Automatically Extracting Challenge Sets for Non-Local Phenomena in Neural Machine Translation

Leshem Choshen, Omri Abend


Abstract
We show that the state-of-the-art Transformer MT model is not biased towards monotonic reordering (unlike previous recurrent neural network models), but that nevertheless, long-distance dependencies remain a challenge for the model. Since most dependencies are short-distance, common evaluation metrics will be little influenced by how well systems perform on them. We therefore propose an automatic approach for extracting challenge sets rich with long-distance dependencies, and argue that evaluation using this methodology provides a complementary perspective on system performance. To support our claim, we compile challenge sets for English-German and German-English, which are much larger than any previously released challenge set for MT. The extracted sets are large enough to allow reliable automatic evaluation, which makes the proposed approach a scalable and practical solution for evaluating MT performance on the long-tail of syntactic phenomena.
Anthology ID:
K19-1028
Volume:
Proceedings of the 23rd Conference on Computational Natural Language Learning (CoNLL)
Month:
November
Year:
2019
Address:
Hong Kong, China
Venue:
CoNLL
SIG:
SIGNLL
Publisher:
Association for Computational Linguistics
Note:
Pages:
291–303
Language:
URL:
https://www.aclweb.org/anthology/K19-1028
DOI:
10.18653/v1/K19-1028
Bib Export formats:
BibTeX MODS XML EndNote
PDF:
http://aclanthology.lst.uni-saarland.de/K19-1028.pdf
Supplementary material:
 K19-1028.Supplementary_Material.pdf