Quantifying the Plausibility of Context Reliance in Neural Machine Translation

TL;DR

This paper introduces PECoRe, a framework for evaluating how plausibly neural machine translation models rely on context, using interpretability techniques to compare model rationales with human annotations and identify context-driven predictions.

Contribution

The paper presents PECoRe, a novel interpretability framework that quantifies and analyzes context reliance in language models, addressing limitations of artificial benchmarks.

Findings

PECoRe effectively identifies context-sensitive tokens in translations.

The framework reveals instances of plausible and implausible context usage.

Comparison with human annotations validates the interpretability approach.

Abstract

Establishing whether language models can use contextual information in a human-plausible way is important to ensure their trustworthiness in real-world settings. However, the questions of when and which parts of the context affect model generations are typically tackled separately, with current plausibility evaluations being practically limited to a handful of artificial benchmarks. To address this, we introduce Plausibility Evaluation of Context Reliance (PECoRe), an end-to-end interpretability framework designed to quantify context usage in language models' generations. Our approach leverages model internals to (i) contrastively identify context-sensitive target tokens in generated texts and (ii) link them to contextual cues justifying their prediction. We use \pecore to quantify the plausibility of context-aware machine translation models, comparing model rationales with human annotations across several discourse-level phenomena. Finally, we apply our method to unannotated model translations to identify context-mediated predictions and highlight instances of (im)plausible context usage throughout generation.