Showing Revision 3 created 05/25/2016 by Udacity Robot.

1. Title:
   Unit 22 15 Translation Examplemp4
2. Description:

3. 133 0:00 - 0:07
   Here's an example of translation. And this is a phrase-based translation that doesn't happen to use any syntactic trees.
4. 7433 0:07 - 0:13
   And in this case we're using an example going from German to English.
5. 12900 0:13 - 0:19
   And in this model we break up the probability of the translation into three components.
6. 18833 0:19 - 0:25
   First, a segmentation model of how do we break up the German into phrases?
7. 25300 0:25 - 0:31
   Here the sentence has been broken up into 1-2-3-4-5 phrases.
8. 30833 0:31 - 0:37
   Then a translation model. For each phrase, what's a good translation into English?
9. 36867 0:37 - 0:45
   And then a distortion model, saying, "Each of these phrases, what order would be a good order to put them into?"
10. 45500 0:46 - 0:50
    So let's look at each of those in turn. First, the segmentation model.
11. 50300 0:50 - 1:00
    We have a database of phrases that we picked out--maybe through a similar process to what went on in the Chinese menu,
12. 59967 1:00 - 1:07
    where we looked for coherent phrases that occurred frequently, and so we're able to apply probabilities to them.
13. 67500 1:08 - 1:15
    So now we have a probability. What's the probability that "morgen" is a phrase and that "fliege" is a phrase by itself?
14. 75000 1:15 - 1:20
    We would also consider the probability that they're considered a phrase together,
15. 80267 1:20 - 1:24
    and come up with a high probability segmentation.
16. 84367 1:24 - 1:30
    Next, the translation model. That's going between the two sides of the Chinese menu.
17. 89833 1:30 - 1:36
    How often, when we saw the phrase "morgen", did it correspond to the phrase "tomorrow" in English?
18. 96000 1:36 - 1:40
    And so on for the other phrases. So far, that's all pretty straight-forward.
19. 100267 1:40 - 1:49
    And the we have the distortion model, saying, "In what order should we put these phrases? Should we swap them around in any order?"
20. 109367 1:49 - 1:54
    And we measure that just by looking at the beginning and the ending of each phrase.
21. 114467 1:54 - 2:06
    So Ai is the beginning of the i phrase, and Bi minus one is the ending of the i minus one phrase.
22. 126467 2:06 - 2:15
    But we measure those in the German, but we consider the indexes, the I's, by looking at the English.
23. 134533 2:15 - 2:22
    So we say, "This is the last phrase, is this phrase, 'in Canada,'" but that corresponds to this one here,
24. 141767 2:22 - 2:31
    and so the beginning of that phrase is at number three, and the next to last phrase is this one, that corresponds to "zero confidence."
25. 151400 2:31 - 2:41
    And the end of that phrase is at seven. And so the distortion there from three to seven is a distortion of four.
26. 161167 2:41 - 2:46
    And our distortion model, then, would just be a probability distribution over those integers.
27. 166133 2:46 - 2:53
    So it's not doing anything fancy in terms of saying what type of phrases occur before or after what other types of phrases.
28. 173367 2:53 - 3:00
    It's just saying, "Are they shifted to the right or to the left? And are they shifted a small amount or a large amount?"
29. 180167 3:00 - 3:06
    And I should note that in this model, if we had a one to one translation where things weren't switched--
30. 186367 3:06 - 3:16
    So say, if the original German sentence had "zur Konferenz" before "nach Kanada," and we translated it into English like this,
31. 195867 3:16 - 3:26
    then the Bi minus one would be five, and the Ai--imagine this being swapped over here--would also be five.
32. 205600 3:26 - 3:31
    In that case the distortion would be zero. And so for a language where the words line up
33. 210767 3:31 - 3:38
    very closely between the source and the target language--for those pairs of languages--then we'd have a high probability
34. 218167 3:38 - 3:43
    mass under a zero distortion, and lower probability for other distortions.
35. 222833 3:43 - 3:50
    In a language where lots of things are swapped far, a more volatile type of translation between the language pairs,
36. 230367 3:50 - 3:58
    then we'd expect the probability mass to be lower for zero distortion, and higher for higher distortions.
37. 237800 3:58 - 4:05
    So this is a very simple model. It takes only into account segmentation, translation between phrases,
38. 244867 4:05 - 4:13
    and just the simplest model of distortion. You can imagine a more complex model based on trees and other components.
39. 252533 4:13 - 4:18
    And I should note that this is just the translation part of the model.
40. 257567 4:18 - 4:23
    And then to make the final choice, we would want to multiply out all these probabilities,
41. 262933 4:23 - 4:28
    but we would also want to take into account the probability of the generated English sentence.
42. 268267 4:28 - 4:34
    Is this a good sentence in English? And we have a probability model for that.
43. 273700 4:34 - 4:40
    That's a monolingual model rather than a bilingual model. And the process of coming up with the best translation, then,
44. 280300 4:40 - 4:48
    is just a search through all possible segmentations, all possible translations, all possible distortions,
45. 288300 4:48 - 4:56
    multiply up these probabilities times the monolingual probability, and find the one that gives you the highest value,
46. 295700 4:56 - 5:02
    and that'll be your best translation. And the tricky part is just coming up with a search technique that can enumerate
47. 302467 5:02 - 5:07
    through many of those possibilities quickly, and choose a good one.