Multilingual Coreference Resolution via Cycle-Consistent Machine Translation

2026-06-03 · Source: Artificial Intelligence · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Natural Language Processing · Depth: Expert, quick

Summary

A novel pipeline addresses the challenge of coreference resolution in low-resource languages, a task crucial for applications like machine translation, question answering, and document summarization. This method leverages machine translation (MT) from English to a target low-resource language to generate or expand necessary training data. To ensure data quality, the pipeline back-translates the generated samples and evaluates their similarity to the original English samples using cosine similarity within a BERT model's latent space. These similarity scores are then integrated into the loss function, weighting training samples based on their MT cycle consistency. Experiments across four low-resource languages demonstrate significant performance gains in coreference resolution, even enabling accurate resolution in languages previously lacking any dedicated corpora.

Key takeaway

For NLP Engineers expanding coreference resolution to low-resource languages, this pipeline offers a robust method to overcome data scarcity. You should consider implementing cycle-consistent machine translation to generate high-quality training data, leveraging back-translation and BERT-based similarity scoring. This approach enables accurate coreference resolution even in languages where no prior corpora exist, significantly broadening your model's applicability.

Key insights

Cycle-consistent machine translation effectively generates training data for low-resource multilingual coreference resolution, improving performance where corpora are scarce.

Principles

• Cycle consistency validates MT data quality.
• Data generation bridges resource gaps.
• Latent space similarity quantifies translation accuracy.

Method

Translate English coreference data to a target language, back-translate, then assess similarity with original English via BERT's latent space cosine similarity. Integrate these scores into the loss function to weight training samples.

In practice

• Apply MT for low-resource NLP data.
• Use BERT latent space for text similarity.
• Weight training data by translation quality.

Topics

• Coreference Resolution
• Machine Translation
• Low-Resource Languages
• Natural Language Processing
• Data Augmentation
• BERT

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

Related on AIssential

• Multilingual Coreference Resolution via Cycle-Consistent Machine Translation — Coreference resolution in low-resource languages can be significantly improved by cycle-consistent machine translation data augmentation.
• Closing the Gap at CRAC 2026: Two-Stage Adaptation for LLM-Based Multilingual Coreference Resolution — Two-stage fine-tuning and optimized input formats enable LLMs to achieve state-of-the-art multilingual coreference resolution.
• Evaluating Cross-lingual Knowledge Consistency in Code-Mixed vis-a-vis Indian Languages using IndicKLAR — Code-mixing inputs significantly improves large language model knowledge recall in Indian languages, nearly matching English performance.
• A Community Roadmap for Including Kapampángan in Modern AI Research — Community-driven data creation is essential for including underrepresented languages in AI research and applications.
• Universal Conceptual Structure in Neural Translation: Probing NLLB-200's Multilingual Geometry — NLLB-200's internal geometry reflects language-universal conceptual structures, akin to human multilingual cognition.
• mmPISA-bench: Do LLMs Reason Equally Well Across 43 Languages? — LLMs demonstrate effective multilingual reasoning across 43 languages, with machine translations proving as effective as human ones.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Artificial Intelligence.