Improving the quality of machine translation using the reverse model
The article proposes a method to improve machine translation consistency and quality between different languages, especially for smaller language areas with limited data. The article presents a theoretical justification for the proposed approach and describes the experiments conducted to validate the method. The authors suggest training the translation model to generate translations that closely match the source text when translated back and forth between the target and source languages. The proposed method uses a reverse model derived from the likelihood of cyclic translations. The paper shows that this method improves both consistency and translation quality by providing the forward model with information from the reverse model. The article discusses the challenges associated with the proposed method, such as the need to simultaneously store gradients for direct and reverse models at each training iteration and high variance of gradients in the initial iterations of training. The authors present a solution to these challenges by fine-tuning the already pre-trained translation model, which saves computational resources and converges faster than training from scratch. The experiments were carried out in the Russian-to-Kazakh and English-to-Finnish language areas, and the results show that the proposed method improves the consistency of translations and the overall translation quality. This article presents an important contribution to the field of natural language text processing by proposing a method to improve machine translation quality and consistency. The theoretical justification for the proposed method and the detailed description of the experiments conducted make this article a valuable resource for researchers and practitioners in the field.
- Improve the performance of neural machine translation with co-training
- An optimized cognitive-assisted machine translation approach for natural language processing
- Statistical machine translation enhancements through linguistic levels: a survey
- Uyghur-to-Chinese neural machine translation based on incremental training
- Improving statistical machine translation for a resource-poor language using related resource-rich languages
- Collocation dictionary optimization using WordNet and k-nearest neighbor learning
- Synchronous bidirectional inference for neural sequence generation
- Machine transliteration survey
- Improving statistical machine translation for a resource-poor language using related resource-rich languages
- Unpaired cross-domain image-to-image translation method using conditional projection
- A comparison of different machine transliteration models
- scientific article; zbMATH DE number 1949711 (Why is no real title available?)
- scientific article; zbMATH DE number 1949715 (Why is no real title available?)
- From feature to paradigm: deep learning in machine translation
- Agreement on target-bidirectional recurrent neural networks for sequence-to-sequence learning
- Statistical machine translation enhancements through linguistic levels: a survey
- Dependency-based correlation guidance for neural machine translation
- Improve the performance of neural machine translation with co-training
- Uyghur-to-Chinese neural machine translation based on incremental training
- Machine translation algorithm of low-resource languages based on semi-supervised learning
- Exploitation of machine learning techniques in modelling phrase movements for machine translation
- Specialized pre-training of neural networks on synthetic data for improving paraphrase generation
This page was built for publication: Improving the quality of machine translation using the reverse model
Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q2691537)
