1
00:00:02,058 --> 00:00:04,882
Hello. Well, let me start
by asking you a question:

2
00:00:05,537 --> 00:00:08,905
How many of you had to fill out
some sort of web form

3
00:00:08,929 --> 00:00:13,318
where you've been asked to read
a distorted sequence of characters?

4
00:00:16,163 --> 00:00:18,097
How many of you found it really annoying?

5
00:00:18,611 --> 00:00:20,761
Okay, outstanding. So I invented that.

6
00:00:20,761 --> 00:00:22,373
(Laughter)

7
00:00:22,397 --> 00:00:25,848
(Applause)

8
00:00:29,954 --> 00:00:31,789
That thing is called a CAPTCHA.

9
00:00:31,789 --> 00:00:35,837
And it is there to make sure
the entity filling out the form

10
00:00:35,861 --> 00:00:38,713
is actually a human and not
some sort of computer program

11
00:00:38,737 --> 00:00:43,084
that was written to submit the form
millions and millions of times.

12
00:00:43,108 --> 00:00:45,014
The reason it works is because humans

13
00:00:45,038 --> 00:00:47,443
have no trouble reading
these squiggly characters,

14
00:00:47,467 --> 00:00:50,308
whereas computer programs
simply can't do it as well yet.

15
00:00:50,332 --> 00:00:54,276
For example, when you're buying tickets
online for attending a concert

16
00:00:54,276 --> 00:00:57,494
the reason you have to type

17
00:00:57,518 --> 00:01:02,183
these distorted characters is to prevent

18
00:01:02,207 --> 00:01:04,155
scalpers from writing a program

19
00:01:04,179 --> 00:01:07,004
that can buy millions of tickets,
two at a time.

20
00:01:07,028 --> 00:01:08,956
CAPTCHAs are used all over the Internet.

21
00:01:08,980 --> 00:01:10,511
And since they're used so often,

22
00:01:10,535 --> 00:01:15,822
a lot of times the precise sequence
of random characters shown to the user

23
00:01:15,846 --> 00:01:17,422
is not so fortunate.

24
00:01:17,446 --> 00:01:20,923
So this is an example from Yahoo.

25
00:01:20,947 --> 00:01:23,314
The random characters that happened

26
00:01:23,338 --> 00:01:27,306
to be shown to the user were W, A, I, T

27
00:01:27,330 --> 00:01:29,621
which spells a word.

28
00:01:29,645 --> 00:01:32,564
But the best part is the message
that the Yahoo help desk

29
00:01:32,588 --> 00:01:35,634
got about 20 minutes later.

30
00:01:35,658 --> 00:01:38,358
["Help! I've been waiting
for over 20 minutes,

31
00:01:38,382 --> 00:01:41,800
and nothing happens."]
(Laughter)

32
00:01:42,390 --> 00:01:45,734
This of course, is not as bad
as this poor person.

33
00:01:45,758 --> 00:01:49,017
[REBOOT]
(Laughter)

34
00:01:49,041 --> 00:01:51,549
I can tell funny stories
about captchas for hours

35
00:01:51,573 --> 00:01:53,409
but since I cannot do that

36
00:01:53,433 --> 00:01:56,218
let me tell you about a project
that we did afterwards

37
00:01:56,242 --> 00:01:58,463
which is sort of the next
evolution of CAPTCHA.

38
00:01:58,488 --> 00:01:59,845
We call it reCAPTCHA,

39
00:01:59,869 --> 00:02:03,251
which is something that
we started at the University,

40
00:02:03,275 --> 00:02:05,431
and then we turned it
into a startup company.

41
00:02:05,455 --> 00:02:07,374
And then Google acquired this company.

42
00:02:07,399 --> 00:02:09,917
so, all what I'm going to say
for the next 5 minutes

43
00:02:09,941 --> 00:02:11,505
is owned by Google.

44
00:02:11,529 --> 00:02:14,751
So, please, do not spread the word.

45
00:02:14,775 --> 00:02:18,633
So let me tell you
how this project started.

46
00:02:18,657 --> 00:02:23,127
It turns out that about 200 million
CAPTCHAs are typed everyday.

47
00:02:23,151 --> 00:02:26,239
When I first heard this,
I was quite proud of myself.

48
00:02:26,263 --> 00:02:29,266
I thought, "look at the impact
that my research has had."

49
00:02:29,290 --> 00:02:31,205
But then I started feeling bad.

50
00:02:31,229 --> 00:02:33,108
They are not only obnoxious, but also

51
00:02:33,133 --> 00:02:36,577
each time you type a CAPTCHA

52
00:02:36,601 --> 00:02:38,933
essentially you waste
10 seconds of your time.

53
00:02:38,957 --> 00:02:42,684
And if you multiply that
by 200 million you get that

54
00:02:42,708 --> 00:02:46,722
humanity as a whole is wasting
about 500,000 hours every day

55
00:02:46,746 --> 00:02:48,949
typing these annoying CAPTCHAs.

56
00:02:48,973 --> 00:02:51,417
So then I started feeling bad.

57
00:02:51,441 --> 00:02:54,469
And then I started thinking,
is there any way

58
00:02:54,493 --> 00:02:58,341
we can use this effort for something
that is good for humanity?

59
00:02:58,365 --> 00:03:03,695
While you're typing a CAPTCHA,
during those 10 seconds,

60
00:03:03,719 --> 00:03:05,844
your brain is doing something amazing.

61
00:03:05,868 --> 00:03:09,576
Your brain is doing something
that computers cannot yet do.

62
00:03:09,600 --> 00:03:11,432
So can we get you to do some

63
00:03:11,456 --> 00:03:13,166
useful work to mankind?

64
00:03:13,190 --> 00:03:14,371
Putting it differently,

65
00:03:14,396 --> 00:03:16,977
is there some humongous problem
that we cannot yet get

66
00:03:17,001 --> 00:03:18,523
computers to solve,

67
00:03:18,547 --> 00:03:20,675
yet we can split into tiny chunks

68
00:03:20,699 --> 00:03:22,856
such that each time
somebody solves a CAPTCHA

69
00:03:22,881 --> 00:03:24,934
they solve a little bit of this problem?

70
00:03:24,957 --> 00:03:27,868
The answer to that is "yes,"
and this is what we're doing now.

71
00:03:27,893 --> 00:03:32,591
So what you may not know is that
nowadays while you're typing a CAPTCHA,

72
00:03:32,615 --> 00:03:35,772
not only are you authenticating
yourself as a human,

73
00:03:35,796 --> 00:03:38,676
but in addition you're actually
helping us to digitize books.

74
00:03:38,827 --> 00:03:40,616
So let me explain how this works.

75
00:03:40,640 --> 00:03:44,297
So there's a lot of projects out there
trying to digitize the existing books.

76
00:03:44,322 --> 00:03:45,749
Google is digitizing books.

77
00:03:45,773 --> 00:03:48,418
Amazon, with the Kindle,
is digitizing books.

78
00:03:48,443 --> 00:03:51,116
The way this works
is you start with an old book.

79
00:03:51,141 --> 00:03:53,142
You've seen those things, right?

80
00:03:53,166 --> 00:03:54,204
Like a book?

81
00:03:54,229 --> 00:03:55,557
(Laughter)

82
00:03:55,582 --> 00:03:58,063
So you start with a book,
and then you scan it.

83
00:03:58,087 --> 00:04:03,549
Now scanning a book is like taking
a digital photograph of every page.

84
00:04:07,567 --> 00:04:09,883
The next step in the process
is that the computer

85
00:04:09,908 --> 00:04:15,128
needs to be able to decipher
all of the words in this image.

86
00:04:15,152 --> 00:04:18,815
Now the problem is that for older books
that were written several years ago

87
00:04:18,839 --> 00:04:21,291
the computer cannot recognize
a lot of the words

88
00:04:21,315 --> 00:04:25,119
because the ink has faded
and the pages have turned yellow.

89
00:04:25,143 --> 00:04:26,953
Thus the words look a bit different

90
00:04:26,977 --> 00:04:29,803
and the computer cannot recognize them.

91
00:04:29,827 --> 00:04:32,468
So, for books that were written
more than 50 years ago,

92
00:04:32,493 --> 00:04:35,895
the computer cannot recognize
about 30 percent of the words.

93
00:04:35,919 --> 00:04:37,084
So what we're doing now

94
00:04:37,109 --> 00:04:40,400
is we're taking all of the words
that the computer cannot recognize

95
00:04:40,425 --> 00:04:43,824
and we're getting people to read them
for us while they're typing

96
00:04:43,848 --> 00:04:45,311
a CAPTCHA on the Internet.

97
00:04:45,335 --> 00:04:47,727
So, the next time you type a CAPTCHA -

98
00:04:47,751 --> 00:04:54,322
(Applause)

99
00:04:54,346 --> 00:04:57,535
these words that you're typing

100
00:04:57,559 --> 00:05:00,927
are actually words that are coming
from books that are being digitized

101
00:05:00,951 --> 00:05:03,145
that the computer could not recognize.

102
00:05:03,169 --> 00:05:06,958
And now the reason we have
two words nowadays instead of one

103
00:05:06,982 --> 00:05:12,167
is because we need to verify
if the answer is correct.

104
00:05:13,588 --> 00:05:17,682
Because one of the words is such
that the system knows what it was,

105
00:05:17,706 --> 00:05:20,728
and the other is a word that
the system just got out of a book,

106
00:05:20,753 --> 00:05:23,538
it didn't know what it was,
and it's presented to you.

107
00:05:24,014 --> 00:05:27,069
We're going to ask you to type both words.

108
00:05:27,093 --> 00:05:29,072
And we won't tell you which one's which.

109
00:05:29,096 --> 00:05:30,682
And if you type the correct word

110
00:05:30,707 --> 00:05:33,483
for the one for which the system
already knows the answer,

111
00:05:33,508 --> 00:05:35,069
it assumes you are human,

112
00:05:35,093 --> 00:05:39,600
and it also gets some confidence
that you typed the other word correctly.

113
00:05:39,624 --> 00:05:43,264
And if we repeat this process
to like 10 different people

114
00:05:43,288 --> 00:05:45,582
and all of them agree
on what the new word is,

115
00:05:45,606 --> 00:05:47,884
we are very confident
that this new word

116
00:05:47,908 --> 00:05:49,850
was accurately digitized.

117
00:05:49,874 --> 00:05:51,920
So this is how the system works.

118
00:05:51,944 --> 00:05:54,396
The good thing is that
it has been very successful.

119
00:05:54,428 --> 00:05:57,825
We're digitizing about
100 million words a day,

120
00:05:57,849 --> 00:06:01,513
which is the equivalent of
about two million books a year.

121
00:06:01,537 --> 00:06:04,436
And this is all being done
one word at a time

122
00:06:04,460 --> 00:06:07,016
by just people typing CAPTCHAs
on the Internet.

123
00:06:07,040 --> 00:06:10,672
Now, since we're doing
so many words per day,

124
00:06:10,696 --> 00:06:14,724
funny things can happen.

125
00:06:14,748 --> 00:06:18,286
And this is especially true
because now we're giving people

126
00:06:18,310 --> 00:06:21,676
two randomly chosen English words
next to each other.

127
00:06:21,700 --> 00:06:23,976
So funny things can happen.

128
00:06:24,000 --> 00:06:25,976
For example, we presented this word.

129
00:06:26,000 --> 00:06:30,284
It's the word "Christians";
there's nothing wrong with it.

130
00:06:30,308 --> 00:06:33,976
But if you present it along with
another randomly chosen word,

131
00:06:34,000 --> 00:06:35,276
bad things can happen.

132
00:06:35,300 --> 00:06:38,128
So we get this.
[bad Christians]

133
00:06:38,152 --> 00:06:39,317
(Laughter)

134
00:06:39,341 --> 00:06:40,860
It's quite funny.

135
00:06:40,884 --> 00:06:43,316
But it's even worse,
because the particular website

136
00:06:43,340 --> 00:06:45,251
where we showed this

137
00:06:45,275 --> 00:06:50,080
actually happened to be called
The Embassy of the Kingdom of God.

138
00:06:50,104 --> 00:06:51,841
(Laughter)

139
00:06:51,865 --> 00:06:52,959
Oops!

140
00:06:52,984 --> 00:06:54,882
Here's another really bad one.

141
00:06:54,906 --> 00:06:59,417
American politician, JohnEdwards.com
[Damn liberal]

142
00:06:59,441 --> 00:07:03,691
(Laughter)

143
00:07:03,715 --> 00:07:06,836
So we keep on insulting people everyday.

144
00:07:06,860 --> 00:07:08,971
Now, we're not just insulting people.

145
00:07:08,995 --> 00:07:11,851
Quite often,
interesting things can happen.

146
00:07:11,875 --> 00:07:15,009
So this actually has given rise
to an Internet meme

147
00:07:15,033 --> 00:07:17,367
that millions of people
have participated in,

148
00:07:17,391 --> 00:07:20,173
which is called CAPTCHA art.

149
00:07:20,197 --> 00:07:21,759
Here's how it works.

150
00:07:21,783 --> 00:07:25,505
Imagine you're using the
Internet and you see a CAPTCHA

151
00:07:25,529 --> 00:07:27,917
that you think is somewhat peculiar,

152
00:07:27,941 --> 00:07:30,204
like this CAPTCHA.

153
00:07:30,228 --> 00:07:33,484
Then what you're supposed to do
is you take a screen shot of it.

154
00:07:33,508 --> 00:07:35,395
Then of course,
you fill out the CAPTCHA

155
00:07:35,420 --> 00:07:38,119
because you help us
digitize a book, please.

156
00:07:38,143 --> 00:07:40,160
But then, first you take a screen shot,

157
00:07:40,184 --> 00:07:44,389
and then you draw something
that is related to it, like this.

158
00:07:44,413 --> 00:07:46,082
[invisible toaster]

159
00:07:46,106 --> 00:07:47,136
(Laughter)

160
00:07:47,161 --> 00:07:50,639
It's just an example of CAPTCHA art.

161
00:07:50,662 --> 00:07:54,359
There are tens of thousands of these.
Some of them are interesting.

162
00:07:54,383 --> 00:07:57,672
Some of them are very cute.
[clenched it!]

163
00:07:57,696 --> 00:08:01,190
Some of them are funnier.

164
00:08:01,214 --> 00:08:07,551
[stoned founders]
(Laughter)

165
00:08:07,575 --> 00:08:11,761
This is my favorite number
of this whole project: 900 millions.

166
00:08:11,785 --> 00:08:14,921
This is the number of distinct people

167
00:08:14,945 --> 00:08:17,307
that have helped us digitize
at least one word

168
00:08:17,331 --> 00:08:19,067
out of a book through reCAPTCHA.

169
00:08:19,091 --> 00:08:21,186
A little over 10%
of the world's population,

170
00:08:21,211 --> 00:08:22,946
has helped digitize human knowledge.

171
00:08:22,970 --> 00:08:25,992
And it is numbers like these
that motivate my research agenda.

172
00:08:26,016 --> 00:08:29,553
So the question that motivates
my research is the following:

173
00:08:29,577 --> 00:08:32,626
If you look at humanity's
large-scale achievements,

174
00:08:32,652 --> 00:08:35,371
these really big things that humanity
has gotten together

175
00:08:35,397 --> 00:08:40,099
like building the pyramids of Egypt
or the Panama Canal

176
00:08:40,123 --> 00:08:42,357
or putting a man on the Moon --

177
00:08:42,381 --> 00:08:44,813
there is a curious fact about them,

178
00:08:44,837 --> 00:08:48,268
and it is that they were all done
with about the same number of people.

179
00:08:48,292 --> 00:08:50,836
They were all done with
about 100,000 people.

180
00:08:50,860 --> 00:08:53,231
We can ask ourselves
why is that all of them used

181
00:08:53,256 --> 00:08:54,842
about the same number of people.

182
00:08:54,867 --> 00:08:57,368
The reason for that is because,
before the Internet,

183
00:08:57,393 --> 00:09:00,797
coordinating more than
100,000 people was impossible.

184
00:09:00,821 --> 00:09:03,535
But now with the Internet,
I've just shown you a project

185
00:09:03,559 --> 00:09:06,433
where we've coordinated
900 million people.

186
00:09:06,457 --> 00:09:08,734
So the question that
motivates my research is,

187
00:09:08,758 --> 00:09:11,238
if we can put a man on the Moon
with 100,000 people,

188
00:09:11,262 --> 00:09:14,219
what can we do with
100 million people?

189
00:09:14,243 --> 00:09:17,371
Based on this question,
we've been working on a lot of projects.

190
00:09:17,395 --> 00:09:19,397
I will not tell you
about all we have done.

191
00:09:19,421 --> 00:09:22,446
But, let me tell you about
one that we are working on now.

192
00:09:22,470 --> 00:09:25,479
We've been working on this
for about two years now.

193
00:09:25,503 --> 00:09:29,737
And we're going to launch it
in about 30 days.

194
00:09:29,761 --> 00:09:32,511
It's called Duolingo.

195
00:09:32,535 --> 00:09:35,484
This project started asking
the following question:

196
00:09:35,508 --> 00:09:39,798
How can we get 100 million people

197
00:09:39,822 --> 00:09:45,173
translating the Web into
every major language for free?

198
00:09:45,197 --> 00:09:47,736
So there's a lot of things
to say about this question.

199
00:09:47,760 --> 00:09:49,441
First of all, translating the Web.

200
00:09:49,466 --> 00:09:51,960
Right now it is partitioned
into multiple languages.

201
00:09:51,984 --> 00:09:54,065
A large fraction of it is in English.

202
00:09:54,089 --> 00:09:56,524
If you don't know any English,
you can't access it.

203
00:09:56,548 --> 00:09:58,618
But large fractions are
in other languages,

204
00:09:58,643 --> 00:10:01,443
and if you don't know the languages
you can't access them.

205
00:10:01,468 --> 00:10:05,026
I would like to translate all
of the Web into every major language.

206
00:10:05,310 --> 00:10:08,209
Now some of you may say,

207
00:10:08,233 --> 00:10:10,686
why can't we use computers to translate?

208
00:10:10,710 --> 00:10:14,677
Machine translation nowadays is starting
to translate some sentences.

209
00:10:14,701 --> 00:10:17,079
Well the problem with that is that

210
00:10:17,103 --> 00:10:18,797
it's not yet good enough,

211
00:10:18,821 --> 00:10:23,275
and it probably won't be
for the next 20 to 30 years.

212
00:10:23,299 --> 00:10:26,531
So let me show you an example of something

213
00:10:26,555 --> 00:10:28,240
that was translated by a machine.

214
00:10:28,264 --> 00:10:32,834
Actually it was a forum
about programming questions.

215
00:10:37,614 --> 00:10:40,208
It was a programming question
translated from Japanese

216
00:10:40,233 --> 00:10:44,667
into English and from then into Spanish,
though my translation is good.

217
00:10:44,691 --> 00:10:46,941
The other one is bad. You'll see.

218
00:10:46,965 --> 00:10:50,447
So I'll just let you read.

219
00:10:50,471 --> 00:10:53,308
This person starts apologizing for
the machine translation.

220
00:10:53,332 --> 00:10:55,954
Indeed, this was done with
the best translation program

221
00:10:55,978 --> 00:10:57,421
from Japanese into English.

222
00:10:57,445 --> 00:11:03,475
Remember, it's a question
about computer programming.

223
00:11:03,499 --> 00:11:06,076
So here you are the preamble
to the question.

224
00:11:06,100 --> 00:11:11,893
[At often, the goat-time install
a error is vomit.] (Laughter)

225
00:11:11,917 --> 00:11:13,995
Then comes the first part of the question.

226
00:11:14,020 --> 00:11:19,726
[How many times like the wind, a pole,
and the dragon?] (Laughter)

227
00:11:19,750 --> 00:11:21,801
Then comes my favorite part
of the question.

228
00:11:21,825 --> 00:11:25,939
[This insult to father's stones?]
(Laughter)

229
00:11:25,963 --> 00:11:28,444
And then comes my favorite
part of the whole thing.

230
00:11:28,468 --> 00:11:32,451
[Please apologize for your stupidity.
There are a many thank you.] (Laughter)

231
00:11:32,475 --> 00:11:34,950
Okay, so computer translation
isn't yet good enough.

232
00:11:34,974 --> 00:11:36,594
We need people to translate.

233
00:11:36,618 --> 00:11:39,332
So what I want is to get
100 million people

234
00:11:39,356 --> 00:11:42,640
translating the Web into
every major language for free.

235
00:11:42,665 --> 00:11:45,600
I couldn't afford paying
100 million people for the job,

236
00:11:45,625 --> 00:11:47,202
so I want them to do it for free.

237
00:11:47,227 --> 00:11:48,940
Now if this is what you want to do,

238
00:11:48,965 --> 00:11:51,485
you pretty quickly realize
you're going to run into

239
00:11:51,509 --> 00:11:55,949
two pretty big obstacles,
needing to be hurdled.

240
00:11:55,973 --> 00:12:00,028
The first one is a lack of bilinguals.

241
00:12:00,052 --> 00:12:03,159
So I don't even know if there
exists 100 million people out there

242
00:12:03,183 --> 00:12:06,501
using the Web who are bilingual
enough to help us translate.

243
00:12:06,525 --> 00:12:07,570
That's a big problem.

244
00:12:07,594 --> 00:12:09,708
The other problem is a lack of motivation.

245
00:12:09,732 --> 00:12:13,720
How are we going to motivate people
to actually translate the Web for free?

246
00:12:13,744 --> 00:12:18,844
After thinking about this for months,

247
00:12:18,868 --> 00:12:20,864
we realized there's actually a way

248
00:12:20,888 --> 00:12:23,610
to solve both these problems
with the same solution.

249
00:12:23,634 --> 00:12:26,720
We realized that there's a way
to kill two birds with one stone.

250
00:12:26,745 --> 00:12:30,892
And that is to transform
language translation

251
00:12:30,916 --> 00:12:33,986
into something that millions
of people want to do,

252
00:12:34,010 --> 00:12:37,882
and that also helps with
the problem of lack of bilinguals,

253
00:12:37,906 --> 00:12:40,416
and that is language education.

254
00:12:40,441 --> 00:12:44,061
It turns out there are millions of people
wanting to learn other languages.

255
00:12:44,085 --> 00:12:49,119
Today there are over 1.2 billion people
learning a foreign language.

256
00:12:49,143 --> 00:12:52,326
And it's not just because
they're being forced to do so in school.

257
00:12:52,350 --> 00:12:54,936
For example, in the United States alone,
there are over

258
00:12:54,961 --> 00:12:58,936
5 million people who have paid
over $500 for software

259
00:12:58,961 --> 00:13:00,459
to learn a new language.

260
00:13:00,483 --> 00:13:03,364
Many people want to learn a new language.

261
00:13:03,388 --> 00:13:06,667
So what we've been working on
for the last two years

262
00:13:06,691 --> 00:13:08,824
is a new website called Duolingo,

263
00:13:08,848 --> 00:13:11,816
where the basic idea is
people learn a new language

264
00:13:11,840 --> 00:13:17,188
for free, while simultaneously
translating the Web.

265
00:13:17,916 --> 00:13:19,977
And so they're learning by doing.

266
00:13:20,001 --> 00:13:21,634
So this is how it works.

267
00:13:21,658 --> 00:13:25,528
The way this works is whenever
you're a just a beginner,

268
00:13:25,552 --> 00:13:28,036
we give you very,
very simple sentences on the Web.

269
00:13:28,060 --> 00:13:31,716
And if you don't know a word
we'll tell you what each word means

270
00:13:31,740 --> 00:13:34,159
though you are asked
to "translate this sentence".

271
00:13:34,184 --> 00:13:36,058
And it turns out that it really works.

272
00:13:36,082 --> 00:13:38,358
Even though people know nothing
of the language

273
00:13:38,383 --> 00:13:41,635
if we explain what each word means,
they'll be able to translate it.

274
00:13:41,660 --> 00:13:43,432
As you see how other people translate

275
00:13:43,456 --> 00:13:45,842
the same sentence,
you start learning the language.

276
00:13:45,866 --> 00:13:47,857
And as you get more and more advanced,

277
00:13:47,881 --> 00:13:51,078
we give you more and more
complex sentences to translate.

278
00:13:51,102 --> 00:13:53,431
This is how
you are going to help us translate.

279
00:13:53,455 --> 00:13:55,104
This is how the site works.

280
00:13:55,128 --> 00:13:57,118
We're mostly done building it,

281
00:13:57,142 --> 00:13:59,415
and now we're testing it.

282
00:13:59,439 --> 00:14:01,184
When we started working on this

283
00:14:01,208 --> 00:14:03,653
I didn't think it could work, really.

284
00:14:03,677 --> 00:14:06,183
But it turns out that
it works, indeed. It's amazing.

285
00:14:06,207 --> 00:14:08,988
First, people really
can learn a language with it.

286
00:14:09,012 --> 00:14:12,369
In this case we are testing it with people

287
00:14:12,393 --> 00:14:14,751
knowing English,
wanting to learn Spanish,

288
00:14:14,775 --> 00:14:16,049
and vice versa.

289
00:14:16,073 --> 00:14:18,309
So people really do learn a language.

290
00:14:18,333 --> 00:14:19,920
And they learn it about as well

291
00:14:19,945 --> 00:14:22,077
as the leading language learning software,

292
00:14:22,102 --> 00:14:24,498
which is very good,
but perhaps more surprisingly,

293
00:14:24,523 --> 00:14:28,924
the translations that we get from people
using the site are very good.

294
00:14:28,948 --> 00:14:30,767
They are as accurate as those

295
00:14:30,791 --> 00:14:33,291
of professional language translators.

296
00:14:33,315 --> 00:14:36,407
Now of course,
we play a trick here and it is that

297
00:14:36,431 --> 00:14:40,748
we combine the translations
of multiple beginners, several students,

298
00:14:40,772 --> 00:14:42,490
and choose the best.

299
00:14:42,514 --> 00:14:44,742
But it turns out
that that best translation

300
00:14:44,767 --> 00:14:47,925
is as good as those of
professional language translators.

301
00:14:47,949 --> 00:14:53,553
Now even though we're combining
multiple translations,

302
00:14:54,857 --> 00:14:57,073
another good thing about Duolingo is that

303
00:14:57,097 --> 00:15:00,188
the site actually
can translate pretty fast.

304
00:15:00,212 --> 00:15:03,531
So let me show you an estimate
of how fast we could translate.

305
00:15:03,555 --> 00:15:06,579
If we wanted to translate Wikipedia
from English into Spanish --

306
00:15:06,604 --> 00:15:09,319
of course, Wikipedia exists in Spanish
but is much smaller

307
00:15:09,343 --> 00:15:12,098
than its English counterpart,
about 20 percent of it --

308
00:15:12,122 --> 00:15:15,711
If we wanted to translate Wikipedia
from English into Spanish using Duolingo

309
00:15:15,735 --> 00:15:20,922
we could do it in five weeks
with 100,000 active users

310
00:15:20,946 --> 00:15:22,500
learning English with Duolingo.

311
00:15:22,524 --> 00:15:25,623
And we could do it in about 80 hours
with a million active users.

312
00:15:25,647 --> 00:15:28,333
Since all the projects that
my group has worked on so far

313
00:15:28,357 --> 00:15:29,994
have gotten millions of users,

314
00:15:30,018 --> 00:15:33,452
we're hopeful that we'll be able
to translate the Web for free.

315
00:15:33,476 --> 00:15:35,693
We haven't yet launched Duolingo,

316
00:15:35,717 --> 00:15:39,113
(Applause)

317
00:15:45,159 --> 00:15:46,895
I'd like to leave you with --

318
00:15:47,871 --> 00:15:49,941
we haven't yet launched Duolingo
we plan to do so in 30 days.

319
00:15:49,966 --> 00:15:52,230
If you visit Duolingo.com, you can sign up

320
00:15:52,254 --> 00:15:56,547
to be part of our private beta
in about 30 days.

321
00:15:56,572 --> 00:15:57,658
Help us.

322
00:15:57,683 --> 00:15:58,833
Thank you.

323
00:15:58,858 --> 00:16:00,373
(Applause)