LLMs as smart concordancers for legal translation practice

There are several literature-driven studies exploring how intelligent machines and humans can interact to streamline and improve legal translation processes. Nonetheless, no scholarly research has focused on the employment of GenAI/LLMs as concordancers in the legal translation workflow. This paper aims at filling this epistemic lacuna by considering Notebook LM. For this purpose, an extract of an Italian certificate of limited partnership (società in accomandita semplice) is translated into English by leveraging Notebook LM and querying it as a corpus concordancer. Hence, 45 certificates of partnerships drawn up in English are uploaded to the LLM system and used as reference documents for the translation task. Prompts are formulated with a view to extracting collocates, disambiguating words in context, verifying word usages, obtaining concordances, and clarifying relevant legilinguistic aspects. To complete the analysis, the US Uniform Partnership Act 1997 – Last Amended 2013 (“UPA”) is also consulted via a separate folder of Notebook LM. The findings indicate the importance of formulating prompts as clearly and completely as possible in order to retrieve translation-salient information. The results also show that LLMs can be seamlessly integrated into the translation workflow: prompts can feature simple language and are articulated in a user-friendly environment, thus enabling accessibility by non-ICT specialists; responses are, most of the times, straightforward and coherent, thereby facilitating the reading and interpretation of corpus data. However, the upload of targeted and field-relevant documents, as well as expert knowledge are necessary to draft ad hoc queries, eschew unrelated responses, and interpret results consistently.