POSTDOCTORAL POSITION

Postdoctoral Researcher in AI/NLP

Creation of synthetic argumentative datasets for preferences: from
simple to complex arguments with integrated validation

GRAIL project | Scientific supervision: Benoit Crabbé and Onn Sherry

The opportunity. Develop AI methods that align texts with structured
arguments and progress from reliable simple cases to challenging
composite arguments.


Research project

The research will first align simple structured arguments with their
corresponding natural-language texts, creating a controlled dataset for
training and evaluating preference-based argument parsers. We will study
several generation and alignment methods involving an understanding of
typical errors. Both structure-to-text and, where useful,
text-to-structure generation may be studied. The work will then progress
to composite arguments that combine simpler arguments through shared
premises, intermediate conclusions, supports, attacks, preferences, and
other interdependencies.

Candidate methods include GNNs, graph attention networks, graph
transformers, language models, and hybrid architectures. Validation will
accompany generation as complexity grows, assessing structural
correctness, semantic fidelity, coherence, and consistency. Automatic
correction is an optional stretch goal.

Main activities

-   Create simple and composite argument structures aligned with
    corresponding texts.
-   Build controlled synthetic datasets and complexity-based evaluations
    for parser training.
-   Compare graph-native, language-model, and hybrid architectures with
    rigorous baselines.
-   Integrate validation into dataset curation and publish reusable
    datasets, models, and code.

Candidate profile

-   PhD in NLP, AI, machine learning, computer science, or a closely
    related field by the starting date.
-   Strong expertise in modern AI/NLP, including Transformers or large
    language models and rigorous evaluation.
-   Strong programming skills, preferably Python and PyTorch, and the
    ability to conduct independent, collaborative research in English.
-   Additional assets. Graph learning, structured prediction, synthetic
    data, argumentation, or logic are welcome. Logic or argumentation
    expertise is advantageous but not required.

APPLICATION INFORMATION

Position details and application Complete the items in square brackets
before publication.

Employment conditions
Host institution: Université Paris Cité
Location: Paris, France
Contract: Fixed-term postdoctoral contract; 2 years

Expected start: January 1st, 2027
Application deadline: November 1st 2026

Research environment

The researcher will be jointly supervised by Benoit Crabbé and Onn
Shehory and will collaborate with members of the GRAIL project:
gathering specialists in NLP, machine learning, argumentation, and
reasoning, with access to project data and computing resources.

How to apply
-   Submit as a single PDF including
-   a Curriculum vitae, including a publication list.
-   Brief cover letter describing motivation and fit.
-   Research statement of up to two pages covering relevant experience 
    and an initial approach.

Contact details for two referees and one or two representative
publications.

Application email: benoit.crabbe@u-paris.fr and onn.shehory@biu.ac.il
Inquiries: 
    Benoit Crabbé benoit.crabbe@u-paris.fr
    Onn Shehory: onn.shehory@biu.ac.il