Multilingual Machine Translation
Multilingual machine translation (MMT) aims to build systems capable of translating between many language pairs using a single model, improving cross-lingual communication and access to information. Current research focuses on leveraging large language models (LLMs), comparing encoder-decoder and decoder-only architectures, and optimizing training strategies like supervised fine-tuning and data augmentation to enhance translation quality, particularly for low-resource languages. These advancements are significant because they address the limitations of traditional NMT approaches, improving translation accuracy and efficiency across diverse linguistic contexts and potentially impacting fields like global communication, information access, and cross-cultural understanding.
Papers
Multilingual Pixel Representations for Translation and Effective Cross-lingual Transfer
Elizabeth Salesky, Neha Verma, Philipp Koehn, Matt Post
Exploring Representational Disparities Between Multilingual and Bilingual Translation Models
Neha Verma, Kenton Murray, Kevin Duh
Beyond Shared Vocabulary: Increasing Representational Word Similarities across Languages for Multilingual Machine Translation
Di Wu, Christof Monz
When Does Monolingual Data Help Multilingual Translation: The Role of Domain and Model Scale
Christos Baziotis, Biao Zhang, Alexandra Birch, Barry Haddow
Target-Agnostic Gender-Aware Contrastive Learning for Mitigating Bias in Multilingual Machine Translation
Minwoo Lee, Hyukhun Koh, Kang-il Lee, Dongdong Zhang, Minsung Kim, Kyomin Jung
Condensing Multilingual Knowledge with Lightweight Language-Specific Modules
Haoran Xu, Weiting Tan, Shuyue Stella Li, Yunmo Chen, Benjamin Van Durme, Philipp Koehn, Kenton Murray