Adapting Neural Machine Translation with Parallel Synthetic Data

Chinea-Rios, Mara; Peris, Álvaro; Casacuberta, Francisco

doi:10.18653/v1/w17-4714

Cited by 47 publications

(23 citation statements)

References 22 publications

(22 reference statements)

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Note that the TER was extremely high in this task. This phenomenon, also observed by Chinea-Rios et al (2017) for the same task, is due to the translation of short sentences with a NMT system trained on long sentences from a different domain (Europarl). Therefore, the system generated hypotheses much longer than the references.…”

Section: Translation Post-editing With Online Learningsupporting

confidence: 55%

See 1 more Smart Citation

Online learning for effort reduction in interactive neural machine translation

Peris

Casacuberta

2019

Computer Speech & Language

Self Cite

View full text Add to dashboard Cite

Neural machine translation systems require large amounts of training data and resources. Even with this, the quality of the translations may be insufficient for some users or domains. In such cases, the output of the system must be revised by a human agent. This can be done in a post-editing stage or following an interactive machine translation protocol.We explore the incremental update of neural machine translation systems during the post-editing or interactive translation processes. Such modifications aim to incorporate the new knowledge, from the edited sentences, into the translation system. Updates to the model are performed on-the-fly, as sentences are corrected, via online learning techniques. In addition, we implement a novel interactive, adaptive system, able to react to single-character interactions. This system greatly reduces the human effort required for obtaining high-quality translations.In order to stress our proposals, we conduct exhaustive experiments varying the amount and type of data available for training. Results show that online learning effectively achieves the objective of reducing the human effort required during the post-editing or the interactive machine translation stages. Moreover, these adaptive systems also perform well in scenarios with scarce resources. We show that a neural machine translation system can be rapidly adapted to a specific domain, exclusively by means of online learning techniques.3. We conduct a large experimentation, using public corpora containing different features from varied domains. We stress the translation systems, applying them in three different translation scenarios. 4. We show that OL brings significant improvements to the translation engines, in terms of translation quality and human effort reduction. Comparisons with other works in the literature also show that our adaptive, interactive systems are able to outperform the existing state-of-the-art. 5. We open-source all code developed in this work, in order to make research reproducible.The rest of this manuscript is structured as follows: the related work is reviewed in Section 2. Next, Section 3 briefly introduces the NMT technology, while the interactive protocol for NMT is presented in Section 4. OL is described and applied together with INMT in Section 5. Section 6 describes the experimental setup of this work. Results are presented and discussed in Section 7. Finally, we conclude the work and trace future lines of work in Section 8. Related workThis work puts together three thoroughly studied fields: neural machine translation, interactive machine translation and online learning. In this section, we briefly review the progress made in the last years in each one of these fields. Neural machine translationAlthough the first attempts of performing machine translation with neural networks date from long ago (Castaño and Casacuberta, 1997; Forcada andÑeco, 1997), NMT only took off recently. Kalchbrenner and Blunsom (2013) reintroduced full neural MT, although the results were non-competitive with re...

show abstract

Section: Translation Post-editing With Online Learningsupporting

confidence: 55%

“…Therefore, in order to match the reference, TER must delete many words. This problem may be addressed via heuristics in the search method (as pointed out by Chinea-Rios et al (2017)), but this is out of the scope of this work.…”

Section: Translation Post-editing With Online Learningmentioning

confidence: 99%

Online learning for effort reduction in interactive neural machine translation

Peris

Casacuberta

2019

Computer Speech & Language

Self Cite

View full text Add to dashboard Cite

show abstract

“…Gains with forward translation are significant, as in (Chinea-Rios et al, 2017), albeit about half as good as with BT, and result in small improvements for the in-domain and for the out-of-domain tests. Experiments combining forward and backward translation (backfwdtrans-nmt), each English→French English→German Figure 1: Learning curves from backtrans-nmt and natural.…”

Section: Bt Quality Does Mattermentioning

confidence: 98%

Using Monolingual Data in Neural Machine Translation: a Systematic Study

Burlot¹,

Yvon²

2018

Proceedings of the Third Conference on Machine Translation: Research Papers

View full text Add to dashboard Cite

Neural Machine Translation (MT) has radically changed the way systems are developed. A major difference with the previous generation (Phrase-Based MT) is the way monolingual target data, which often abounds, is used in these two paradigms. While Phrase-Based MT can seamlessly integrate very large language models trained on billions of sentences, the best option for Neural MT developers seems to be the generation of artificial parallel data through back-translation -a technique that fails to fully take advantage of existing datasets. In this paper, we conduct a systematic study of back-translation, comparing alternative uses of monolingual data, as well as multiple data generation procedures. Our findings confirm that back-translation is very effective and give new explanations as to why this is the case. We also introduce new data simulation techniques that are almost as effective, yet much cheaper to implement.

show abstract

“…Similarly to this paper, the use of artificiallygenerated sentences to fine-tuned models has also been explored by Chinea-Rios et al (2017) where they select monolingual authentic sentences in the source-side and translate them into the target language, or the work of Poncelas et al (2019a) where they use back-translated sentences only to adapt the models.…”

Section: Use Of Artificially-generated Data Tomentioning

confidence: 99%

Selecting Artificially-Generated Sentences for Fine-Tuning Neural Machine Translation

Poncelas¹,

Way²

2019

Proceedings of the 12th International Conference on Natural Language Generation

View full text Add to dashboard Cite

Neural Machine Translation (NMT) models tend to achieve best performance when larger sets of parallel sentences are provided for training. For this reason, augmenting the training set with artificially-generated sentence pairs can boost performance.Nonetheless, the performance can also be improved with a small number of sentences if they are in the same domain as the test set. Accordingly, we want to explore the use of artificially-generated sentences along with data-selection algorithms to improve Germanto-English NMT models trained solely with authentic data.In this work, we show how artificiallygenerated sentences can be more beneficial than authentic pairs, and demonstrate their advantages when used in combination with dataselection algorithms.

show abstract

Adapting Neural Machine Translation with Parallel Synthetic Data

Cited by 47 publications

References 22 publications

Online learning for effort reduction in interactive neural machine translation

Online learning for effort reduction in interactive neural machine translation

Using Monolingual Data in Neural Machine Translation: a Systematic Study

Selecting Artificially-Generated Sentences for Fine-Tuning Neural Machine Translation

Contact Info

Product

Resources

About