Until 2022, machine translation, in its various paradigms (from phrase-based models to neural machine translation, NMT), served as the main reference point for studying the incorporation of artificial intelligence (AI) in translation. The diffusion of generative AI at the end of that year marked a turning point by introducing systems capable of producing communicatively effective texts and being incorporated into professional workflows. Nevertheless, the complexity of legal discourse and the differences between legal systems continue to limit its full integration into legal translation, making human intervention essential to ensure accuracy and adequacy. Research on the application of AI in legal translation remains incipient and is concentrated in academic or institutional settings, overlooking its real integration into professional workflows as well as the transformations these tools bring about in translation practice. This thesis contributes to filling this research gap by analyzing the impact of AI on the workflows of legal translators in Spain and Italy, with special attention to NMT and generative AI. Its main objective is to assess the degree of adoption of these technologies, considering both the perceptions of market actors and the organization of workflow as well as the final quality of the translated texts. The thesis is organized into two main sections: a theoretical-methodological one and a practical-empirical one. The first section begins with a literature review, followed by a definition of the theoretical-methodological framework, in which the main technologies applied to legal translation are examined, evaluating their functioning, advantages, limitations, and implications for professional practice. This first part concludes with an analysis focused on the translation of the textual genre that is the subject of the experimental study presented in the second section. The second empirical section combines two studies under a mixed-methods approach: a study of the legal translation market in Spain and Italy, and an experimental study involving professional legal translators from Spain. On one hand, the market study, based on surveys of translators and companies, evaluates how the sector adjusts its practices in response to technological advances and measures the impact of NMT and generative AI on productivity, profitability, professional perception, and translation quality. On the other hand, the experimental study directly observed the workflow and translation output in three dimensions: efficiency and speed, translation process, and final product quality. Two groups participated in the experiment: group A worked on segments pre-translated with NMT, while group B translated from scratch according to the standard modality identified in the market study. A triangulation design was applied, including pre- and post-task questionnaires, screen recordings, and product analysis using an adapted version of the MQM-Core model. The results confirmed that the problems in AI integration are not due exclusively to biases or unfounded resistance from translators. The editing of pre-translated segments emerges as an effective strategy, whose success depends on advanced competencies in documentation, functional analysis of legal discourse, and critical revision, activities that, at least in the short and medium term, can only be performed by human translation professionals. The conclusions offer insights on both the professional practice and the academic field. It is expected that these results will contribute to a better understanding of the relationship between legal translators and AI, and will guide both the development of technologies and the adoption of practices more closely aligned with the reality of a professional category essential to the functioning of the sector.
Hasta 2022, la traducción automática, en sus distintos paradigmas (desde los modelos basados en frases hasta la traducción automática neuronal, TAN), constituyó la referencia principal para estudiar la incorporación de la inteligencia artificial (IA) en traducción. La llegada de la IA generativa a finales de ese año supuso un punto de inflexión al introducir sistemas capaces de generar textos con eficacia comunicativa e integrarse en los flujos profesionales. No obstante, la complejidad del discurso jurídico y las diferencias entre los ordenamientos jurídicos sigue limitando su completa integración en la traducción jurídica, lo que hace imprescindible la intervención humana para garantizar la precisión y la adecuación al encargo. La investigación sobre la aplicación de la IA en la traducción jurídica sigue siendo incipiente y se concentra en entornos académicos o institucionales, descuidando su integración real en los flujos de trabajo profesionales, así como las transformaciones que estas herramientas generan en la práctica traductora. Esta tesis contribuye a llenar este vacío investigador analizando el impacto de la IA en los flujos de trabajo de traductores jurídicos en España e Italia, con especial atención a la TAN y a la IA generativa. El objetivo principal es evaluar el grado de adopción de estas tecnologías, considerando tanto las percepciones de los actores del mercado como la organización del flujo de trabajo y la calidad final de los textos traducidos. La tesis se articula en dos bloques principales: uno de carácter teórico-metodológico y otro de carácter práctico-empírico. El primer bloque empieza con una revisión bibliográfica, seguida por una definición del marco teórico-metodológico, en el que se examinan las principales tecnologías aplicadas a la traducción jurídica para evaluar su funcionamiento, ventajas, limitaciones e implicaciones para la práctica profesional. Esta parte termina con un análisis orientado a la traducción del género discursivo objeto del estudio experimental presentado en el segundo bloque. La segunda parte empírica combina dos estudios bajo un enfoque de métodos mixtos: un estudio del mercado de la traducción jurídica de España y de Italia, y un estudio experimental con traductores jurídicos profesionales españoles. Por un lado, el estudio de mercado, basado en encuestas a traductores y empresas, evalúa cómo el sector ajusta sus prácticas ante los avances tecnológicos y mide el impacto de la TAN y de la IA generativa sobre productividad, rentabilidad, percepción profesional y calidad de la traducción. Por otro lado, el estudio experimental observa el flujo de trabajo y el producto de traducción en tres dimensiones: eficiencia y velocidad, proceso de traducción y calidad del producto final. En el experimento participan dos grupos: el grupo A trabaja sobre segmentos pretraducidos con TAN, mientras que el grupo B traduce desde cero según la modalidad estándar identificada en el estudio del mercado. Se aplica un diseño de triangulación con cuestionarios pre- y post-task, grabaciones de pantalla y análisis del producto mediante una versión adaptada del modelo MQM-Core. Los resultados muestran que los problemas en la integración de la IA no derivan solo de una resistencia infundada por parte de los traductores. La edición de segmentos pretraducidos es eficaz, pero su éxito exige competencias en documentación, análisis del discurso jurídico y revisión crítica, tareas que, al menos a corto y medio plazo, solo pueden realizar profesionales humanos de la traducción. Se espera que estos hallazgos ayuden a comprender mejor la relación entre traductores jurídicos e IA y a orientar tanto el desarrollo tecnológico como la adopción de prácticas más acordes con la realidad de un colectivo esencial para el sector.
La traducción jurídica en la era de la inteligencia artificial: una aproximación empírica / Francesca Accogli , 2026 Mar 23. 38. ciclo, Anno Accademico 2024/2025.
La traducción jurídica en la era de la inteligencia artificial: una aproximación empírica
ACCOGLI, FRANCESCA
2026-03-23
Abstract
Until 2022, machine translation, in its various paradigms (from phrase-based models to neural machine translation, NMT), served as the main reference point for studying the incorporation of artificial intelligence (AI) in translation. The diffusion of generative AI at the end of that year marked a turning point by introducing systems capable of producing communicatively effective texts and being incorporated into professional workflows. Nevertheless, the complexity of legal discourse and the differences between legal systems continue to limit its full integration into legal translation, making human intervention essential to ensure accuracy and adequacy. Research on the application of AI in legal translation remains incipient and is concentrated in academic or institutional settings, overlooking its real integration into professional workflows as well as the transformations these tools bring about in translation practice. This thesis contributes to filling this research gap by analyzing the impact of AI on the workflows of legal translators in Spain and Italy, with special attention to NMT and generative AI. Its main objective is to assess the degree of adoption of these technologies, considering both the perceptions of market actors and the organization of workflow as well as the final quality of the translated texts. The thesis is organized into two main sections: a theoretical-methodological one and a practical-empirical one. The first section begins with a literature review, followed by a definition of the theoretical-methodological framework, in which the main technologies applied to legal translation are examined, evaluating their functioning, advantages, limitations, and implications for professional practice. This first part concludes with an analysis focused on the translation of the textual genre that is the subject of the experimental study presented in the second section. The second empirical section combines two studies under a mixed-methods approach: a study of the legal translation market in Spain and Italy, and an experimental study involving professional legal translators from Spain. On one hand, the market study, based on surveys of translators and companies, evaluates how the sector adjusts its practices in response to technological advances and measures the impact of NMT and generative AI on productivity, profitability, professional perception, and translation quality. On the other hand, the experimental study directly observed the workflow and translation output in three dimensions: efficiency and speed, translation process, and final product quality. Two groups participated in the experiment: group A worked on segments pre-translated with NMT, while group B translated from scratch according to the standard modality identified in the market study. A triangulation design was applied, including pre- and post-task questionnaires, screen recordings, and product analysis using an adapted version of the MQM-Core model. The results confirmed that the problems in AI integration are not due exclusively to biases or unfounded resistance from translators. The editing of pre-translated segments emerges as an effective strategy, whose success depends on advanced competencies in documentation, functional analysis of legal discourse, and critical revision, activities that, at least in the short and medium term, can only be performed by human translation professionals. The conclusions offer insights on both the professional practice and the academic field. It is expected that these results will contribute to a better understanding of the relationship between legal translators and AI, and will guide both the development of technologies and the adoption of practices more closely aligned with the reality of a professional category essential to the functioning of the sector.I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.


