My dad has a story he likes to tell about one of his friends, a scientist. The scientist was giving a lecture in Japan, and opened with a joke that lasted a couple minutes. After delivering the joke in English, he waited for his translator to relay it to the audience. The translator spoke for only a few seconds, and then the crowd burst out laughing.
After the presentation was over, the scientist asked the translator how she managed to distill the humor of his joke down into such a concise form. She shrugged and said, “I said that the American visitor just told a very funny joke, and that they should all laugh now.”
The scientist's story illustrates the subjective, human quality of translation. Moving between languages is rarely a matter of transposing literal meaning; it requires the constant triage of unexpected inputs, endless judgment calls, and some social awareness. In other words, it’s something that humans are cut out for, and that computers are not.
With enough coaching, could a computer become as good as a human? At Microsoft, hundreds of humans are trying to train a machine to listen, translate, and then speak. Last December, the company announced the limited release of Skype Translator, which can translate a conversation between two people videochatting in different languages, in real time. The software, which is still invite-only, can handle English, Spanish, Italian, and Mandarin. Google, too, has a smartphone app—free and available to the public—that can transcribe spoken text in one language, translate it, and then speak the result aloud in another. It’s not hard to imagine Google embedding this technology into its own videochat platform to a similar end.