Simul-LLM: A Framework for Exploring High-Quality Simultaneous Translation with Large Language Models

TL;DR

This paper introduces Simul-LLM, an open-source framework for fine-tuning and evaluating large language models specifically for simultaneous translation, addressing unique challenges and validating classical methods in this context.

Contribution

It presents the first dedicated framework for fine-tuning and assessing LLMs for simultaneous translation, bridging a gap in current research and practice.

Findings

Validated classical SimulMT concepts with LLMs

Developed an open-source fine-tuning pipeline

Explored adaptation of NMT-tuned LLMs for SimulMT

Abstract

Large language models (LLMs) with billions of parameters and pretrained on massive amounts of data are now capable of near or better than state-of-the-art performance in a variety of downstream natural language processing tasks. Neural machine translation (NMT) is one such task that LLMs have been applied to with great success. However, little research has focused on applying LLMs to the more difficult subset of NMT called simultaneous translation (SimulMT), where translation begins before the entire source context is available to the model. In this paper, we address key challenges facing LLMs fine-tuned for SimulMT, validate classical SimulMT concepts and practices in the context of LLMs, explore adapting LLMs that are fine-tuned for NMT to the task of SimulMT, and introduce Simul-LLM, the first open-source fine-tuning and evaluation pipeline development framework for LLMs focused on SimulMT.