XII Международная студенческая научная конференция Студенческий научный форум - 2020

Neural machine translation includes a number of rapidly developing technologies that are widely used, in particular, in communications and business. Among them, there are three main ways of using machine translation: rule-based machine translation, phrase-based machine translation and neural machine translation.

Rule-based machine translation is the oldest approach when several simple levels of analysis are applied:

Targeting parts of speech;

Morphological analysis;

Semantic analysis: some words are assigned a semantic category;

Compound analysis: some words are grouped, for example, into pairs of nouns and corresponding adjectives;

Analysis of dependencies: the object and subject of the main verb are identified.

Phrase-based machine translation is the simplest and most popular version of statistical machine translation. Today it is still used in the majority of online translation services. The technology can generate several different translation options and the essence of the approach is to choose the best option. It makes use of a phrase table that gives translation options and the probability of their use in this sequence in the source language. Then it applies a reordering table that indicates how words can be rearranged when translating from source to target language. Next, a language model that shows the probability for each possible sequence of words in the target language is implemented. Despite the lack of accuracy this method is still the most popular due to the combination of efficiency and simplicity.

Neural machine translation is the most modern and efficient type of machine translation. However, the training and use of neural networks is an ambiguous task that requires careful planning and has a significant cost in terms of computing power.

A bidirectional recurrent neural network (RNN) also known as an encoder is used by a neural network to encode the source sentence for a second recurrent network also known as a decoder which is used to predict words in a target language.

The process itself is divided into two phases. Firstly each word of the original sentence passes through encoder, which generates the original context, relying on the current word and the previous context. Secondly a complete sequence is formed taking into account the initial context, after which target words are generated using target context formed in conjunction with the previous word and providing some information about the status of the translation process.

NMT models require only a small part of memory in comparison with traditional systems of statistical machine translation (SMT). In addition unlike traditional translation systems all parts of the neural translation model are trained together (from start to finish) to maximize translation efficiency.

Despite the fact that neural networks are currently rarely used in machine translation technologies, the results demonstrated by them are superior to the results of using other technologies, and therefore it can be concluded that over time neural machine translation will supersede other word processing methods.

References

1.Sutskever, Ilya; Vinyals, Oriol; Le, Quoc Viet. Sequence to sequence learning with neural networks (неопр.) // NIPS. — 2014.

2. Kyunghyun Cho; Bart van Merrienboer; Dzmitry Bahdanau & Yoshua Bengio (3 September 2014), "On the Properties of Neural Machine Translation: Encoder–Decoder Approaches", arΧiv:1409.1259.

3. Yonghui Wu, Mike Schuster, Zhifeng Chen, Quoc V. Le, Mohammad Norouzi, Wolfgang Macherey, Maxim Krikun, Yuan Cao, Qin Gao, Klaus Macherey, Jeff Klingner, Apurva Shah, Melvin Johnson, Xiaobing Liu, Łukasz Kaiser, Stephan Gouws, Yoshikiyo Kato, Taku Kudo, Hideto Kazawa, Keith Stevens, George Kurian, Nishant Patil, Wei Wang, Cliff Young, Jason Smith, Jason Riesa, Alex Rudnick, Oriol Vinyals, Greg Corrado, Macduff Hughes, Jeffrey Dean. (26 September 2016), “Google's Neural Machine Translation System: Bridging the Gap between Human and Machine Translation”, https://arxiv.org/abs/1609.08144.