1
00:00:02,930 --> 00:00:06,605
Hi everyone. I am excited to see you on board,

2
00:00:06,605 --> 00:00:08,510
and welcome to our course.

3
00:00:08,510 --> 00:00:13,070
I want to start our lesson with the informal discussion of who we are,

4
00:00:13,070 --> 00:00:15,225
and who is this course for.

5
00:00:15,225 --> 00:00:20,025
Then, we will have a brief introduction to the area of Natural Language Processing.

6
00:00:20,025 --> 00:00:25,100
You know, it might feel a little hand wavy as any introduction actually,

7
00:00:25,100 --> 00:00:27,515
but I hope that after our course,

8
00:00:27,515 --> 00:00:32,585
you will know exactly everything that will be mentioned in our lesson now.

9
00:00:32,585 --> 00:00:35,555
So, ready? Let us get started.

10
00:00:35,555 --> 00:00:40,555
My name is Anna, and we have a big nice team creating the course for you.

11
00:00:40,555 --> 00:00:43,120
So, we have Sergey, Alexey, Andrey,

12
00:00:43,120 --> 00:00:46,535
and one more Anna preparing the materials.

13
00:00:46,535 --> 00:00:50,795
I have a background on computer science and machine learning,

14
00:00:50,795 --> 00:00:56,410
and I'm now applying this background in natural language processing in different ways.

15
00:00:56,410 --> 00:01:00,160
And you know these different activities like research, teaching,

16
00:01:00,160 --> 00:01:04,700
and industry, you've had different perspectives to the same area.

17
00:01:04,700 --> 00:01:09,230
So, for example, when you come to the industry very soon,

18
00:01:09,230 --> 00:01:15,460
you realize that not a new paper from academia is useful in the particular settings,

19
00:01:15,460 --> 00:01:22,430
like large scale implementation or some noisy dater or specific needs of your business.

20
00:01:22,430 --> 00:01:24,625
So, probably, you need to build

21
00:01:24,625 --> 00:01:30,410
some more simple solution but that would work nicely in your specific settings.

22
00:01:30,410 --> 00:01:34,525
Okay. Now, who is this course for?

23
00:01:34,525 --> 00:01:40,340
When I was thinking about what would be one word to characterize our audience,

24
00:01:40,340 --> 00:01:43,435
I thought that it would be the word curious.

25
00:01:43,435 --> 00:01:50,155
So, this course is for curious people who want to know what is inside some applications.

26
00:01:50,155 --> 00:01:53,950
For example, you have differently used machine translation.

27
00:01:53,950 --> 00:01:56,725
Do you know how it works or

28
00:01:56,725 --> 00:02:01,295
dialogue agents that are so popular nowadays? What is inside there?

29
00:02:01,295 --> 00:02:06,860
And you know, this popularity of certain applications is couldn't bet.

30
00:02:06,860 --> 00:02:09,710
So, for example, for dialogue agents,

31
00:02:09,710 --> 00:02:13,495
we have so much hype around so that it is not

32
00:02:13,495 --> 00:02:17,965
that easy to distinguish what is just some beautiful words,

33
00:02:17,965 --> 00:02:21,930
and what is something that will really work in practice.

34
00:02:21,930 --> 00:02:27,215
So, hopefully, one outcome of our course for you would be

35
00:02:27,215 --> 00:02:32,920
the ability to distinguish between the hype and something that really works.

36
00:02:32,920 --> 00:02:36,775
Now, our course is rather in-depth.

37
00:02:36,775 --> 00:02:41,745
So, I want to go with some details through several methods in NLP

38
00:02:41,745 --> 00:02:47,630
because these will give you the ability to distinguish the hype from the methods.

39
00:02:47,630 --> 00:02:50,210
Okay? Also, we will cover

40
00:02:50,210 --> 00:02:54,550
real state-of-the-art approaches both in research and production.

41
00:02:54,550 --> 00:02:56,505
And as I have already said,

42
00:02:56,505 --> 00:02:59,365
this could be rather different approaches.

43
00:02:59,365 --> 00:03:03,410
Now, another goal that's a little bit contradict to

44
00:03:03,410 --> 00:03:08,250
going in-depth would be to have a big picture of the area.

45
00:03:08,250 --> 00:03:13,125
So, I feel like it is really important to have some expertise like,

46
00:03:13,125 --> 00:03:16,745
I am given a task, what should I do with it?

47
00:03:16,745 --> 00:03:20,290
What approaches would work in this certain case?

48
00:03:20,290 --> 00:03:22,285
To have this intuition,

49
00:03:22,285 --> 00:03:27,310
we will try to discuss as many different settings and tasks as possible,

50
00:03:27,310 --> 00:03:30,005
and cover some approaches for them.

51
00:03:30,005 --> 00:03:36,365
And obviously, we should not only talk and you should not only listen and read about it,

52
00:03:36,365 --> 00:03:40,710
but you have to do some practice to get a hands-on experience.

53
00:03:40,710 --> 00:03:45,000
So, we are preparing materials for you for home assignments in

54
00:03:45,000 --> 00:03:50,795
Python for some popular NLP tasks like text classification,

55
00:03:50,795 --> 00:03:54,445
or duplicate detection, named entity recognition,

56
00:03:54,445 --> 00:03:59,355
and some others, so that you have some experience with your own hands.

57
00:03:59,355 --> 00:04:02,985
Also, this home tasks will help you to build

58
00:04:02,985 --> 00:04:08,030
the project of our course that would be a conversational chat-bot.

59
00:04:08,030 --> 00:04:15,525
So now, I feel like it is really important also to see what is our course not about,

60
00:04:15,525 --> 00:04:21,450
because NLP is so big that obviously our course cannot feed everyone's needs.

61
00:04:21,450 --> 00:04:25,550
So, I feel like if you only want to know

62
00:04:25,550 --> 00:04:31,085
some black box implementations and stock them together to build some solution,

63
00:04:31,085 --> 00:04:34,210
then probably, this course is not for you.

64
00:04:34,210 --> 00:04:38,550
Also, I think that it is a good idea to take machine learning and

65
00:04:38,550 --> 00:04:44,075
deep learning courses first to fill it is with some names and formulas.

66
00:04:44,075 --> 00:04:47,420
For example here, I have a quick test for you.

67
00:04:47,420 --> 00:04:50,450
Do you know what is Recurrent Neural Networks?

68
00:04:50,450 --> 00:04:53,910
Or have you heard about likelihood maximization?

69
00:04:53,910 --> 00:04:57,500
Just take a moment to see how comfortable you are with

70
00:04:57,500 --> 00:05:01,190
these words and see whether you need to take,

71
00:05:01,190 --> 00:05:03,070
for example, deep learning course in

72
00:05:03,070 --> 00:05:07,470
our specialization first before going to this course.

73
00:05:07,470 --> 00:05:11,420
Also, we expect that you have some experience with Python.

74
00:05:11,420 --> 00:05:14,910
Probably, you don't have any experience with TensorFlow,

75
00:05:14,910 --> 00:05:16,600
and this is maybe okay,

76
00:05:16,600 --> 00:05:22,495
and then this is a good moment for you to try to go through some tutorials,

77
00:05:22,495 --> 00:05:25,965
and this course could be a good reason to go through them.

78
00:05:25,965 --> 00:05:28,975
Actually, TensorFlow has really nice tutorials,

79
00:05:28,975 --> 00:05:31,940
so I think that it shouldn't be a problem for you.

80
00:05:31,940 --> 00:05:34,385
I hope you are still not frightened.

81
00:05:34,385 --> 00:05:38,700
And I hope you are ready for our journey to the NLP.

82
00:05:38,700 --> 00:05:44,690
And I want to start this journey with the survey of the main approaches.