language
Ability to communicate and coordinate with different partners: field linguists, computer scientists, engineers
Rigorous, organised and methodical
Management and leadership skills

Contexte de travail

Work in accordance with the standard provisions of the laboratory's internal regulations.