Translate Meanings, Not Just Words: IdiomKB's Role in Optimizing Idiomatic Translation with Language Models

TL;DR

This paper introduces IdiomKB, a multilingual idiom knowledge base built with large language models, to improve idiomatic translation in machine translation systems by enhancing context understanding and scalability.

Contribution

The paper presents a scalable, context-aware multilingual idiom KB created with large LMs, improving idiomatic translation for smaller models and introducing a GPT-4-based human-aligned evaluation metric.

Findings

IdiomKB significantly improves translation quality of idiomatic expressions.

Smaller models with IdiomKB outperform baseline models without it.

Human evaluations confirm the high quality of the idiom knowledge base.

Abstract

To translate well, machine translation (MT) systems and general-purposed language models (LMs) need a deep understanding of both source and target languages and cultures. Therefore, idioms, with their non-compositional nature, pose particular challenges for Transformer-based systems, as literal translations often miss the intended meaning. Traditional methods, which replace idioms using existing knowledge bases (KBs), often lack scale and context awareness. Addressing these challenges, our approach prioritizes context awareness and scalability, allowing for offline storage of idioms in a manageable KB size. This ensures efficient serving with smaller models and provides a more comprehensive understanding of idiomatic expressions. We introduce a multilingual idiom KB (IdiomKB) developed using large LMs to address this. This KB facilitates better translation by smaller models, such as BLOOMZ (7.1B), Alpaca (7B), and InstructGPT (6.7B), by retrieving idioms' figurative meanings. We present a novel, GPT-4-powered metric for human-aligned evaluation, demonstrating that IdiomKB considerably boosts model performance. Human evaluations further validate our KB's quality.