It offers a website interface, a mobile app for Android and iOS, as well as an API that helps developers build browser extensions and software applications.

Launched in April 2006 as a statistical machine translation service, it originally used United Nations and European Parliament documents and transcripts to gather linguistic data.

[9] During a translation, it looked for patterns in millions of documents to help decide which words to choose and how to arrange them in the target language.

It uses this broader context to help it figure out the most relevant translation, which it then rearranges and adjusts to be more like a human speaking with proper grammar".

Despite this, Google initially did not hire experts to resolve this limitation due to the ever-evolving nature of language.

[12] In January 2010, Google introduced an Android app and iOS version in February 2011 to serve as a portable personal interpreter.

[14] It uses deep learning techniques to translate whole sentences at a time, which has been measured to be more accurate between English and French, German, Spanish, and Chinese.

In the web interface, users can suggest alternate translations, such as for technical terms, or correct mistakes.

Until March 2023, some less widely spoken languages used the open-source eSpeak synthesizer for their speech; producing a robotic, awkward voice that may be difficult to understand.

[51] A January 2011 Android version experimented with a "Conversation Mode" that aims to allow users to communicate fluidly with a nearby person in another language.

[52] Originally limited to English and Spanish, the feature received support for 12 new languages, still in testing, the following October.

[56][57][13] The original January launch only supported seven languages, but a July update added support for 20 new languages, with the release of a new implementation that utilizes convolutional neural networks, and also enhanced the speed and quality of Conversation Mode translations (augmented reality).

The technology underlying Instant Camera combines image processing and optical character recognition, then attempts to produce cross-language equivalents using standard Google Translate estimations for the text as it is perceived.

[63][64][65] The Translate API page stated the reason as "substantial economic burden caused by extensive abuse" with an end date set for December 1, 2011.

[66] In response to public pressure, Google announced in June 2011 that the API would continue to be available as a paid service.

The system's original creator, Franz Josef Och, has criticized the effectiveness of rule-based algorithms in favor of statistical approaches.

Och was the head of Google's machine translation group until leaving to join Human Longevity, Inc. in July 2014.

Hence, publishing in English, using unambiguous words, providing context, or using expressions such as "you all" may or may not make a better one-step translation depending on the target language.

To acquire this huge amount of linguistic data, Google used United Nations and European Parliament documents and transcripts.

Google representatives have been involved with domestic conferences in Japan where it has solicited bilingual data from researchers.

[113] In August 2016, a Google Crowdsource app was released for Android users, in which translation tasks are offered.

Tests in 44 languages showed that the "suggest an edit" feature led to an improvement in a maximum of 40% of cases over four years.

It uses this broader context to help it figure out the most relevant translation, which it then rearranges and adjusts to be more like a human speaking with proper grammar".

[2] GNMT's "proposed architecture" of "system learning" has been implemented on over a hundred languages supported by Google Translate.

[120] With the end-to-end framework, Google states but does not demonstrate for most languages that "the system learns over time to create better, more natural translations.

[125] GNMT was first enabled for eight languages: to and from English and Chinese, French, German, Japanese, Korean, Portuguese, Spanish and Turkish.

[2][119] In March 2017, it was enabled for Hindi, Russian and Vietnamese,[126] followed by Bengali, Gujarati, Indonesian, Kannada, Malayalam, Marathi, Punjabi, Tamil and Telugu in April.

When text is well-structured, written using formal language, with simple sentences, relating to formal topics for which training data is ample, it often produces conversions similar to human translations between English and a number of high-resource languages.

[citation needed] Since its English reference material contains only "you" forms, it has difficulty translating a language with "you all" or formal "you" variations.

(English database designed and developed for Foras na Gaeilge by Lexicography MasterClass Ltd.) Welsh language data from Gweiadur by Gwerin.