Google Neural Machine Translation

From Wikipedia, the free encyclopedia
Jump to navigation Jump to search

Google Neural Machine Translation (GNMT) is a neural machine translation (NMT) system developed by Google and introduced in November 2016, that uses an artificial neural network to increase fluency and accuracy in Google Translate.[1][2][3][4]

GNMT improves on the quality of translation by applying an example based (EBMT) machine translation method in which the system "learns from millions of examples".[2] GNMT's proposed architecture of system learning was first tested on over a hundred languages supported by Google Translate.[2] With the large end-to-end framework, the system learns over time to create better, more natural translations.[1] GNMT is capable of translating whole sentences at a time, rather than just piece by piece.[1] The GNMT network can undertake interlingual machine translation by encoding the semantics of the sentence, rather than by memorizing phrase-to-phrase translations.[2][5]


The Google Brain project was established in 2011 in the "secretive Google X research lab"[6] by Google Fellow Jeff Dean, Google Researcher Greg Corrado, and Stanford University Computer Science professor Andrew Ng.[7][8][9] Ng's work has led to some of the biggest breakthroughs at Google and Stanford.[6]

In September 2016, a research team at Google announced the development of the Google Neural Machine Translation system (GNMT) and by November Google Translate began using neural machine translation (NMT) in preference to its previous statistical methods (SMT)[1][10][11][12] which had been used since October 2007, with its proprietary, in-house SMT technology.[13][14]

Google Translate's NMT system uses a large artificial neural network capable of deep learning.[1][2][3] By using millions of examples, GNMT improves the quality of translation,[2] using broader context to deduce the most relevant translation. The result is then rearranged and adapted to approach grammatically based human language.[1] GNMT's proposed architecture of system learning was first tested on over a hundred languages supported by Google Translate.[2] GNMT did not create its own universal interlingua but rather aimed at commonality found in between many languages, considered to be of more interest to psychologists and linguists than to computer scientists.[15] The new translation engine was first enabled for eight languages: to and from English and French, German, Spanish, Portuguese, Chinese, Japanese, Korean and Turkish in 2016.[16] In March 2017, three additional languages were enabled: Russian, Hindi and Vietnamese along with Thai for which support was added later.[17][18] Support for Hebrew and Arabic was also added with help from the Google Translate Community in the same month.[19] In mid April 2017 Google Netherlands announced support for Dutch and other European languages related to English.[20] Further support was added for nine Indian languages, viz. Hindi, Bengali, Marathi, Gujarati, Punjabi, Tamil, Telugu, Malayalam and Kannada at the end of April 2017.[21]

Languages Supported by GNMT[edit]

This is a list of language translation pairs supported by Google Translate's Neural Machine Translation (NMT) model. As of July 2017 all languages currently only support translation to and from English:[22]

Language Codes Language Pair
afen Afrikaans ↔ English
sqen Albanian ↔ English
amen Amharic ↔ English
aren Arabic ↔ English
hyen Armenian ↔ English
azen Azerbaijani ↔ English
euen Basque ↔ English
been Belarusian ↔ English
bnen Bengali ↔ English
bsen Bosnian ↔ English
bgen Bulgarian ↔ English
yueen Cantonese ↔ English
caen Catalan ↔ English
ceben Cebuano ↔ English
zh-CN * ↔ en Chinese (Simplified) ↔ English
zh-TWen Chinese (Traditional) ↔ English
coen Corsican ↔ English
hren Croatian ↔ English
csen Czech ↔ English
daen Danish ↔ English
dven Dhivehi ↔ English
nlen Dutch ↔ English
eoen Esperanto ↔ English
eten Estonian ↔ English
fien Finnish ↔ English
fren French ↔ English
fyen Frisian ↔ English
glen Galician ↔ English
kaen Georgian ↔ English
deen German ↔ English
elen Greek ↔ English
guen Gujarati ↔ English
hten Haitian Creole ↔ English
haen Hausa ↔ English
hawen Hawaiian ↔ English
iwen Hebrew ↔ English
hien Hindi ↔ English
hmnen Hmong ↔ English
huen Hungarian ↔ English
isen Icelandic ↔ English
igen Igbo ↔ English
iden Indonesian ↔ English
gaen Irish ↔ English
iten Italian ↔ English
jaen Japanese ↔ English
jwen Javanese ↔ English
knen Kannada ↔ English
kken Kazakh ↔ English
kmen Khmer ↔ English
koen Korean ↔ English
kuen Kurdish ↔ English
kyen Kyrgyz ↔ English
loen Lao ↔ English
laen Latin ↔ English
lven Latvian ↔ English
lten Lithuanian ↔ English
lben Luxembourgish ↔ English
mken Macedonian ↔ English
mgen Malagasy ↔ English
msen Malay ↔ English
mlen Malayalam ↔ English
mt <- en Maltese** <- English
mien Maori ↔ English
mren Marathi ↔ English
mnen Mongolian ↔ English
neen Nepali ↔ English
noen Norwegian ↔ English
nyen Nyanja (Chichewa) ↔ English
psen Pashto ↔ English
faen Persian ↔ English
plen Polish ↔ English
pten Portuguese (Portugal, Brazil) ↔ English
paen Punjabi ↔ English
roen Romanian ↔ English
ruen Russian ↔ English
smen Samoan ↔ English
gden Scots Gaelic ↔ English
sren Serbian ↔ English
sten Sesotho ↔ English
snen Shona ↔ English
sden Sindhi ↔ English
sien Sinhala (Sinhalese) ↔ English
sken Slovak ↔ English
slen Slovenian ↔ English
soen Somali ↔ English
esen Spanish ↔ English
swen Swahili ↔ English
sven Swedish ↔ English
tlen Tagalog (Filipino) ↔ English
tgen Tajik ↔ English
taen Tamil ↔ English
teen Telugu ↔ English
then Thai ↔ English
tren Turkish ↔ English
uken Ukrainian ↔ English
uren Urdu ↔ English
uzen Uzbek ↔ English
vien Vietnamese ↔ English
cyen Welsh ↔ English
xhen Xhosa ↔ English
yien Yiddish ↔ English
yoen Yoruba ↔ English
zuen Zulu ↔ English

Zero-shot translation[edit]

The GNMT system is said to represent an improvement over the former Google Translate in that it will be able handle "zero-shot translation", that is it directly translates one language into another (for example, Japanese to Korean).[2] Google Translate previously first translated the source language into English and then translated the English into the target language rather than translating directly from one language to another.[5] Currently, no non-English language pairs are supported.

See also[edit]


  1. ^ a b c d e f Barak Turovsky (November 15, 2016), "Found in translation: More accurate, fluent sentences in Google Translate", Google Blog, retrieved January 11, 2017
  2. ^ a b c d e f g h Mike Schuster, Melvin Johnson, and Nikhil Thorat (November 22, 2016), "Zero-Shot Translation with Google's Multilingual Neural Machine Translation System", Google Research Blog, retrieved January 11, 2017CS1 maint: uses authors parameter (link)
  3. ^ a b Gil Fewster (January 5, 2017), "The mind-blowing AI announcement from Google that you probably missed", freeCodeCamp, retrieved January 11, 2017
  4. ^ Wu, Yonghui; Schuster, Mike; Chen, Zhifeng; Le, Quoc V.; Norouzi, Mohammad. "Google's neural machine translation system: Bridging the gap between human and machine translation" (PDF). Retrieved October 1, 2018. Cite journal requires |journal= (help)
  5. ^ a b Boitet, Christian; Blanchon, Hervé; Seligman, Mark; Bellynck, Valérie (2010). "MT on and for the Web" (PDF). Retrieved December 1, 2016.
  6. ^ a b Robert D. Hof (August 14, 2014). "A Chinese Internet Giant Starts to Dream: Baidu is a fixture of online life in China, but it wants to become a global power. Can one of the world's leading artificial intelligence researchers help it challenge Silicon Valley's biggest companies?". Technology Review. Retrieved January 11, 2017.
  7. ^ Jeff Dean and Andrew Ng (June 26, 2012). "Using large-scale brain simulations for machine learning and A.I." Official Google Blog. Retrieved January 26, 2015.
  8. ^ "Google's Large Scale Deep Neural Networks Project". Retrieved October 25, 2015.
  9. ^ Markoff, John (June 25, 2012). "How Many Computers to Identify a Cat? 16,000". New York Times. Retrieved February 11, 2014.
  10. ^ Katyanna Quach (November 17, 2016), Google's neural network learns to translate languages it hasn't been trained on: First time machine translation has used true transfer learning, retrieved January 11, 2017
  11. ^ Lewis-Kraus, Gideon (December 14, 2016). "The Great A.I. Awakening". The New York Times. Retrieved January 11, 2017.
  12. ^ Le, Quoc; Schuster, Mike (September 27, 2016). "A Neural Network for Machine Translation, at Production Scale". Google Research Blog. Google. Retrieved December 1, 2016.
  13. ^ Google Switches to its Own Translation System, October 22, 2007
  14. ^ Barry Schwartz (October 23, 2007). "Google Translate Drops SYSTRAN for Home-Brewed Translation". Search Engine Land.
  15. ^ Chris McDonald (January 7, 2017), Commenting on Gil Fewster's January 5th article in the Atlantic, retrieved January 11, 2017
  16. ^ Turovsky, Barak (November 15, 2016). "Found in translation: More accurate, fluent sentences in Google Translate". The Keyword Google Blog. Google. Retrieved December 1, 2016.
  17. ^ Perez, Sarah (March 6, 2017). "Google's smarter, A.I.-powered translation system expands to more languages". TechCrunch. Oath Inc.
  18. ^ Turovsky, Barak. "Higher quality neural translations for a bunch more languages". The Keyword Google Blog. Google. Retrieved March 6, 2017.
  19. ^ Novet, Jordan (March 30, 2017). "Google now provides AI-powered translations for Arabic and Hebrew". VentureBeat.
  20. ^ Finge, Rachid (April 19, 2017). "Grote verbetering voor het Nederlands in Google Translate" [Big improvement for Dutch in Google Translate]. Google Netherlands Blog (in Dutch).
  21. ^ Turovsky, Barak (April 25, 2017). "Making the internet more inclusive in India". The Keyword.
  22. ^ "Translation API Language Support". Google Cloud Platform. May 4, 2017.

External links[edit]