Digging Errors in NMT: Evaluating and Understanding Model Errors from Partial Hypothesis Space

TL;DR

This paper introduces a novel evaluation protocol for neural machine translation (NMT) that assesses model errors based on ranking hypotheses within the entire hypothesis space, revealing significant ranking issues in state-of-the-art models.

Contribution

It proposes a new evaluation framework and approximation methods to analyze NMT model errors beyond traditional metrics, highlighting limitations of current models and search algorithms.

Findings

Transformer models perform at chance level in top hypothesis ranking.

Model errors are strongly correlated with human judgment.

Search algorithms like beam search exhibit inductive biases affecting translation quality.

Abstract

Solid evaluation of neural machine translation (NMT) is key to its understanding and improvement. Current evaluation of an NMT system is usually built upon a heuristic decoding algorithm (e.g., beam search) and an evaluation metric assessing similarity between the translation and golden reference. However, this system-level evaluation framework is limited by evaluating only one best hypothesis and search errors brought by heuristic decoding algorithms. To better understand NMT models, we propose a novel evaluation protocol, which defines model errors with model's ranking capability over hypothesis space. To tackle the problem of exponentially large space, we propose two approximation methods, top region evaluation along with an exact top-$k$ decoding algorithm, which finds top-ranked hypotheses in the whole hypothesis space, and Monte Carlo sampling evaluation, which simulates hypothesis space from a broader perspective. To quantify errors, we define our NMT model errors by measuring distance between the hypothesis array ranked by the model and the ideally ranked hypothesis array. After confirming the strong correlation with human judgment, we apply our evaluation to various NMT benchmarks and model architectures. We show that the state-of-the-art Transformer models face serious ranking issues and only perform at the random chance level in the top region. We further analyze model errors on architectures with different depths and widths, as well as different data-augmentation techniques, showing how these factors affect model errors. Finally, we connect model errors with the search algorithms and provide interesting findings of beam search inductive bias and correlation with Minimum Bayes Risk (MBR) decoding.