Title: Advancing Responsible and Epistemically Robust Multilingual LLMs for the Social Sciences and Humanities

Abstract:

The rapid advancement of multilingual capabilities and reasoning prowess in large language models has paved the way for their incorporation into research workflows within the Social Sciences and Humanities. However, current evaluation methods are predominantly tied to task-oriented NLP benchmarks, leaving critical issues such as interpretive validity, cultural context, and epistemic mediation largely unaddressed. This study reframes multilingual reasoning models as hermeneutic tools that play an active role in shaping meaning production across diverse linguistic and cultural landscapes. By synthesizing insights from hermeneutics, the philosophy of technology, science and technology studies, multilingual NLP, and computational social science methodologies, we propose a theoretically robust framework for assessing multilingual reasoning in SSH research. Our approach includes a detailed experimental protocol featuring operationalized metrics for cultural alignment, cross-lingual consistency, and reasoning fidelity, alongside transparency standards specifically designed for interpretive research tasks. We demonstrate the utility of this framework through a practical case study focused on multilingual political discourse analysis. Ultimately, this work provides a conceptual and methodological basis for the ethical and effective integration of multilingual reasoning LLMs into computational social science systems.

Source: arXiv Generated at: 2026-06-02 00:00:00 UTC