Less Lost in Translation
Non-native English speakers attempting to express themselves in the global language of business and science get a software assist from Microsoft’s Beijing lab.
Imagine that you have to write something in a language that is not your mother tongue. This happens frequently to, for example the 22 million native Spanish-speaking people in the United States. A technophile might say: let them write in Spanish and then have the computer translate it to English. The problem is that such automatic machine translation is not yet good enough. The technology can help professional translators do their job faster, but it is of little use to someone who doesnt recognize the sometimes subtle mistakes made by the translation program. If the writer is not confident in the target language, the “help” from machine translation holds little value.
A tool introduced recently in China by Microsoft helps writers who are not native in English to write better English. Called the English Writing Wizard, it is the first product that addresses the difficult task of giving suggestions to someone who has little or no ability to distinguish between good and bad advice. Although the wizard can help with translation, it is not, strictly speaking, a machine translation tool. It is more akin to the grammar checkers familiar to users of common word-processing programsbut enhanced to work with people not native in English. A writer uncertain about how to phrase an idea in English can type it directly in Chinese and get a high-quality translation. The wizard can also detect errors typical for Chinese speakers writing in English, such as forgotten or incorrect articlesa mistake made by few native English speakers.
The writing wizard is the result of work by Ming Zhou, a natural language processing and machine translation specialist at Microsoft Research Asia in Beijing. Using large numbers of Chinese and English texts, Zhou and his colleagues have learned what mistakes are common and what the writer probably intended. Using this quantitative approachlarge sets of texts rather than complicated language rulesthey also join the statistical trend from the last couple of years. It has proven far more effective than expected.
Before coming to Microsoft, Zhou was a professor of computer science at Tsinghua University, where he designed software that translated between Chinese and Japanese. He also designed the CEMT-I machine translation system, the first experiment of Chinese-English machine translation in China. Despite this work, Zhou doesn’t think that machine translation technology is, for the most part, good enough to serve as a writing aid.
In our new tool we only translate words and phrases, never whole sentences because the user must feel confident with the translation, says Zhou. The ideal user is someone who has good knowledge of the second language, but still needs advice on specific words and phrases, as well as help in detecting errors characteristic of Chinese speakers.
One feature the Writing Wizard provides is to help users find word combinations that can be used together. Enter the word book as a verb, for example, and you could find phrases such as book flight or book hotel. If you are looking for the best way to thank someone for a gift, you can find a list of adjectives that work with the word present.
Zhou says that the Writing Wizard has been well received by the users, but he concedes that it sometimes gives bad advice. This is a problem that the program has in common with all other language tools: We call it the selection problem,” Zhou says. “Users tend to use the word selected by the wizard even if it is not ideal, because they don’t know any better.
The Writing Wizard is not yet integrated into a word processor the way grammar and spellcheck software is. It runs more like old time checkers, in a separate window. We would like to improve the automation and integration, Zhou says.
Zhou would also like develop the wizard so that it mimics the users’ own texts. When you answer an e-mail, he points out, it is easy to just copy some phrases from the received message to get the correct level of formality or casualness. He envisions a software tool that could help in this process in a more advanced wayfor example, using the words in a question when answering it.
For now, the Writing Wizard is sold only in China and only to large corporate customers. But Zhou wants to add more languages to the system’s repertoire; Japanese and Korean versions would probably come first. And he thinks that with the large base of knowledge his team has built for the Chinese-English wizard, they could easily port the system to other languages, including Spanish, French, and German. Software help for those 22 million Spanish speakers in the U.S. may just be on its way.
The race is on to define the new blockchain era. Get a leg up at Business of Blockchain 2019.Register now