Answers are grounded in this paper and cite their sources.
LLMs are largely applied in legal settings, as well as in legal translation. This paper wishes to explore the reliability and accuracy of the automatic translations of an extract of a law report sourced from the Corpus of Historical English Law Reports (CHELAR). To this aim, two translation tasks are performed: a first automatic translation from English into Italian, and a second machine back-translation into English of a post-edited version of the target text. The first task is carried out by using the following platforms: ChatGPT, Gemini, ChatSonic, DeepL, Google Translate, and Reverso. In the first task, the automatically generated target texts are post-edited via corpus analysis and on the basis of monolingual dictionary entries. For this purpose, amongst others, the Bononia Legal Corpus (BoLC) and the Black’s Law Dictionary are consulted. The second machine translation task (namely, a back-translation into English) is performed by ChatGPT and Google Translate. The back-translations are then compared to the initial source text. The paper findings report soma inaccuracies in the first automated translations by the LLM and MT platforms. Such interfaces generate shortcomings at both syntactic and collocational level, where word meanings and usages in context are not always catered for. Conversely, the second task does not give rise to issues, and the back-translations are overall satisfactory. As a whole, this paper underscores that human supervision is necessary, thereby highlighting the importance of translators as mediators and post-editors. In particular, corpus users can compensates for LLM-driven and MT shortcomings, provided that the legal principles referred to in the source text are acknowledged.
No takes yet. Share an insight, caveat, or question.
Patrizia Giampieri (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: