Quality estimation with transformer and RNN architectures

Yulin Zhang, Chong Feng*, Hongzheng Li

*Corresponding author for this work

Research output: Chapter in Book/Report/Conference proceedingConference contributionpeer-review

Abstract

The goal of China Conference on Machine Translation (CCMT 2019) Shared Task on Quality Estimation (QE) is to investigate automatic methods for estimating the quality of Chinese↔English machine translation results without reference translations. This paper presents the submissions of our team for the sentence-level Quality Estimation shared task of CCMT19. Considering the good performance of neural models in previous shared tasks of WMT, our submissions also include two neural-based models: one is Bi-Transformer which proposes the model as a feature extractor with a bidirectional transformer and then processes the semantic representations of source and the translation output with a Bi-LSTM predictive model for automatic quality estimation, and the other BiRNN architecture uses only two bi-directional RNNs (bi-RNN) with Gated Recurrent Units (GRUs) as encoders, and learns representation of the source and translation sentence pairs to predict the quality of translation outputs.

Original languageEnglish
Title of host publicationMachine Translation - 15th China Conference, CCMT 2019, Revised Selected Papers
EditorsShujian Huang, Kevin Knight
PublisherSpringer
Pages69-76
Number of pages8
ISBN (Print)9789811517204
DOIs
Publication statusPublished - 2019
Event15th China Conference on Machine Translation, CCMT 2019 - Nanchang, China
Duration: 27 Sept 201929 Sept 2019

Publication series

NameCommunications in Computer and Information Science
Volume1104 CCIS
ISSN (Print)1865-0929
ISSN (Electronic)1865-0937

Conference

Conference15th China Conference on Machine Translation, CCMT 2019
Country/TerritoryChina
CityNanchang
Period27/09/1929/09/19

Keywords

  • Quality Estimation
  • Transformer
  • Translation evaluation

Fingerprint

Dive into the research topics of 'Quality estimation with transformer and RNN architectures'. Together they form a unique fingerprint.

Cite this