Post-editing (PE) of machine translation (MT) has become a prevailing translation modality both in the translation industry and in specialized professional settings with interlinguistic communication needs. As English is the lingua franca in academic contexts, many researchers require a level of proficiency that is often supplied by MT. Against this background, the serious game of the GAMETRAPP project was developed. This article highlights the most frequent MT errors found in scientific abstracts in the field of medicine, while also reflecting on the task of PE and the design of gamified activities for healthcare professionals.
Nominal multiword terms (MWTs) consisting of three or more elements frequently exhibit attachment ambiguities (Aguilar et al., 2011), which pose significant challenges for translation, particularly in the biomedical domain where semantic precision is essential. This study investigates the translation of 15 English biomedical noun compounds into German and Spanish after being translated using ChatGPT (version 5.1), DeepL (in its two versions: classic and next-gen), and Google Translate. The limited morphosyntactic marking of English increases the risk of ambiguity in these structures, complicating accurate transfer across languages. The results reveal persistent limitations of machine translation systems in correctly resolving syntactic relations within nominal multiword terms and underscore the necessity of expert human review for their accurate translation.
The digital age is transforming every sector of society, including translation and interpreting. Rapid advances in artificial intelligence (Al) have created a dire need for change for these fields to adapt and evolve. Given the burgeoning demand for post-editing, this paper presents a comparative study of five Al-powered gamification platforms used to automatically generate five gamified activities (GAs) tailored to train translators and interpreters in pinpointing and correcting potential Al errors in medical text translation. Following a four-phase systematic methodology, Al is examined from a dual perspective-both as a tool for the automatic generation of GAs and for the translation of medical texts-yielding promising findings that pave the way for further innovative avenues of research, such as the automatic creation of complete GAs.
The aim of this article is to evaluate the quality of asynchronous automatic interpreting tools. The study puts forward a definition of these tools and describes their architecture and modalities. We then review the quality parameters set out in the literature to construct a model of error analysis applicable to human and automatic interpreting. Lastly, this model is implemented in the analysis of a professional and an automatic interpretation of a specialized health sciences discourse.
Medical textsfrequently exhibitfeatures that compromise their clarity and hinder the effective transmission of information to patients. In this context, plain language has emerged as a key instrument for the adaptation and simplification of specialized knowledge, thereby enhancing comprehensibility. Nevertheless, the extent to which Artificial Intelligence tools may streamline these processes has not yet been examined in depth. This study compiles a corpus of previously translated and adapted oncology texts with the aim of analyzing and comparing ChatGPT's effectiveness in the clarification and de-terminologisation of medical discourse.
