AMHERST, Mass. — English readers of foreign-language digital novels have long despaired of poor translation quality, especially when the original versions were published in a non-Romance language and written with great literary sensibility. But that may soon change, thanks to an $822,365 grant to University of Massachusetts Amherst computer and information science professor Mohit Iyyer of Open Philanthropy.
Traditionally, novels have been translated by experts who not only master the denotative meaning of words in two or more languages, but are also sensitive to the subtle nuances and connotations that distinguish literature from more technical writing. It may take years for such a translator to arrive at a faithful interpretation that preserves the language play and image of the original – if such a translator can even be found. Since linguists estimate that there are over 7,000 languages spoken on earth today, much of what is written in one language will only be mistranslated into another, if translated at all. .
Although the rise of AI-based translation software has helped reduce the bottleneck, it is far from perfect. “French to English translates relatively well,” says Iyyer, “but Japanese to English is notoriously bad, and anything with a literary sensibility is hopeless.” To illustrate this point, Iyyer cites two translations of Norwegian Wood by Japanese novelist Haruki Murakami. The first, written by a professional human translator, reads as follows:
A cold November rain darkens the land, turning the scene into a dark Flemish painting. Airport workers in rain gear, flags atop faceless airport buildings, BMW billboards, everything. Just great, I think, Germany again.
Compare this to the same Japanese source text run through Google Translate:
The freezing November rain darkened the land, and the mechanics wearing rain feathers, the flag flying over the flat airport building, the BMW billboard and all were a gloomy picture of the Flemish school. It looked like the background of. I wondered if it was still Germany.
“Status quo AI translators are often much too literal,” says Iyyer, “because they are trained on news articles and parliamentary debates”
Iyyer’s solution is to bring humans back into the equation. Over the next two years, Iyyer and his team will build an online platform that will host a wide range of previously untranslated novels, which will be available in English through an AI model that his team will develop. These translations will be interactive, and readers will be able to highlight sections of text that they believe are incorrect and suggest alternatives that are easier to read. Another AI model, a post-editing model, will collect these user-generated corrections and update the AI translation model with them. This is a way for the AI translation model to “learn”.
Iyyer is quick to point out that this process cannot replace the expertise of a dedicated human translator. “But,” he says, “hopefully we can give these expert translators a head start, and in the meantime, we can help deliver readable versions of the world’s greatest literature.”