3 ms·
I remember it being not all that long ago, as China (and greater Asia) ramped up on the internet, that the West was deeply concerned about the mono-lingual Asia
by bane 12y ago
I remember it being not all that long ago, as China (and greater Asia) ramped up on the internet, that the West was deeply concerned about the mono-lingual Asians not being able to take advantage of all the wisdom, science and art the West had produced.
In some ways this is true, in some ways it's self-enforced, and in some ways the reverse is true as well.
Automatic machine translation, and query expansion was supposed to make these inter-lingual barriers go away. But it feels like in many ways, the effort has died. Sure you can still go to translate.google.com and put in a URL and get a reasonable translated page back for a large number of languages. But I can't go to google.com and put in a query for refrigerator and it brings back matches for 冰箱 or 电冰箱 and translates those automagically. That next step just never seemed to happen, and what were fast advances in East Asian language machine translation just seems to have died out (Korean/Japanese <-> English is virtually unusable).
Even within like-language groups, ones that have "good enough" translations, this next step didn't happen. If I want to search for réfrigérateur I still have to translate the word myself, then go back to google.com, paste the word in and then hit "translate page" next to the search results (and hope that chrome's automatic translation kicks in on the search results so I have some idea what I'm clicking).
- paublyrne 12y agoYes. We will all have to become a lot more multi-lingual instead. A large amount of secondary students here now choose Chinese over French, German, or Spanish. It's easy to be 'lazy' as a native English speaker.
- davmre 12y ago> Automatic machine translation, and query expansion was supposed to make these inter-lingual barriers go away. But it feels like in many ways, the effort has died. Machine translation is still very much an active area of research. It sucked for many years as people hand-crafted more and more complicated systems; the big breakthroughs of the past ten years or so have come from realizing that, given very large (internet-scale) training datasets, you can apply some simple statistical methods and do much better than the previous hard-coded methods that couldn't exploit this wealth of data. There's been lots of tuning of the shallow statistical approach, but my understanding is that significant further progress will involve models that capture a deeper understanding of semantics while still maintaining the ability to be trained from large datasets. Identifying such models is hard and is an active area of research. Neural networks have been receiving a lot of study recently as one possible approach (see e.g. this demo http://104.131.78.120/ http://104.131.78.120/ from Yoshua Bengio's lab), but there are other avenues as well. Basically I don't think it's accurate to say "the effort has died". Research takes time, the technology is improving, and it's quite possible that within another ten or fifteen years we'll be able do to the sort of seamless, accurate translation across the Internet that you'd like to see. (translation between European languages, which are fundamentally similar in many ways, is already close to that point, but bridging the gap to Asian languages has been much harder...)