ADAPT DCU researchers have had three papers accepted into AMTA this year. AMTA 2022 is the 15th biennial conference of the Association for Machine Translation in the Americas. It brings together researchers and users of machine translation from business and government in a single hybrid event with in-person and online participation.
Pintu Lohar (ADAPT DCU), Maja Popović (ADAPT DCU) and Tanya Habruseva (LinkedIn) have co-written the paper “Building Machine Translation System for Software Product Descriptions Using Domain-specific Sub-corpora Extraction”. This work is as a result of their three month collaboration with LinkedIn. The end goal was to build an English-to-French Neural Machine Translation (NMT) system that could translate the software product descriptions from English to French.
Wandri Jooste (ADAPT DCU), Andy Way (ADAPT DCU), Rejwanul Haque (ADAPT NCI) and Riccardo Superbo (KantanAI) co-wrote the paper “Knowledge Distillation for Sustainable Neural Machine Translation” which has been accepted to the user track. The paper details the process of distilling knowledge to reduce model size and training time without a significant loss in performance. The results of their investigation will demonstrate to the translation industry a cost effective and quality alternative to standard Knowledge Distillation training methods.
Yasmin Moslem (ADAPT DCU), Rejwanul Haque (ADAPT NCI), John Kelleher (ADAPT TU Dublin), and Andy Way (ADAPT DCU) wrote the paper “Domain-Specific Text Generation for Machine Translation”. The paper proposes a novel approach to domain adaptation leveraging state-of-the-art pretrained language models (LMs) for domain-specific data augmentation for MT, simulating the domain characteristics of either a small bilingual dataset, or the monolingual source text to be translated.
This year’s conference will take place in Orlando, Florida from the 12 – 16 September. More information about the conference is available on their website here.