With the global expansion of the Internet, more people with different language backgrounds are in communication than ever before. In response, an abundance of online translation services have emerged; some are easy-to-use programs which translate text instantly from one language to another, while others link the user to someone who is fluent in multiple languages of interest.
The main hurdle for further development of online translation services is the fundamental reality that people acquire language through social interactions, and computers cannot engage in such interactions and so must be programmed to understand language in a way that is artificial. Even with brilliant minds working for decades on translation service development, these services still depend on research from the field of Natural Language Processing, a field characterized by widespread disagreements about basic principles of language learning. Even if software can use language to accomplish a multitude of complex tasks, this does not imply that the software “understands” language as we understand the meaning of the word. The best examples of why Natural Language Processing is so difficult to program are in dealing with slang, expressions, dialects, irony and humor. Researchers are making progress in modeling all of these phenomena, but there is much work to be done, and therefore, highly effective automatic translation services are likely still years away.
Google Translate leads the way in instant, online translations. Aside from 'auto' translating sites on the web, users can provide text input in one language and select the target language for translation. But the translations that result are often awkward and hard to understand, because languages only rarely have one-to-one correlations for words and meanings.
Google has enlisted the help of expert polyglots, including professional linguists, to assist in improving the accuracy and effectiveness of Translate. They have a new forum that allows for people who know multiple languages to analyze translations and to edit them to improve the algorithms used by Google Translate. All of the translations would be open for all volunteers to edit, and Translate will determine the best new translations through statistical analysis.
Duolingo is taking a different approach from Google’s, attempting to revolutionize both online translation and language learning in one fell swoop. Luis Von Ahn the co-creator of Duolingo is focusing on developing a language-learning program online that is completely free for users. Users will be able to log in and start practicing with easy translations, gradually progressing to more advanced translations. Duolingo will then compare everyone’s translations to determine progress for each person as well as the most accurate translation, much as Google Translate’s new forum does. Duolingo has currently generated more than $38M in 3 funding rounds.
Log in to DataFox to access the full Duolingo company profile. Image from DataFox:
One startup in this space to keep a close eye on is Unbabel, a "human corrected machine translation service that enables business to communicate globally." Unbabel gives online retailers a path for multi-language customer service support, as well as a transcription service for email newsletters ("Communicate with your customers in their language"). The company was launched in March 2014, backed by Y Combinator. According to a recent TechCrunch article, Unbabel recently took a $1.5M round from "Jared Fliesler of Matrix Partners and Kevin Rose of Google Ventures, among others".
Translation via Distributed Teams
Another type of translation service worth mentioning is services done by distributed teams. These are relatively slow services, but companies that effectively produce translation work this way have shown some potential. But as with all firms in this space, major breakthroughs in productivity and profitability will only come through advances in natural language processing technology, and such advances have so far been short of revolutionary.