​​​​​Starting date: between September and November 2024

Place of work: Laboratoire d’Informatique de Grenoble, CNRS, Grenoble, France

Duration of contract: 3 years

Contact: maximin.coavoux@univ-grenoble-alpes.fr

Mission: research oriented (teaching possible but not mandatory)

Description:

The SynPaX ANR project (Syntactic Parsing of Spontaneous speech with cross-modal transfer learning) aims at investigating modal fusion systems to develop end-to-end speech parsing models. Most research on syntactic parsing of speech focused on parsing transcriptions, and face two limitations: (i) the predicted transcriptions may be very noisy, (i) they do not contain prosodic cues that are important to infer the syntactic structure.

A first end-to-end speech parsing model has been proposed by Pupier et al (2022) and evaluated on French spoken treebank Orféo. The main issues of the model stems from: (i) the scarcity of training data (ii) the difficulty of parsing spontaneous speech. Hence, the PhD project will build on this work to develop new methods for end-to-end dependency parsing of speech. Potential research directions include:

• multilingual evaluation of Pupier et al’s parser
• data augmentation methods (e.g. leveraging audio corpora with no syntactic annotation)
• investigate the use of multimodal pretrained models or modality fusion methods aimed to combine BERT-style models with wav2vec-style models.
• any research directions proposed by the applicant is most welcome (within the general topic of speech parsing).

Scientific environment:

The PhD will be conducted within the Getalp team of the LIG laboratory (https://lig-getalp.imag.fr/). The GETALP team has a strong expertise and track record in Natural Language Processing. The recruited person will be welcomed within the team which offer a stimulating, multinational and pleasant working environment. The PhD will be supervised by Maximin Coavoux, and Benjamin Lecouteux.

Salary: ~2000€ gross/month (social security included)

Profile:

• Master’s degree in Computer Science / NLP / Computational Linguistics
• background in NLP and/or speech processing
• Python (transformers, pytorch)
• Good communication skills in English. Rudimentary knowledge of French is welcome since it will be the focus language for the parsing experiments.

How to apply:

• please send CV + cover letter + recent (last ~2 years) academic transcripts + master thesis (if available) to maximin.coavoux@univ-grenoble-alpes.fr
• deadline: April 8th
• Informal inquiries are welcome!

References:

End-to-End Dependency Parsing of Spoken French (Interspeech 2022) Adrien Pupier, Maximin Coavoux, Benjamin Lecouteux, Jérôme Goulian https://hal.science/hal-03713551/

Wave to Syntax: Probing spoken language models for syntax Gaofei Shen, Afra Alishahi, Arianna Bisazza, Grzegorz Chrupała https://arxiv.org/abs/2305.18957