ADAPT DCU and MTU researchers are set to showcase their work at the upcoming AMTA conference, which will be held from September 30th to October 2nd in Chicago. The team includes Inacio Vieira (Dublin City University), Will Allred (Dublin City University), Séamus Lankford (Munster Technological University), Sheila Castilho (Dublin City University), and Andy Way (Dublin City University), who will present two key papers:
“How Much Data is Enough Data? Fine-Tuning Large Language Models for In-House Translation: Performance Evaluation Across Multiple Dataset Sizes”
Co-authored by Inacio, Will, Séamus, Sheila and Andy, this paper investigates the effectiveness of fine-tuning Large Language Models (LLMs) utilising client-specific translation memories to improve translation accuracy and efficiency.
While decoder-only LLMs have demonstrated strong performance in machine translation (MT) due to their ability to learn from vast datasets and produce high-quality translations, they often struggle with capturing the specific nuances and style required for organisation-specific translations. The study examines translations for five language-pairs with varying resource levels (English to Brazilian Portuguese, Czech, German, Finnish, and Korean) and assesses the effect of different dataset sizes on translation quality.
“Leveraging LLMs for MT in Crisis Scenarios: A Blueprint for Low-Resource Languages” Presented by Séamus and Andy, this study explores the use of LLMs and Multilingual LLMs to enhance MT in crisis situations. The research addresses the unique challenges of crisis scenarios, where speed, accuracy, and the ability to handle a range of resource-poor languages are critical.
It proposes a new approach that integrates cutting-edge LLM capabilities with fine-tuning techniques and community-driven corpus development strategies to improve translation outcomes for low-resource languages.
AMTA is the North American chapter of the International Association for Machine Translation (IAMT), which also includes the Asian-Pacific and European Associations for Machine Translation. For more information about the conference, visit AMTA.
