Miloslav's engagement with applied linguistics began during high school, where he was first introduced to the morphology of German, English, Italian, and Latin. He pursued further education at Comenius University in Bratislava, culminating in a PhD (Permanent head Damage:) in Slovak-Turkish relations. Academic journey included research stints at Charles University in Prague, Czechia, Ankara Üniversitesi and Uludağ Üniversitesi in Bursa, Türkiye. Additionally, he interned at the Consulate General of the Slovak Republic in Istanbul and attended postgraduate studies at the Institute of International Relations and Legal Comparative Studies in Bratislava.
His professional experience spans across London, Istanbul and Bratislava within an international corporation. Since 2014, he has been working for a private SME, focusing on translating content into various languages using machine learning, artificial intelligence and human translation.
Fine-Tuning LLMs with Client Data: A Case Study in AI-Powered Localization
Inside the Session
The translation division of exe embarked on its AI journey by leveraging its vast repository of translation memories (TMs) to fine-tune large language models (LLMs). Recognizing the limitations of generic neural machine translation (NMT), exe saw an opportunity to enhance domain-specific accuracy and consistency using its curated bilingual corpora.
The team began by aligning and preprocessing TMs to serve as high-quality training data, ensuring linguistic precision and contextual relevance. To meet the demanding hardware and software requirements, exe invested in scalable cloud infrastructure with GPU acceleration and adopted open-source frameworks.
Upskilling internal staff was crucial - linguists collaborated with data scientists to bridge the gap between language expertise and machine learning.
The results were compelling: the fine-tuned models outperformed general and customized NMT engines, especially in specialized terminology and stylistic fidelity.