Machine Translation in the Age of Artificial Intelligence: Opportunities, Limitations and Professional Implications

The present thesis explores the transformative impact of Artificial Intelligence (AI) on Machine Translation (MT), with particular emphasis on literary translation. Through a systematic review of recent literature, the study adopts a qualitative methodology in order to investigate the opportunities, limitations and professional implications generated by the application of Neural Machine Translation (NMT) systems and Large Language Models (LLMs) within translation practice. The analysis provides a historical overview of MT, establishing the evolutionary path from rule and statistical-based methods to the current neural and transformer-based systems. The main opportunities presented by AI-supported translation are outlined, including major increases in efficiency, speed and cost-effectiveness, democratisation of access to multilingually published content and improved quality of output where terminological consistency and volume are prioritised. Specifically, AI tools have been analysed as potential assistants in post-editing workflows, allowing translators to devote more attention to higher-level interpretive and stylistic decisions rather than mechanical elements of the translation process. Nonetheless, there is a series of limitations. The main characteristics of literary language, including the voice of the author, intertextuality, creative deviation from linguistic norms, and genre-specific stylistic conventions, are problematic for contemporary AI systems to address. In particular, the distinctive elements of literary texts, such as metaphors and idiomatic expressions, create significant barriers to AI. Another major limitation relates to the phenomenon of translationese, which can result in over-simplified syntax and lexical uniformity, particularly damaging in the literary context. Moreover, AI systems struggle to handle uncommon vocabulary, domain-specific vocabulary and low-resource language pairs, thus raising concerns regarding cultural homogenisation. Ethical and social dimensions are also central to the analysis. Particularly, the thesis makes reference to algorithmic bias; transparency and accountability issues in the use of automated systems; data privacy; and the risk of what UNESCO defines as “the great linguistic flattening”, the systematic underrepresentation of minority and low-resource languages in training data, which risks perpetuating and deepening existing linguistic inequalities at a global scale. From a professional perspective, the study argues that the translator’s role is being redefined rather than replaced. Human translators are increasingly embracing the role of cultural curators, with an emphasis on an expanded set of translation-related skills that comprise an understanding of AI-enabled tools, enhanced post-editing capabilities, and a deepened awareness of the ethical responsibilities related to mediated cultural communication. As the profession evolves, translation education might need to integrate linguistic and analytical skills with critical AI literacy. In conclusion, the thesis advocates for a balanced, knowledge-based use of AI translation technologies. While it demonstrates the efficiency and benefit of providing global access to multilingual communication through AI translation, it also sheds light on the need of human judgement, knowledge, and ethical responsibility. The development of AI-based translation technology could rest on a human-AI collaboration, where human agency is considered to be essential for safeguarding the richness and authenticity of human expression.