ExPosST: Explicit Positioning with Adaptive Masking for LLM-Based Simultaneous Machine Translation

TL;DR

ExPosST introduces an explicit position allocation framework for LLM-based simultaneous translation, improving efficiency and consistency across various positional encodings and policies.

Contribution

The paper proposes a novel explicit position allocation method and a policy-consistent fine-tuning strategy to enhance LLM-based simultaneous translation.

Findings

Supports diverse translation policies effectively.

Enables efficient decoding with fixed positional slots.

Bridges the gap between training and inference behaviors.

Abstract

Large language models (LLMs) have recently demonstrated promising performance in simultaneous machine translation (SimulMT). However, applying decoder-only LLMs to SimulMT introduces a positional mismatch, which leads to a dilemma between decoding efficiency and positional consistency. Existing approaches often rely on specific positional encodings or carefully designed prompting schemes, and thus fail to simultaneously achieve inference efficiency, positional consistency, and broad model compatibility. In this work, we propose ExPosST, a general framework that resolves this dilemma through explicit position allocation. ExPosST reserves fixed positional slots for incoming source tokens, enabling efficient decoding with KV cache across different positional encoding methods. To further bridge the gap between fine-tuning and inference, we introduce a policy-consistent fine-tuning strategy that aligns training with inference-time decoding behavior. Experiments across multiple language pairs demonstrate that ExPosST effectively supports simultaneous translation under diverse policies.