Rethinking Human-like Translation Strategy: Integrating Drift-Diffusion Model with Large Language Models for Machine Translation

2024-02-16

arXiv.org

AI Summary

To address the lack of human-like decision-making mechanisms in machine translation (MT), this paper proposes Thinker-DDM, the first framework to incorporate the cognitive science–inspired drift-diffusion model (DDM) into MT. Thinker-DDM explicitly models translators’ dynamic speed–accuracy trade-offs under resource constraints. It integrates large language models (LLMs) with a learnable, adaptive threshold-based decision mechanism, thereby redefining the MT generation paradigm. Experiments on WMT22 and CommonMT demonstrate significant improvements over state-of-the-art baselines across both high- and low-resource translation settings. Further commonsense translation analysis confirms enhanced semantic consistency and robustness. This work pioneers the integration of cognitive modeling with LLM-based MT, establishing a novel paradigm for building more interpretable and human-aligned translation systems.

Technology Category

• Machine Learning: Large Multimodal Models (LMMs)
• Cognitive Modeling & Cognitive Systems: Simulating Human Behavior
• Natural Language Processing: Machine Translation, Multilinguality, Cross-Lingual NLP

Application Category

• Semantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactions
• Economics, Online Markets and Human Computation: Cost models of using LLMs in production systems
• User Modeling, Personalization and Recommendation: Large Language Models (LLM) for user modeling and recommendation

Abstract

Large language models (LLMs) have demonstrated promising potential in various downstream tasks, including machine translation. However, prior work on LLM-based machine translation has mainly focused on better utilizing training data, demonstrations, or pre-defined and universal knowledge to improve performance, with a lack of consideration of decision-making like human translators. In this paper, we incorporate Thinker with the Drift-Diffusion Model (Thinker-DDM) to address this issue. We then redefine the Drift-Diffusion process to emulate human translators' dynamic decision-making under constrained resources. We conduct extensive experiments under the high-resource, low-resource, and commonsense translation settings using the WMT22 and CommonMT datasets, in which Thinker-DDM outperforms baselines in the first two scenarios. We also perform additional analysis and evaluation on commonsense translation to illustrate the high effectiveness and efficacy of the proposed method.

Problem

• Modeling human-like deliberation in machine translation
• Improving translation under resource constraints
• Enhancing commonsense translation performance

Innovation

• Incorporates Thinker with Drift-Diffusion Model
• Redefines Drift-Diffusion for dynamic decision-making
• Outperforms baselines in resource-constrained scenarios