This web-book tells the story of Google Translate, the world’s most-used service to convert words among languages. Google claims to serve 103 languages, but nobody has ever before investigated what actually happens when you put in text for translation. Sometimes your results will be spot-on, sometimes they’re spotty at best, and sometimes they’re as intelligible as a red-spotted toad. What are those times, and why such variable results? Are some languages better than others? Do some things translate better than others? Can Google Translate be trusted?

The pages that follow dive into the question of how well Google Translate accomplishes the job it says it does. This started out as a technical research study, with measurements designed with academic rigor to compare translations from English to the other 102 Google languages. Along the way, though, it became evident that there was a much bigger story than whether Google found the right words in one language or another. The story involved human psychology, media gullibility, and corporate mendacity – a rich mixture of hope and hubris, built on a platform of data and deceit.

What originated as a project for an academic journal, then, has transformed into a book for the wider public. This is a story for anyone who wonders what is going on when they use Google Translate. It is written using the sorts of English expressions that regular people use every day and that a translation service should be able to translate, not the highfalutin language meant for professors to impress each other with their genius. The study should stand up to academic standards, and the peer Commentary section is an open invitation for experts working on language and technology to discuss topics surrounding the research and its interpretation. As you read this book, you will find humor and videos and digressions that you would not find in a computer science journal, with the intent that the study’s findings will be as enjoyable as they are enlightening to the most number of people.

This is not your standard study, and it is not your standard website (and it is not optimized for mobile phones – it looks better on a larger screen). There are no pop-up alerts, no ads, and no craven attempts to get your data. Please do comment in the discussion threads if you have something to say, especially if you have examples from languages that are not deeply treated in the research. Please also share with all your friends and colleagues who might use Google Translate, and therefore might like to know what they are getting when they do. And please enjoy.

How to cite: Martin Benjamin 2019, Teach You Backwards: An In-Depth Study of Google Translate for 103 Languages, Kamusi Project International, https://teachyoubackwards.com

Empirical Evaluation of Google Translate across 102 Languages

I conducted a scientific evaluation of translations produced by Google, for all 102 languages vs. English in their roster. Such a study, using native speakers for every language, has never been done before. Evaluators everywhere from Samoa to Uzbekistan examined Google translations in their language. Here is what they reported.

Qualitative Analysis of Google Translate across 103 Languages

Universal translation is a dream, like the flying car. Yet several myths lead people to believe Google has achieved it already. Artificial Intelligence. Neural Networks. Zero-Shot Translation. With these buzzwords and more, we think that Google has solved language. This chapter looks at the myths, and looks at what really happens.

The Astounding Mathematics of Machine Translation

Translation is not an easy game to play. One wrong word, or the right word in the wrong place, can flip your meaning upside-down. The opportunities for error stack one atop another. Reliable automatic translation using Google’s techniques is not just a dream. The way language works, it is a mathematical impossibility.

Disruptive Approaches for Next Generation Machine Translation

The missing link in using machines to produce great translations is, ironically, people. Language resides in our heads, so we must transmit that data to our computers before the computers can learn how to use it. This has not been done systematically, even for the few languages that Google and others have invested in heavily. Moreover, we need new ways to guide our machines through our words and towards our meaning. There is a path that could work for every language.

When & How to Use Google Translate

In some situations, GT is an excellent tool. In some, it is helpful as part of a broader translation strategy. In many, though, it should be avoided or cannot be used. Based on this study, here is a guide on how to use Google Translate, and when to run, run, run, run away.

Hello, cookie lover! We use cookies. Yummy, yummy cookies. To enjoy this site, please accept the use of cookies. Privacy Policy

The cookie settings on this website are set to "allow cookies" to give you the best browsing experience possible. If you continue to use this website without changing your cookie settings or you click "Accept" below then you are consenting to this.

Close