Call for postdoc applications in Natural Language Processing for the automatic detection of gender stereotypes in the French media (Grenoble Alps University, France)
Starting date: flexible, November 30, 2023, at the latest
Duration: full-time position for 12 months
Salary: according to experience (up to 4142€/ month)
Application Deadline: Open until filled
Location: The position will be based in Grenoble, France. This is not a remote work.
Keywords: natural language processing, gender stereotypes bias, corpus analysis, language models, transfer learning, deep learning
*Context* The University of Grenoble Alps (UGA) has an open position for a highly motivated postdoc researcher to joint the multidisciplinary GenderedNews project. Natural Language Processing models trained on large amount of on-line content, have quickly opened new perspectives to process on-line large amount of on-line content for measuring gender bias in a daily basis (see our project https://gendered-news.imag.fr/ https://gendered-news.imag.fr/ ). Regarding research on stereotypes, most recent works have studied Language Models (LM) from a stereotype perspective by providing specific corpora such as StereoSet (Nadeem et al., 2020) or CrowS-Pairs (Nangia et al. 2020). However, these studies are focusing on the quantifying of bias in the LM predictions rather than bias in the original data (Choenni et al., 2021). Furthermore, most of these studies ignore named entities (Deshpande et al., 2022) which account for an important part of the referents and speakers in news. In this project, we intend to build corpora, methods and NLP tools to qualify the differences between the language used to describe groups of people in French news.
*Main Tasks*
The successful postdoc will be responsible for day-to-day running of the research project, under the supervision of François Portet (Prof UGA at LIG) and Gilles Bastin (prof UGA at PACTE). Regular meetings will take place every two weeks.
- Defining the dimensions of stereotypes to be investigated and the possible metrics that can be processed from a machine learning perspective.
- Exploring, managing and curating news corpora in French for stereotypes investigation, with a view to making them widely available to the community to favor reproducible research and comparison.
- Studying and developing new computational models to process large number of texts to reveal stereotype bias in news. Make use of pretrained models for the task.
- Evaluate the methods on curated focused corpus and apply it to the unseen real longitudinal corpus and analyze the results with the team.
- Preparing articles for submission to peer-reviewed conferences and journals.
- Organizing progress meetings and liaising between members of the team.
The hired person will interact with PhD students, interns and researchers being part of the GenderedNews project. According to his/her background his/her own interests and in accordance with the project's objective, the hired person will have the possibility to orient the research in different directions.
*Scientific Environment*
The recruited person will be hosted within the GETALP teams of the LIG laboratory (https://lig-getalp.imag.fr/ https://lig-getalp.imag.fr/), which offers a dynamic, international, and stimulating environment for conducting high-level multidisciplinary research. The person will have access to large datasets of French news, GPU servers, to support for missions as well as to the scientific activities of the labs. The team is housed in a modern building (IMAG) located in a 175-hectare landscaped campus that was ranked as the eighth most beautiful campus in Europe by the Times Higher Education magazine in 2018.
The person will also closely work with Gilles Bastin (PACTE, a Sociology lab in Grenoble) and Ange Richard (PhD at LIG and PACTE). The project also includes an informal collaboration with "Prenons la une" (https://prenonslaune.fr/ https://prenonslaune.fr/) a journalists’ association which promotes a fair representation of women in the media.
*Requirements*
The candidate must have a PhD degree in Natural Language Processing or computer science or in the process of acquiring it. The successful candidate should have
- Good knowledge of Natural Language Processing - Experience in corpus collection/formatting and manipulation. - Good programming skills in Python - Publication record in a close field of research - Willing to work in multidisciplinary and international teams - Good communication skills - Good mastering of French is required
*Instructions for applying*
Applications will be considered on the fly and must be addressed to François Portet (Francois.Portet@imag.fr mailto:Francois.Portet@imag.fr). It is therefore advisable to apply as soon as possible. The application file should contain
- Curriculum vitae - References for potential letter(s) of recommendation - One-page summary of research background and interests for the position - Publications demonstrating expertise in the aforementioned areas - Pre-defense reports and defense minutes; or summary of the thesis with the date of defense for those currently in doctoral studies
*References*
Deshpande et al. (2022). StereoKG: Data-Driven Knowledge Graph Construction for Cultural Knowledge and Stereotypes. arXiv preprint arXiv:2205.14036.
Choenni et al. (2021). Stepmothers are mean and academics are pretentious: What do pretrained language models learn about you? arXiv preprint arXiv:2109.10052.
Nadeem et al. (2020) StereoSet: Measuring stereotypical bias in pretrained language models. ArXiv.
Nangia et al. (2020) CrowS-Pairs: A Challenge Dataset for Measuring Social Biases in Masked Language Models. In EMNLP2020.