The presentation is about the experiment and findings from my Master’s in Technopreneurship thesis at the University of Luxembourg, focusing on the business context and application within the Translation Centre for the Bodies of the EU (CdT).
The technopreneurship project investigates whether an LLM can automate the post-editing of machine translation (MT) output—a resource-intensive task currently performed by human linguists—and identify recurring MT error patterns.
The proof of concept uses OpenAI’s GPT 4-o accessing it through the European Commission’s pilot project GPT@JRC platform, supported with a structured, prompt, domain-specific reference material and a dataset English to French from European Medicines Agency in the public health domain. The LLM generated output was benchmarked against the final human version of translation.
The results demonstrated that the text generated by the LLM was in average 1) slightly higher in terms of the number of edits (effort) needed to match the reference 2) semantically closer to the human reference, more consistent, context aware and better aligned with the templates and product names.
The balancing of the above factors is not clear cut, however edit distance metric used for the quantitative measurement may have overestimated the effort of making long edits that could be single edits for humans. Besides, having a text that preserves consistently product names, the meaning of the original, is more context aware and executes fast, can lead to efficiency gains.
Therefore, the proof of concept confirmed feasibility of the approach, encouraging for further experimentation with other LLMs and state-of-the-art technologies such as retrieval augmented generation (RAG) and advanced prompt engineering techniques.
Technical standardisation and standards related to translation services and AI, contributed in a multifaceted way to the project i.e., in identifying risks related to biases, inaccuracies or hallucinations, and aspects of digital trust including trustworthiness, data privacy, security, transparency.
I am currently Head of Section (a.i.) at the Translation Centre for the Bodies of the EU, with 30 years of experience across research, industry and EU institutions. My expertise spans linguistic services, machine translation and artificial intelligence, with a particular focus on bridging language technologies with business innovation and real-world applications. I have extensive experience coordinating teams and multilingual R&D projects and initiatives, and driving change in a rapidly evolving technological landscape. My academic background includes a Master’s degree in Technopreneurship, an MSc in Machine Translation, a Postgraduate Diploma in Computer-assisted Translation and Language Learning, and a Bachelor’s degree in Mathematics. Fluent in Greek, English, Spanish, Portuguese, French and Italian, and with intermediate German, I am passionate about multilingualism and impactful AI, and particularly interested in how emerging technologies can create practical value while supporting Europe’s linguistic and cultural diversity.
