Log-linear Combinations of Monolingual and Bilingual Neural Machine Translation Models for Automatic Post-Editing

By Marcin Junczys-Dowmunt, Roman Grundkiewicz. 2016

110 citations recorded by the source corpus.

This paper describes the submission of the AMU (Adam Mickiewicz University) team to the Automatic Post-Editing (APE) task of WMT 2016. We explore the application of neural translation models to the APE problem and achieve good results by treating different models as components in a log-linear model, allowing for multiple inputs (the MT-output and the source) that are decoded to the same target language (post-edited translations). A simple string-matching penalty integrated within the log-linear model is used to control for higher faithfulness with regard to the raw machine translation output. To overcome the problem of too little training data, we generate large amounts of artificial data. Our submission improves over the uncorrected baseline on the unseen test set by -3.2\% TER and +5.5\% BLEU and outperforms any other system submitted to the shared-task by a large margin.

Subfields: Artificial Intelligence, Natural Language Processing, Machine Translation

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

In Natural Language Processing

In Machine Translation

Citations in the corpus

Findings of the 2016 Conference on Machine Translation (2016, background): “AMU Adam Mickiewicz University, Poland (Junczys-Dowmunt and Grundkiewicz, 2016) CUNI Univerzita Karlova v Praze, Czech Republic (Libovický et al.”
Findings of the 2017 Conference on Machine Translation (WMT17) (2017, methodology): “The data released for the 2016 round of the task (15,000 instances) and the artificially generated post-editing triplets (4 million instances) used by last year’s winning system (Junczys-Dowmunt and Grundkiewicz, 2016) were also provided as additional training material.”
Can Automatic Post-Editing Improve NMT? (2020, background): “…Correia and Martins, 2019) even when trained with limited number of post-edited training instances (generally “triplets” consisting of source, translated, and post-edited segments), with or without additional large-scale artificial data (Junczys-Dowmunt and Grundkiewicz, 2016; Negri et al., 2018).”
Findings of the 2021 Conference on Machine Translation (WMT21) (2021, background): “5 million artificially-generated post-editing triplets described in (Junczys-Dowmunt and Grundkiewicz, 2016), and ii) the 14.”
Back-Translation-Style Data Augmentation for end-to-end ASR (2018, unknown): “Similar techniques have also led to performance improvements in related tasks such as automatic post edition [17].”
A Shared Attention Mechanism for Interpretation of Neural Automatic Post-Editing Systems (2018, methodology): “Since the training set provided by the shared task is too small to effectively train neural networks, (Junczys-Dowmunt and Grundkiewicz, 2016) have proposed a method for creating extra, “artiﬁcial” training data using round-trip translations.”
APE at Scale and Its Implications on MT Evaluation Biases (2019, methodology): “Probably most similar to our work, Junczys-Dowmunt and Grundkiewicz (2016, 2018) uses RTT as additional training data for the automatic post-editing (APE) task of the WMT evaluation campaign (Chatterjee et al., 2018).”
FELIX: Flexible Text Editing Through Tagging and Insertion (2020, methodology): “We follow the procedures introduced in (Junczys-Dowmunt and Grundkiewicz, 2016) and train our models using two synthetic corpora of 4M and 500K examples merged with a corpus of 11K real examples oversampled 10 times.”
Pre-Translation for Neural Machine Translation (2016, background): “Recently, it was shown that models based on neural MT are very successful in this task (Junczys-Dowmunt and Grundkiewicz, 2016).”
Findings of the WMT 2020 Shared Task on Automatic Post-Editing (2020, background): “5 million artificially-generated post-editing triplets described in (Junczys-Dowmunt and Grundkiewicz, 2016).”
An Exploration of Neural Sequence-to-Sequence Architectures for Automatic Post-Editing (2017, background): “, 2016) and compare our performance against the shared task winner, the system submitted by the Adam Mickiewicz University (AMU) team (Junczys-Dowmunt and Grundkiewicz, 2016), and a more recent system by Pal et al.”
Explaining and Generalizing Back-Translation through Wake-Sleep (2018, background): “…back-translation in state-of-the-art (e.g. Hassan et al. (2018)) and unsupervised neural machine translation (Artetxe et al., 2018) dates back to large empirical gains found by Sennrich et al. (2016a) in the context of neural MT and automatic post-editing (Junczys-Dowmunt and Grundkiewicz, 2016).”
Ensembling Factored Neural Machine Translation Models for Automatic Post-Editing and Quality Estimation (2017, background): “A relatively recent line of research has tried to build models which correct errors in MT automatically (Simard et al., 2007; Bojar et al., 2015; Junczys-Dowmunt and Grundkiewicz, 2016).”