The field of large language models (LLMs) is witnessing significant developments, particularly in the area of multilingual capabilities. Researchers are exploring innovative approaches to improve the performance of LLMs in multiple languages, addressing the challenges of context-aware and creative translation, safety moderation, and evaluation. Notably, the use of large language models is being extended to support literary translators, reducing editing time while maintaining creativity. Moreover, multilingual safety moderation tools are being developed to bridge the gaps in existing moderation capabilities. The development of open multilingual LLM judges and online document-level context incorporation agents is also advancing the field.

Some noteworthy papers include: PolyGuard, which introduces a state-of-the-art multilingual safety model for safeguarding LLM generations. M-Prometheus, which provides a suite of open multilingual LLM judges that can evaluate multilingual outputs and improve generated outputs. DoCIA, which enhances speech translation performance by incorporating document-level context. Llama-3-Nanda-10B-Chat, which is a state-of-the-art Hindi-centric instruction-tuned generative LLM that pushes the boundaries of open-source Hindi language models.