Integrating Knowledge Graphs and Multilingual Scholarly Corpora for Domain-Adaptive LLMs in SSH

Jul 16, 2026

Upload by Miles

The integration of Large Language Models (LLMs) into scientific research workflows—especially for bibliographic discovery and literature synthesis—creates methodological, epistemic, and regulatory challenges for Social Sciences and Humanities (SSH). The paper reports an ongoing use case in the European LLMs4EU project and ALT-EDIC infrastructure, adapting foundation models to SSH research practices. It supports question answering, comparative document analysis, and literature review, evaluated through both quantitative benchmarking and qualitative assessment by Digital Humanities experts.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages8

Reading Time20 min

Abstract

Introduction

Challenges in SSH Language and Validation

Limitations of English-centric AI Literature Tools

Interpretation-Driven SSH and Methodological Affinities

SSH Infrastructures and Opportunities for LLMs

arXiv :2607 .05956v 1 [ cs .AI ] 7 Jul 2026 Integrating Knowledge Graphs and Multilingual Scholarly Corpora for Domain-Adaptive LLMs in SSH Adam Faci1 , Alessio Miaschi2 , Anne Combe3 , Pascal Cuxac4 , Francesca Frontini2 , Nicolas Larrousse1 , Stéphane Pouyllau1 1 Huma-Num, CNRS, Paris, France 2 CNR-Institute for Computational Linguistics “A. Zampolli” (CNR-ILC), Pisa, Italy 3 Inria, France 4 Inist, CNRS, 2 rue Jean Zay, 54500 Vandoeuvre-lès-Nancy, France {[name.surname}@huma-num.fr](name.surname}@huma-num.fr), {[name.surname}@ilc.cnr.it](name.surname}@ilc.cnr.it) [anne.combe@inria.fr](anne.combe@inria.fr), [pascal.cuxac@inist.fr](pascal.cuxac@inist.fr) 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 Social Sciences and Humanities (SSH), especially with regard to disciplinary diversity, multilingual access to sources and the evaluation of results. This paper presents an on-going use case developed within the European project LLMs4EU and the ALT-EDIC infrastructure, aimed at adapting foundation models to SSH research practices and supporting tasks such as question answering, comparative document analysis and literature review. The evaluation framework follows the LLMs4EU protocol and encompasses both independent quantitative benchmarking (retrieval, summarisation, traceability and hallucination detection) and a qualitative assessment involving a panel of Digital Humanities experts. By embedding model adaptation within research infrastructures and a structured legal and ethical compliance framework, the use case explores how domain-sensitive and regulation-aware generative AI can support SSH scholarship while preserving reliability and epistemic responsibility. Keywords: Scientific research workflows, Digital Humanities, Retrieval Augmented Generation 1. Introduction In recent years, Large Language Models (LLMs) have profoundly transformed the landscape of language technologies and research practices in Social Sciences and Humanities (SSH-see, among possible examples, Arachchige et al. 2025) . At the same time, the development of both LLMs and AIenhanced scientific discovery platforms remains strongly concentrated on dominant languages and cultural contexts. It also tends to privilege epistemological models and validation practices that are typical of the so-called “hard” sciences. This imbalance is reflected in many established and emerging AI-powered platforms for exploring scientific literature. Semantic search engines (e.g. Semantic Scholar), AI-assisted discovery tools (e.g. Elicit), large-scale citation indexes (e.g. Scopus and Web of Science), and conversational systems integrated into publishing ecosystems (e.g. Scite Assistant) are predominantly trained on large corpora of English-language journal articles. Inmost cases, they allow querying primarily or exclusively in English, prioritise peer-reviewed journal publications over other forms of scholarly output (such as monographs, critical editions, or research data), and rely heavily on citation-based metrics—such as impact factor or h-index—to rank and filter results. Such design choices are not neutral. They re- flect and reinforce specific models of scientific production, visibility, and authority that may not align with the diversity of practices in SSH, where multilingualism, book-based scholarship, local research traditions, and qualitative evaluation criteria remain central. As a consequence, SSH scholars working in less-resourced languages or within epistemic traditions that do not conform to citationdriven models risk being further marginalised in the emerging AI-mediated research ecosystem. Moreover, research in the SSH often relies on interpretation and contextualisation. Texts, data, and metadata are not treated as fixed or neutral objec

FAQs

What problem does the paper address for Social Sciences and Humanities (SSH) when using LLMs?

It focuses on methodological, epistemic, and regulatory challenges for bibliographic discovery and literature synthesis in SSH, where disciplinary diversity and multilingual access are central.

Which project and infrastructure are used in the proposed use case?

The work is an ongoing use case developed within the European project LLMs4EU and the ALT-EDIC infrastructure.

How is the system evaluated in the paper?

Evaluation follows the LLMs4EU protocol and combines independent quantitative benchmarking (retrieval, summarisation, traceability, hallucination detection) with a qualitative assessment by Digital Humanities experts.

Jul 16, 2026

Upload by Miles