Improving Machine Translation of Idioms: A Spanish–Galician Parallel Dataset and Synthetic Augmentation Approach

2026-04-12 · Source: Paper Index on ACL Anthology · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Data Science & Analytics · Depth: Expert, quick

Summary

A new systematic approach addresses the challenge of translating idiomatic expressions between Spanish and Galician using neural machine translation. Researchers first constructed a high-quality parallel dataset of idioms, manually aligning expressions across both languages. This initial dataset was then significantly expanded into a large synthetic parallel corpus through the use of Large Language Models (LLMs). The augmentation strategy prioritized the most frequently observed idioms in authentic corpora. The expanded dataset was subsequently used to retrain a sequence-to-sequence (seq2seq) translation model. Evaluation demonstrated a significant improvement in idiom translation, along with a slight enhancement in the model's overall performance, outperforming baseline models and state-of-the-art LLM-based translators like SalamandraTA.

Key takeaway

For research scientists developing machine translation systems for low-resource or idiom-rich language pairs, consider implementing a two-stage data augmentation strategy. Manually curate a small, high-quality idiom dataset, then use LLMs to generate a larger synthetic corpus, focusing on high-frequency idioms. This approach can significantly enhance idiom translation accuracy and overall model performance, as demonstrated for Spanish-Galician.

Key insights

Synthetic data augmentation using LLMs significantly improves idiom translation in neural machine translation.

Principles

• Manual alignment ensures high-quality idiom datasets.
• Prioritize frequent idioms for synthetic augmentation.

Method

Build a manually aligned idiom dataset, then use LLMs to synthetically augment it, prioritizing frequent idioms, and retrain a seq2seq model.

In practice

• Create high-quality parallel idiom datasets.
• Augment data with LLMs for low-resource language pairs.

Topics

• Idiom Translation
• Neural Machine Translation
• Spanish-Galician Translation
• Large Language Models
• Data Augmentation

Best for: Research Scientist, AI Scientist, NLP Engineer, Machine Learning Engineer

Related on AIssential

• A Dataset for Probing Translationese Preferences in English-to-Swedish Translation — Language models often prefer literal "translationese" over idiomatic phrasing, especially when exposed to the source language.
• G-IdiomAlign: A Gloss-Pivoted Benchmark for Cross-Lingual Idiom Alignment — Cross-lingual idiom alignment in LLMs benefits from explicit semantic glosses, yet literal translation bias remains a significant challenge.
• Data Augmentation for Historical NER: A Systematic Comparison of Lexical and LLM-based Approaches — This paper systematically compares lexical and LLM-based data augmentation for Historical Named Entity Recognition.
• Ancient Greek to Modern Greek Machine Translation: A Novel Benchmark and Fine-Tuning Experiments on LLMs and NMT Models — A new AG-MG parallel corpus and benchmark significantly advance Ancient Greek to Modern Greek machine translation.
• Datasets for Verb Alternations across Languages: BLM Templates and Data Augmentation Strategies — This work addresses the underexplored ability of large language models (LLMs) to capture cross-sentence paradigmatic patterns, specifically verb alternations.
• Multilingual Idioms in Sentences and Conversations Across High-, Medium-, and Low-Resource Languages — Idiomatic expressions challenge multilingual NLP, with literal meanings and low-resource languages posing significant hurdles.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Paper Index on ACL Anthology.