I agree with pretty much everything in this video. Quite a good overview.
A few thoughts.
1. One of the reasons why Google's machine translation is so successful is that they were brave enough to use a corpus-based approach, when everyone else previously had attempted direct analysis of a sentence and tried to produce a corresponding translation. A corpus, in case anyone is unaware, basically means a collection of texts. So, originally, people were trying to go, "Okay, this is an English sentence so it probably follows a subject-verb-object structure. Find the subject, find the verb, find the object, plug in the corresponding words in the other language, and output." Except that doesn't work, for so many reasons. For example, not all English sentences follow set grammatical rules, grammatical rules vary widely between languages, there are untranslatable concepts, etc. This is how you end up with stuff like "all your base are belong to us."
What Google did was collect tons of pre-translated texts. So they'd go, "Okay, we have an English version of this text, and a French version. Break down each sentence to its fundamental structure, and map it to the corresponding parts of the other text. We don't care what's a subject or object or verb, all we care about is that they correspond to each other in these texts that have been confirmed to be correct translations." Google's been able to do this because they had unmatched processing and storage capabilities, compared to other machine translation developers. They could afford to store and analyze gazillions of texts. This approach still isn't perfect yet, as shown in the video, but it does produce translations that often are much more legible than the previous approaches.
2. For the thing about time in the video, I get what he's saying, but I don't think that's the best possible example, because it's not really a translation issue. It's more about differences in cultural norms. In English, "meet me at 7pm" means exactly that, but if you're translating into a language or culture where punctuality isn't as emphasized, generally speaking you're not going to add a line "and I mean 7pm precisely, you chronically late idiot." You're still just going to say "meet me at 7pm" in Italian or Chinese, and hope that your reader knows enough about timekeeping standards in the English-speaking world.
The guy in the video is trying to make the point that translations need to take into account the cultural backgrounds of the source and target languages. So, for example, if you're translating a letter from English to French, in French you need to tailor the end of the letter depending on who the letter is being addressed to. In English, you can use "yours sincerely" for pretty much everything that's not informal, ranging from your colleague to your boss to your son's teacher to the President. In French, the ending of the letter will change based on its recipient, ranging from "Cordialement" to "Je vous prie de croire, Monsieur, à l'assurance de mes salutations distinguées". This is something that a computer cannot understand, at least not yet.
3. Finally, context is very important. Being able to tell the difference between a "ball" that you kick and a "ball" that you dance at (or a "ball" that comes in pairs and that you should, for the love of god, not kick) is easy for a human, depending on context. For a computer, that's much harder, because it now needs to start
understanding the text rather than just mapping it to the corresponding text in a foreign language. This reminds me of a translation error I recently encountered in Deus Ex: Human Revolution. In the China parts of the game, if you start killing people, civilians will cower and shout 沒了, 沒了... (
meile, meile...). This, in English, is "no more." Which makes sense, right? The civilians are witnessing a heavily-augmented guy going on a murder spree, and they're wailing "no more, no more!" Except "meile" in Chinese does not mean "I do not want any more of this," it means "there is none left" or "it's all gone." So this was a very obvious translation error to me, it showed that the translator wasn't given enough context to translate correctly. I had to go through Hengsha listening to people who sounded like they were very sad about there being no more cookies, or something like that.
