published the release of the machine translation system Opennmt 0.28.0 (Open Neural Machine Translation) using machine learning methods. To build a neural network, the project uses the capabilities of the Tensorflow Deep Machine Learning Library. The code developed by the Opennmt module project is written in Python and subjects under the MIT license. Ready-made models prepared for English, German and Catalan languages, for other languages you can independently form a data set from the project opus (two files are transmitted for teaching the system – one with sentences in the original language, and the second with a qualitative translation of these sentences into the target language).
project It develops with the participation of the company systran , specializing in creating machine translation funds, and a group of researchers harvard , developing a model of the human language for machine learning systems. The user interface is simplified as much as possible and requires only specifying the input file with the text and the file to save the translation result. The expansion system makes it possible to implement additional functionality on the basis of OpennMT, for example, abstracting, classification of texts and generation of subtitles.
The use of TensorFlow allows you to use the GPU capabilities (to speed up the process of teaching a neural network. To simplify the distribution of the product, the design also develops a self -sufficient version of the translator in C ++ – ctranslate2 , which uses previously trained models without reference to additional dependencies.
In the new version, the Initial_learning_rate parameter has been added and several new arguments were implemented (mha_bias and output_layer_bias) to configure the Transformer models generator. The rest is noted by the correction of errors.