English

On Long-Tailed Phenomena in Neural Machine Translation

Computation and Language 2020-10-13 v1 Artificial Intelligence Machine Learning

Abstract

State-of-the-art Neural Machine Translation (NMT) models struggle with generating low-frequency tokens, tackling which remains a major challenge. The analysis of long-tailed phenomena in the context of structured prediction tasks is further hindered by the added complexities of search during inference. In this work, we quantitatively characterize such long-tailed phenomena at two levels of abstraction, namely, token classification and sequence generation. We propose a new loss function, the Anti-Focal loss, to better adapt model training to the structural dependencies of conditional text generation by incorporating the inductive biases of beam search in the training process. We show the efficacy of the proposed technique on a number of Machine Translation (MT) datasets, demonstrating that it leads to significant gains over cross-entropy across different language pairs, especially on the generation of low-frequency words. We have released the code to reproduce our results.

Keywords

Cite

@article{arxiv.2010.04924,
  title  = {On Long-Tailed Phenomena in Neural Machine Translation},
  author = {Vikas Raunak and Siddharth Dalmia and Vivek Gupta and Florian Metze},
  journal= {arXiv preprint arXiv:2010.04924},
  year   = {2020}
}

Comments

Accepted to Findings of EMNLP 2020

R2 v1 2026-06-23T19:13:50.904Z