Microsoft Translator

Microsoft Translator

Logo and favicon of Microsoft Translator, which appears along the title in the web browser
Type of site
 Machine translation
Owner Microsoft
Website www.microsoft.com/translator
Registration Optional
Current status Active

Microsoft Translator is a multilingual machine translation cloud service provided by Microsoft. The Microsoft Translator API is integrated across multiple consumer, developer, and enterprise products; including Bing, Microsoft Office, SharePoint, Microsoft Lync, Yammer, Skype Translator, Visual Studio, Internet Explorer, and Microsoft Translator apps for Windows, Windows Phone, iPhone and Apple Watch, and Android phone and Android Wear.

Microsoft Translator also offers text and speech translation through a cloud API as a service for businesses. Service for text translation ranges from a free tier supporting two million characters per month to paid tiers supporting billions of characters per month.[1] The speech translation API, released in March 2016,[2] is offered based on time of the audio stream and ranges from a free tier of 2 hours per month to paid tiers of up to 100 of hours per month.[3]

The service supports 60 language systems as of December 2016. It also supports 9 speech translation systems that currently power Skype Translator and Skype for Windows Desktop, and the Microsoft Translator Apps for iOS and Android.[4]

Development

History

The first version of Microsoft’s machine translation system was developed between 1999 and 2000 within Microsoft Research. This system was based on semantic predicate-argument structures known as logical forms (LF), and was spun from the grammar correction feature developed for Microsoft Word. This system was eventually used to translate the entire Microsoft Knowledge Base into Spanish, French, German, and Japanese.[5]

Microsoft’s approach to machine translation, like most of current modern machine translation systems, is “data-driven”[6] —rather than relying on writing explicit rules to translate natural language, algorithms are trained to understand and interpret translated parallel texts, allowing them to automatically learn how to translate new natural language text. Microsoft’s experience with the LF-system led directly to a treelet translation system which simplified the LF to dependency trees, and eventually to an order template model which led to significant improvements in speed and the incorporation of new target languages.

The consumer-facing translation site known as Bing Translator (previously known as Windows Live translator) was launched in 2007 and provides free text and website translations on the web. Text is translated directly within the Bing Translator webpage while websites are translated through the Bilingual Viewer tools.

In 2011, the service was extended to include numerous Microsoft Translator products through a cloud-based API, which supports products available to both consumer and enterprise users. An additional speech translation capability was introduced in March 2016.[2]

Translation methodology and research

Microsoft Translator uses machine translation to create instantaneous translations from one natural language to another. This system is based on four distinct areas of computer learning research seen below.[6]

Type of Learning Impact on Translation
Neural Networks
Main article: Deep Learning
 Neural networks try to mimic how the brain works to translate between languages. At a high-level, neural network translation works in two stages. First, a first stage models the word that needs to be translated based on the context of this word (and its possible translations) within the full sentence. Second, the neural network translates this word model (not the word itself but the model the neural networks built of it), within the context of the sentence, into the other language. [7]
Syntax-Based SMT Syntax-based translation is based on the idea of translating syntactic units, rather than a word or string of words. Microsoft has used Syntax-based SMT to translate much of its computer-related texts from English into multiple target languages. Ongoing research in this area has produced improvements in word inflections and word ordering.
Phrase-Based SMT In Phrase-Based SMT, the machine learns correspondence between languages from parallel text without the aid of linguist knowledge. This produces better translations in less time than other systems.
Word Alignment SMT systems rely on existing translated data to learn how to automatically translate from one language to another. To train the systems, identifying word correspondences (or word alignments) is crucial. Microsoft has developed work in both discriminative[8] and generative[9] approaches to word alignment, resulting in faster algorithms and higher quality and translations.
Language Modeling Language Modeling uses n-gram models to construct comprehensible translations in the target language. This ensures that the output translation is fluent and readable.

Accuracy

The quality of Microsoft Translator’s machine translation outputs are evaluated using a method called the BLEU score.[10]

BLEU (Bilingual Evaluation Understudy) is an algorithm for evaluating the quality of text which has been machine-translated from one natural language to another. Quality is considered to be the correspondence between a machine's output and that of a human. BLEU was one of the first metrics to achieve a high correlation with human judgments of quality, and remains one of the most popular automated and inexpensive metrics.

Because machine translation is based on statistical algorithms rather than human translators, the automatic translations it produces are not always entirely accurate. Microsoft Translator has introduced various feedback features, such as the Collaborative Translations Framework, into its products to allow users to suggest alternative translations. These alternative translations are then integrated into the Microsoft Translator algorithms to improve future translations.

In November 2016, Microsoft Translator introduced translation using deep neural networks in nine of its highest traffic languages, including all of its speech languages and Japanese. Neural networks provide better translation than industry standard statistical machine learning.[11]

Core products

Microsoft Translator is a cloud-based API that is integrated into numerous Microsoft products and services.[12] The Translator API can be used on its own and can be customized for use in a pre-publishing or post-publishing environment. The API, which is available through subscription, is free for lower translation volumes, and is charged according to a tiered payment system for volumes exceeding two million characters per month.[1] The remaining core products are available for free.

Microsoft Translator API

The Microsoft Translator API is a cloud-based automatic translation service that can be used to build applications, websites, and tools requiring multi-language support.

Collaborative Translations Framework (CTF)

The Collaborative Translations Framework (CTF) is an extension of the Microsoft Translator API that allows post-publishing improvement of translated text.[13] By using the CTF, readers have the ability to suggest alternative translations to those provided by the API, or vote on previously offered alternatives. This information is then delivered to the API to improve future translations.

Microsoft Translator Hub

The Microsoft Translator Hub allows enterprises and language service providers to build their own translation systems that understand business- and industry-specific terminology.[14] The Hub can also be used in conjunction with the CTF, allowing administrators to approve CTF results and add them directly to the Hub.

The Hub has also been used for language preservation, allowing communities to create their own language translation systems for language and cultural preservation.[15] The Hub has been used to create translation systems for languages such as Hmong, Mayan, Nepali, and Welsh.

Translator Web Widget

The Translator Web Widget is a translation tool that can be added to web pages by pasting a predefined snippet of JavaScript code into the page.[16] The web widget is offered for free by Microsoft, and supports both pre-publishing customized translations using the Translator Hub, and post-publishing improvements using the Collaborative Translations Framework.

Multilingual App Toolkit (MAT)

The Multilingual App Toolkit (MAT) is an integrated Visual Studio tool, which allows developers to streamline localization workflows of their Windows, Windows Phone and desktop apps.[17] MAT improves localization of file management, translation support, and editing tools.

Bing Translator website

Main article: Bing § Translator
Bing Translator
Type of site
 Statistical machine translation
Available in 60 languages, see supported languages
Owner Microsoft
Website www.bing.com/translator/
Commercial No
Registration Optional
Launched June 3, 2009 (2009-06-03)
Current status Active

Bing Translator (previously Live Search Translator and Windows Live Translator)[18] is a user facing translation portal provided by Microsoft as part of its Bing services to translate texts or entire web pages into different languages. All translation pairs are powered by the Microsoft Translator, a statistical machine translation platform and web service, developed by Microsoft Research, as its backend translation software. Two transliteration pairs (between Chinese (Simplified) and Chinese (Traditional)) are provided by Microsoft's Windows International team.[19]

As of December 2016, Bing Translator offers translations in 60 different language systems.[20]

The Bilingual Viewer showing the English translation of the French Wikipedia's main page.

Bing Translator can translate phrases entered by the user or acquire a link to a web page and translate its entirely. When translating an entire web page, or when the user selects "Translate this page" in Bing search results, the Bilingual Viewer is shown, which allows users to browse the original web page text and translation in parallel, supported by synchronized highlights, scrolling, and navigation.[21] Four Bilingual Viewer layouts are available:[21]

Bing Translator integrates with several other Microsoft products. The following is a table of products into which Bing Translator is integrated or may be integrated:

Integrates into Mean of integration
Bing Instant Answers Already integrated
Internet Explorer An Accelerator for Internet Explorer 8 or higher[22]

Supported products

Through its core product offerings, Microsoft Translator supports the translation features of many Microsoft products at the consumer and enterprise levels. These products fall broadly into three categories[23]— communication products, Microsoft Office, and apps.

Communication

Microsoft Office

Apps

Supported languages

As of December 2016, Microsoft Translator supports 60 different language systems.[4] The list of supported languages is available at the Microsoft Translator website and can also be retrieved programmatically through the API.[24]

Community partners

Microsoft Translator has engaged with community partners to increase the number of languages and to improve overall language translation quality. Below is a list of community partners that Microsoft Translator has teamed with.[25]

Additionally, Microsoft has teamed with the Klingon Language Institute, which promotes the constructed language, Klingon, which is used within the fictional Star Trek universe produced by Paramount and CBS Studios. Klingon has been supported by Microsoft Translator since May, 2013.[26]

See also

References

  1. 1 2 "Azure Data Marketplace- Microsoft Translator".
  2. 1 2 "Microsoft Translator brings end-to-end speech translation to everyone with the world's first Speech Translation API".
  3. ↑ "Azure Data Marketplace- Microsoft Translator Speech API".
  4. 1 2 "Microsoft Translator- Languages".
  5. ↑ "Microsoft Research- Arul Menezes".
  6. 1 2 "Microsoft Research- Machine Translation".
  7. ↑ "What is neural network based translation?".
  8. ↑ "A Discriminative Framework for Bilingual Word Alignment" (PDF).
  9. ↑ "Using Word Dependent Transition Models in HMM based Word Alignment for Statistical Machine Translation" (PDF).
  10. ↑ "Microsoft Translator Hub: Discussion of BLEU Score".
  11. ↑ "Microsoft Translator launching Neural Network based translations for all its speech languages".
  12. ↑ "Microsoft Translator- Products".
  13. ↑ "Microsoft Translator- Collaborative Translations Framework".
  14. ↑ "Microsoft Translator- Translator Hub".
  15. ↑ "Where Language Meets the World: Microsoft Translator Hub".
  16. ↑ "Microsoft Translator- Translator Web Widget".
  17. ↑ "Microsoft Translator- Multilingual App Toolkit".
  18. ↑ Antoniou, Grigoris; Grobelnik, Marko; Simperl, Elena; Parsia, Bijan; Plexousakis, Dimitris; de Leenheer, Pieter; Pan, Jeff Z. (2011). The Semantic Web: Research and Applications. Berlin: Springer Science+Business Media. p. 341. ISBN 9783642210334.
  19. ↑ "Translation powered by....Microsoft Translator!". Microsoft Translator (and Bing Translator) Official Team Blog. Microsoft Corporation. 8 September 2008. Retrieved 21 October 2010.
  20. ↑ "Microsoft Translator- FAQ".
  21. 1 2 "About Bing Translator". Bing Translator. Microsoft. Retrieved 20 December 2014.
  22. ↑
  23. ↑ "Microsoft Translator- Products".
  24. ↑ "Translator Language Codes".
  25. ↑ "Microsoft Translator- Community Partners".
  26. ↑ "New Bing Translator Option Lets You Translate... Klingon".
This article is issued from Wikipedia - version of the 12/5/2016. The text is available under the Creative Commons Attribution/Share Alike but additional terms may apply for the media files.