Improving The Automatic Detection Of Word Meaning Negotiation Indicators In Conversation Through Active Learning H/F - 75

A propos d'Inria

Inria est l'institut national de recherche dédié aux sciences et technologies du numérique. Il emploie 2600 personnes. Ses 215 équipes-projets agiles, en général communes avec des partenaires académiques, impliquent plus de 3900 scientifiques pour relever les défis du numérique, souvent à l'interface d'autres disciplines. L'institut fait appel à de nombreux talents dans plus d'une quarantaine de métiers différents. 900 personnels d'appui à la recherche et à l'innovation contribuent à faire émerger et grandir des projets scientifiques ou entrepreneuriaux qui impactent le monde. Inria travaille avec de nombreuses entreprises et a accompagné la création de plus de 200 start-up. L'institut s'eorce ainsi de répondre aux enjeux de la transformation numérique de la science, de la société et de l'économie.Improving the Automatic Detection of Word Meaning Negotiation Indicators in Conversation through Active Learning
Le descriptif de l'offre ci-dessous est en Anglais
Type de contrat : CDD

Contrat renouvelable : Oui

Niveau de diplôme exigé : Bac +5 ou équivalent

Fonction : Chercheur contractuel

Niveau d'expérience souhaité : Jeune diplômé

Contexte et atouts du poste

Place of work: INRIA Paris, ALMAnaCH project-team

Starting date: FromJune 2026 (flexible)
Duration: 6 months

Context
In conversation, speakers sometimes signal that a word or expression is unclear or disputed, leading to short exchanges where meaning is clarified or negotiated. This internship focuses on improving the automatic detection of such interactions in conversational data. Building on existing models, the project will explore active learning to efficiently improve detection performance with limited annotation. The work aims to support the large-scale analysis of meaning-related misunderstandings and may result in improved models, new annotated data, and insights on current models' capabilities.

Keywords: natural language processing, word meaning (negotiation), computational lexical semantics, semantic alignment, semantic coordination, interactional linguistics, misunderstanding, clarification requests

Supervision: Chloé Clavel, Aina Garí Soler

Mission confiée

Description:

In natural, everyday conversation, speakers are sometimes faced with unclear, ambiguous, or contested words or expressions. These situations can give rise to Word Meaning Negotiations (WMNs, Myrendal, 2015): exchanges in which participants explicitly ask for clarification or question or negotiate the meaning of a word or phrase. WMNs are valuable for studying how people detect, signal, and deal with misunderstandings, and offer a lens on collaborative meaning construction and repair mechanisms in dialog (Bazzanella and Damiano, 1999). They can also support the detection of unclear word usages and contribute to the development of dialogue systems that can detect, avoid and resolve such situations as humans do.

A typical WMN sequence consists of three parts: a trigger (a problematic word usage), an indicator (an utterance signaling the need to clarify or challenge word meaning), and a negotiation (one or more turns where speakers resolve the issue). Among these, indicators are the easiest to identify automatically, making them a key target for the detection of this kind of interaction.

Previous work on detecting indicators has relied either on regular expressions, followed by manual annotation (Garí Soler et al., 2025a) or on automatic detection through language models (Garí Soler et al., 2025b). While automatic detection shows promise, performance is still limited, and further improvements are constrained by data scarcity. This makes the problem particularly well suited to an active learning approach, allowing models to improve iteratively and focusing annotation effort on the most informative examples.

Principales activités

The goal of this internship is to improve the automatic detection of WMN indicators by designing and evaluating an active learning framework, potentially combined with simple regular-expression based methods, whereby models will iteratively improve by selectively annotating the most uncertain or informative instances. This work is a crucial step toward reducing annotation costs for the collection of WMN data from diverse conversational corpora. During the internship, the candidate will:

- Become familiar with WMN, conversational repair, existing active learning methodologies and the previous models and datasets used for this task;
- Propose and implement active learning strategies for WMN indicator detection (human annotation can be provided),
- Select and work with different types of conversational data;
- Experiment with the use of conversational context;
- Perform error analysis and evaluate and compare to previous approaches.

Ideally, the internship will result in (1) improved methods and/or models for WMN indicator detection, (2) additional annotated data for this task, and (3) a clearer understanding or model limitations.

Depending on its outcome, the work done during this internship can lead to a research paper to be submitted to a relevant NLP conference or workshop.

References:

Carla Bazzanella and Rossana Damiano. 1999. . Journal of Pragmatics, 31(6):817-836.

Aina Garí Soler, Jenny Myrendal, Chloé Clavel, and Staffan Larsson. 2025b. . PREPRINT (Version 1) available at Research Square.

Aina Garí Soler, Matthieu Labeau, and Chloé Clavel. 2025.. In Findings of the Association for Computational Linguistics: EMNLP 2025, Suzhou, China, November 5-7.

Jenny Myrendal. 2015. . Ph.D. thesis, University of Gothenburg.

Relevant links

ALMAnaCH:

Chloé Clavel:
Aina Garí Soler:

Compétences

Technical skills and level required :

Languages :

Relational skills :

Other valued appreciated :

Avantages

- Subsidized meals
- Partial reimbursement of public transport costs
- Leave: 7 weeks of annual leave + 10 extra days off due to RTT (statutory reduction in working hours) + possibility of exceptional leave (sick children, moving home, etc.)
- Possibility of teleworking (after 6 months of employment) and flexible organization of working hours
- Professional equipment available (videoconferencing, loan of computer equipment, etc.)
- Social, cultural and sports events and activities
- Access to vocational training
- Social security coverage