Yandex.Translate

from Wikipedia, the free encyclopedia
Globe icon of the infobox
Yandex.Translate
Machine translation
languages 94
operator Yandex
Registration Yandex account (optional)
On-line March 2011 (currently online)
https://translate.yandex.com/

Yandex.Translate (also Yandex.Perevod ) is an online service from Yandex that can machine translate words, texts and entire websites .

history

The service was started in 2011 and initially only covered English, Ukrainian and Russian. 94 languages ​​are now offered. Yandex emphasizes that it also wants to cover rare languages ​​- including Bashkir , Papiamentu and Sindarin , among others . In 2017, a corpus of one million sentences was specified for the English-Russian language combination, which corresponded to around half of the corresponding inventory in the TraMOOC Project .

Mobile apps including the offline function have been available for Android and iOS since 2012 .

functionality

Yandex's machine translation was initially based on statistical methods. In September 2017, Yandex announced that it would introduce a hybrid system for its machine translation that combines statistical and neural functions .

The service consists of three main internal components for each language: translation model, language model and decoder.

The translation model was created through the selection of parallel documents and the subsequent selection of sentence pairs and word pairs or word combination pairs, i.e. it is a table with all words and phrases in the system and their possible translation variants into the other language, including the probability of these variants. So not only individual words are compared, but also combinations of several consecutive words. The translation model comprises one hundred million word and word combination pairs for each language pair.

For the language model, the system examines hundreds of thousands of different texts in the desired language and creates a list of all the words and word combinations used in them, including their frequency. This is the system knowledge of the language to be translated into.

The decoder acts as a translator itself. He selects all translation variants for each sentence of the entered text by combining the phrases from the translation model and sorting them according to descending probability.

The decoder estimates all variants of the output combinations using the language model. He thus selects the sentence with the best combination of probability (translation model) and frequency of use (language model).

The service can try to automatically recognize the source language of a text. But he is not able to include neighboring sentences as context.

Yandex.Translate offers a programming interface (REST API) to incorporate translations into your own websites and offers. There is an easy to use implementation for Python .

Reception and evaluation

In a small practical trial in May 2014, Yandex.Translate only played a very marginal role among translation students at the University of Zielona Góra compared to numerous other providers. In a comparison with Google Translate, Yandex.Translate 2015 achieved better results in the combination of Russian and Croatian .

Yandex.Translate, along with Google Translate, was named one of the best machine translation providers available in 2018 , but it produced less fluid translations than its competitor. Together with Google Translate, it also offers the largest range of Slavic languages , but ultimately has to admit defeat to its competitor.

Web links

Individual evidence

  1. ↑ a b E.I. Gimazitdinov, DA Morel Morel: Machine Translation Technologies used in Online Translation Industry . In: dspace.bsu.edu.ru . On-line
  2. ^ Yandex History - 2011. In: yandex.com. Retrieved May 16, 2019 .
  3. ^ List of supported languages ​​- Yandex.Translate. Help. In: yandex.com. Retrieved May 16, 2019 .
  4. ↑ What Google can't: Russian search engine saves dying languages. In: de.sputniknews.com. April 19, 2017. Retrieved May 17, 2019 .
  5. ↑ Sheila Castilho, Joss Moorkens, Federico Gaspari, Rico Senn Rich, Andy Way. Panayota Georgakopoulou: Evaluating MT for massive open online courses. A multifaceted comparison between PBSMT and NMT systems . In: Machine Translation . tape 32 , 2018, p. 255-278 , doi : 10.1007 / s10590-018-9221-y .
  6. ^ Yandex History - 2012. In: yandex.com. Retrieved May 16, 2019 .
  7. ↑ One model is better than two. Yandex.Translate launches a hybrid machine translation system. In: yandex.com. September 14, 2017, accessed on May 16, 2019 .
  8. ^ Yandex - Technologies - Machine Translation. Retrieved February 15, 2019 .
  9. ↑ TA Tohmetov, AO Ushakov, IS Vanushin: The Problems of Machine Translation . In: earchive.tpu.ru . On-line
  10. ↑ a b Nikolay Arefyev, Pavel Ermolaev, Alexander Panchenko: How much does a word weigh? Weighting word embeddings for word sense induction . In: Proceedings of the 24th International Conference on Computational Linguistics and Intellectual Technologies . May 23, 2018, arxiv : 1805.09209 [abs] .
  11. ↑ Maarten van Hees, Paulina Kozłowska, Nana Tian: Web-based automatic translation: the Yandex.Translate API . In: mediatechnology.leiden.edu . Online .
  12. ↑ James Axl: yandex-translater: Python API for Yandex Translate. In: PyPI . Retrieved May 19, 2019 .
  13. ↑ Agnieszka Kałużna: Machine translation tools in the students' translation training . In: Łukasz Grabowski, Tadeusz Piotrowski (eds.): The Translator and the Computer 2. Proceedings of a Conference held in Wrocław, October 25-26, 2014 . Wrocław 2015, p. 39-50 . ( Online )
  14. ↑ S. Seljan, I. Dunđer: Machine Translation and Automatic Evaluation of English / Russian-Croatian . In: Proceedings of Corpus Linguistics . 2015. Online .
  15. ↑ Lana Soglasnova: Dealing with False Friends to Avoid Errors in Subject Analysis in Slavic Cataloging: An Overview of Resources and Strategies . In: Cataloging & Classification Quarterly . tape 56 , no. 5-6 , 2018, pp. 404-421 , doi : 10.1080 / 01639374.2018.1438551 .