Language models already help researchers read, translate, search, summarise, code, and analyse text. The question is not whether they can be useful, but how to decide which language technology is appropriate for a particular research purpose, when adaptation is justified, and whether its output is credible enough to support research. Beyond general-purpose chat models lies a wider landscape of language-based systems: encoder models commonly used for classification, semantic search, and matching; domain-adapted and fine-tuned models; and retrieval-based systems, each involving different trade-offs in performance, scale, cost, reproducibility, and data requirements. Drawing on recent examples from economics and the social sciences, the talk works through how to choose among these options for a given research task and how to improve a chosen system through prompting, retrieval, fine-tuning, or domain-adaptive pretraining.
Friday, 26 June, the 10th edition of the Grande Région Cahiers was officially presented in Esch-Belval. Launched in 2019 by Franz Clément and published by LISER, this publication series explores key issues affecting the Greater Region.
