Domain-Adaptive LLMs for Social Sciences and Humanities Research

Adam Faci, Alessio Miaschi, Anne Combe, Pascal Cuxac, Francesca Frontini, Nicolas Larrousse, St\'ephane Pouyllau· July 8, 2026 View original

Summary

This paper details an ongoing project within LLMs4EU and ALT-EDIC to adapt foundation models for Social Sciences and Humanities (SSH) research, integrating knowledge graphs and multilingual scholarly corpora. The initiative aims to support tasks like question answering and literature review while addressing methodological, epistemic, and regulatory challenges.

The integration of large language models (LLMs) into scientific research, particularly for tasks like bibliographic discovery and literature synthesis, presents unique challenges for the Social Sciences and Humanities (SSH). These challenges include accommodating disciplinary diversity, ensuring multilingual access to sources, and rigorously evaluating results. An ongoing project, part of the European LLMs4EU initiative and ALT-EDIC infrastructure, is addressing these issues by adapting foundation models specifically for SSH research practices. This involves integrating knowledge graphs with multilingual scholarly corpora to enhance the LLMs' domain-specific understanding. The goal is to support critical SSH tasks such as question answering, comparative document analysis, and literature reviews. The project employs a comprehensive evaluation framework, including both quantitative benchmarking (for retrieval, summarization, traceability, and hallucination detection) and qualitative assessment by Digital Humanities experts. By embedding model adaptation within established research infrastructures and adhering to strict legal and ethical compliance, the project aims to develop reliable and epistemically responsible generative AI tools for SSH scholarship.

Why it matters

For professionals in research, education, and public policy, this work demonstrates how LLMs can be responsibly adapted for specialized domains like SSH, improving literature review, knowledge discovery, and comparative analysis while addressing critical ethical and methodological concerns.

How to implement this in your domain

  1. 1Explore domain-adaptive LLM strategies for specialized knowledge domains beyond SSH, such as legal, medical, or engineering fields.
  2. 2Investigate integrating knowledge graphs and multilingual corpora to enhance the accuracy and relevance of LLM outputs for specific professional contexts.
  3. 3Develop robust evaluation frameworks, combining quantitative metrics and expert qualitative assessment, for domain-adapted LLMs.
  4. 4Establish clear legal and ethical compliance frameworks for deploying generative AI in sensitive research or professional applications.
  5. 5Collaborate with domain experts to fine-tune and validate LLMs for specific tasks like question answering or literature synthesis within their fields.

Who benefits

AcademiaResearch & DevelopmentEducationGovernmentPublishing

Key takeaways

  • LLMs are being adapted for Social Sciences and Humanities research.
  • Integration of knowledge graphs and multilingual corpora is key.
  • The project addresses methodological, ethical, and regulatory challenges.
  • It aims to support tasks like question answering and literature review responsibly.

Original post by Adam Faci, Alessio Miaschi, Anne Combe, Pascal Cuxac, Francesca Frontini, Nicolas Larrousse, St\'ephane Pouyllau

"arXiv:2607.05956v1 Announce Type: new Abstract: The integration of Large Language Models (LLMs) into scientific research workflows, particularly for bibliographic discovery and literature synthesis, raises significant methodological, epistemic and regulatory challenges for the So…"

View on X

Originally posted by Adam Faci, Alessio Miaschi, Anne Combe, Pascal Cuxac, Francesca Frontini, Nicolas Larrousse, St\'ephane Pouyllau on X · view source

Want to go deeper?

Turn these trends into skills with Learnijoy's hands-on AI & tech courses.

Explore courses

More in AI Research

AI ResearchAI Engineering & DevTools

New Theory Explains Neural Network Generalization Beyond Overfitting

This research proposes a new theoretical framework to explain why neural networks can generalize effectively even when over-parameterized. It links this phenomenon to a phase transition in the training process, marked by broken ergodicity and a breakdown of the fluctuation-dissipation theorem.

Chan Li, Nigel GoldenfeldJul 8, 2026
AI Engineering & DevToolsAI Research

K-ABENA Optimizes Backpropagation by Excluding Low-Loss Samples

K-ABENA (K-Adaptive Backpropagation with Error-based N-exclusion Algorithm) is a new selective gradient computation framework that reduces neural network training costs by excluding a fraction of low-loss observations from the backward pass. It uses a compensated sampling design and reweighting to achieve unbiased gradient estimation, maintaining high accuracy while significantly saving computational resources.

Jean-Francois BonbhelJul 8, 2026
AI Engineering & DevToolsAI Research

New Auditor Evaluates AI Unlearning Algorithm Effectiveness

Researchers propose a practical auditor that uses membership inference attacks to compute data-dependent lower bounds on the unlearning parameter, evaluating how effectively unlearning algorithms remove training data influence. The auditor reveals a sharp separation, confirming rigorous unlearning guarantees for methods like model clipping while exposing poor unlearning in empirical methods.

Sahasrajit Sarmasarkar, Anastasia Koloskova, Sanmi KoyejoJul 8, 2026