A Representation Level Analysis of NMT Model Robustness to Grammatical Errors

2025-05-27

AI Summary

This study investigates the robustness mechanisms of neural machine translation (NMT) models against grammatical errors at the representation level. To this end, we introduce the novel concept of *robustness-oriented attention heads* and systematically characterize how encoder representations evolve across layers when processing ungrammatical inputs. Our methodology integrates grammatical error detection (GED) probing, representational similarity analysis (RSA), attention pattern diagnostics, and cross-layer representation tracking. We identify a “detect-then-correct” paradigm in encoder behavior: early layers detect errors, while deeper layers progressively refine representations. Crucially, we localize and validate linguistically interpretable robustness-oriented attention heads—these attend to morphological features and syntactic boundaries, and their reliance increases significantly after robust fine-tuning. This work provides the first representation-based, interpretable evidence and analytical framework for robust NMT, bridging linguistic insight with internal model dynamics.

Technology Category

• Natural Language Processing: Safety and Robustness
• Machine Learning: Adversarial Learning & Robustness
• Computer Vision: Adversarial Attacks & Robustness

Application Category

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

Abstract

Understanding robustness is essential for building reliable NLP systems. Unfortunately, in the context of machine translation, previous work mainly focused on documenting robustness failures or improving robustness. In contrast, we study robustness from a model representation perspective by looking at internal model representations of ungrammatical inputs and how they evolve through model layers. For this purpose, we perform Grammatical Error Detection (GED) probing and representational similarity analysis. Our findings indicate that the encoder first detects the grammatical error, then corrects it by moving its representation toward the correct form. To understand what contributes to this process, we turn to the attention mechanism where we identify what we term Robustness Heads. We find that Robustness Heads attend to interpretable linguistic units when responding to grammatical errors, and that when we fine-tune models for robustness, they tend to rely more on Robustness Heads for updating the ungrammatical word representation.

Problem

• Analyzing NMT model robustness to grammatical errors
• Studying internal model representations of ungrammatical inputs
• Identifying Robustness Heads in attention mechanism

Innovation

• Analyzing NMT robustness via internal representations
• Using GED probing and similarity analysis
• Identifying Robustness Heads in attention mechanism