Human–machine interaction and legal translation

There is growing recognition that linguistic data within institutional or corporate environments constitutes a strategic asset. This paper presents a 12-month workplace research project at the Italian Ministry of Justice aimed at introducing translation technologies into a low-technology work environment staffed by highly specialized legal linguists, with a focus on monitoring human–machine interaction. Adopting a translator-centered, bottom-up approach in line with the principles of augmented translation, the participants were supported by the researcher-trainer in creating translation memories (TMs) and termbases (TBs), and in collecting sufficient linguistic data to customize an adaptive neural machine translation (NMT) engine. Participants leveraged years of linguistic expertise and maintained complete control over the technologies by combining TMs, TBs, NMT, and MT evaluation tools. This approach addressed the pitfalls of an NMT engine, which requires substantial time, effort, and linguistic data to be trained properly, and helped identify a suitable workflow for the Ministry's file formats, document types, timing, and confidentiality requirements. This study contributes to potential solutions for challenges in adopting (N)MT in public administration or institutions: lack of technical expertise, heterogeneous needs that cannot be covered by a single engine, confidentiality constraints, difficulties in evaluating MT performance, and limitations of online platforms.