NSL-MT: Linguistically Informed Negative Samples for Efficient Machine Translation in Low-Resource Languages

2025-11-12

AI Summary

To address the scarcity of parallel corpora and frequent grammatical errors in low-resource neural machine translation (NMT), this paper proposes NSL-MT, a linguistics-informed negative sampling learning framework. During training, NSL-MT constructs a negative space by synthetically generating target-language sentences violating grammatical constraints and introduces a severity-weighted negative sample loss to explicitly penalize illegal syntactic structures. Crucially, NSL-MT requires no additional annotations or external linguistic tools and is plug-and-play compatible with standard NMT architectures. Experiments across multiple low-resource language pairs demonstrate that NSL-MT improves BLEU scores by 3–89% and enhances data efficiency fivefold: using only 1,000 parallel sentence pairs, it matches the performance of baseline models trained on 5,000 pairs. This yields substantially improved grammatical robustness and generalization under few-shot settings.

Technology Category

• Natural Language Processing: Machine Translation, Multilinguality, Cross-Lingual NLP
• Machine Learning: Neuro-Symbolic Learning
• Search and Optimization: Learning to Search

Application Category

• Search and Retrieval-Augmented AI: Multilingual and cross-lingual Web search
• Web Mining and Content Analysis: Mining multimedia, multimodal, multilingual, cross-lingual Web data
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs

Abstract

We introduce Negative Space Learning MT (NSL-MT), a training method that teaches models what not to generate by encoding linguistic constraints as severity-weighted penalties in the loss function. NSL-MT increases limited parallel data with synthetically generated violations of target language grammar, explicitly penalizing the model when it assigns high probability to these linguistically invalid outputs. We demonstrate that NSL-MT delivers improvements across all architectures: 3-12% BLEU gains for well-performing models and 56-89% gains for models lacking descent initial support. Furthermore, NSL-MT provides a 5x data efficiency multiplier -- training with 1,000 examples matches or exceeds normal training with 5,000 examples. Thus, NSL-MT provides a data-efficient alternative training method for settings where there is limited annotated parallel corporas.

Problem

• Improves machine translation efficiency for low-resource languages
• Uses linguistic constraints to penalize invalid grammatical outputs
• Enables effective training with limited parallel data through negative samples

Innovation

• Uses linguistic constraints as weighted penalties
• Generates synthetic grammar violations as negative samples
• Improves data efficiency fivefold for low-resource training