WEBVTT

1
00:00:04.940 --> 00:00:08.109 A:middle L:90%
Bill Hart Davidson is an associate professor of rhetoric and

2
00:00:08.109 --> 00:00:10.380 A:middle L:90%
writing at Michigan State University, and he's the co

3
00:00:10.380 --> 00:00:14.109 A:middle L:90%
director of the Writing and Digital Environments, or Wide

4
00:00:14.119 --> 00:00:17.440 A:middle L:90%
Research Center at Matrix, a digital humanities and social

5
00:00:17.440 --> 00:00:21.390 A:middle L:90%
sciences research center. He teaches courses in technical communication

6
00:00:21.399 --> 00:00:24.449 A:middle L:90%
, interaction, design and research methods. He recently

7
00:00:24.449 --> 00:00:27.210 A:middle L:90%
can pleaded a term as president of the Association of

8
00:00:27.219 --> 00:00:30.350 A:middle L:90%
Teachers of Technical Writing, and in January he will

9
00:00:30.350 --> 00:00:33.439 A:middle L:90%
begin an appointment as associate dean of graduate studies in

10
00:00:33.439 --> 00:00:35.960 A:middle L:90%
the College of Arts and Sciences at Michigan State.

11
00:00:36.640 --> 00:00:38.890 A:middle L:90%
Bill is added at a book along with Jim Rodolfo

12
00:00:38.890 --> 00:00:41.859 A:middle L:90%
of the University of Kentucky called Rhetoric in the Digital

13
00:00:41.859 --> 00:00:46.530 A:middle L:90%
Humanities Rhetoric and Or In Rhetoric and the Digital Humanities

14
00:00:46.539 --> 00:00:49.590 A:middle L:90%
, forthcoming from the University of Chicago Press and available

15
00:00:49.600 --> 00:00:52.380 A:middle L:90%
early next year. Bills writing has also appeared in

16
00:00:52.380 --> 00:00:55.960 A:middle L:90%
journals such as Technical Communication, Technical Communication Quarterly,

17
00:00:56.140 --> 00:00:58.859 A:middle L:90%
The Journal of Business and Technical Communication, Computers and

18
00:00:58.859 --> 00:01:02.509 A:middle L:90%
Composition, Cairo's and the Journal of Community Informatics,

19
00:01:02.509 --> 00:01:04.060 A:middle L:90%
and in lots and lots of edited collections as well

20
00:01:04.439 --> 00:01:07.099 A:middle L:90%
through his work at wide. He also produces writing

21
00:01:07.099 --> 00:01:10.730 A:middle L:90%
software and is a co founder of a spinoff venture

22
00:01:10.730 --> 00:01:14.599 A:middle L:90%
, Drawbridge. Drawbridges, a public private partnership created

23
00:01:14.599 --> 00:01:18.870 A:middle L:90%
to make Eli Review a software service that supports writing

24
00:01:18.879 --> 00:01:21.530 A:middle L:90%
instruction. And on a more personal note, his

25
00:01:21.530 --> 00:01:26.590 A:middle L:90%
Twitter bio describes him as father, academic, juggler

26
00:01:26.810 --> 00:01:30.280 A:middle L:90%
, cyclist cook, and he describes his philosophy of

27
00:01:30.280 --> 00:01:33.069 A:middle L:90%
life in a single word. Give. And we

28
00:01:33.069 --> 00:01:34.000 A:middle L:90%
here at Virginia Tech over the past two days have

29
00:01:34.000 --> 00:01:38.290 A:middle L:90%
taken him up on that philosophy of life with our

30
00:01:38.290 --> 00:01:42.000 A:middle L:90%
philosophy, which is paid. We have put Bill

31
00:01:42.000 --> 00:01:44.620 A:middle L:90%
through his paces with lots of meetings and lots of

32
00:01:44.620 --> 00:01:47.769 A:middle L:90%
meals, and he has performed with good spirits and

33
00:01:47.769 --> 00:01:51.620 A:middle L:90%
admirably all along the way. Yesterday I emailed my

34
00:01:51.620 --> 00:01:53.560 A:middle L:90%
old faculty mentor from my PhD days and told him

35
00:01:53.560 --> 00:01:56.859 A:middle L:90%
that Bill Hart Davidson was on campus and he responded

36
00:01:56.859 --> 00:01:59.659 A:middle L:90%
with one line prince of a man, one of

37
00:01:59.659 --> 00:02:01.829 A:middle L:90%
my favorite people in the field. And that certainly

38
00:02:01.829 --> 00:02:05.140 A:middle L:90%
has been my experience as well. Please join me

39
00:02:05.140 --> 00:02:12.530 A:middle L:90%
in welcoming Bill Heart. Dave, I want to

40
00:02:12.530 --> 00:02:15.599 A:middle L:90%
say thank you to Quinn. That was embarrassingly nice

41
00:02:15.610 --> 00:02:17.830 A:middle L:90%
. Uh, introduction and thank you all for giving

42
00:02:17.830 --> 00:02:21.539 A:middle L:90%
me part of your afternoon sharing with with me and

43
00:02:21.550 --> 00:02:23.349 A:middle L:90%
, um for coming to talk with me. Um

44
00:02:23.360 --> 00:02:28.150 A:middle L:90%
, I'm going to speak a little bit about a

45
00:02:28.159 --> 00:02:30.530 A:middle L:90%
research project in wide that we have been doing now

46
00:02:30.530 --> 00:02:34.659 A:middle L:90%
for about a year, but which is really still

47
00:02:34.669 --> 00:02:38.219 A:middle L:90%
quite experimental. And so it's exciting to have moments

48
00:02:38.219 --> 00:02:42.620 A:middle L:90%
like this one where we have assembled groups from all

49
00:02:42.620 --> 00:02:44.360 A:middle L:90%
over the campus. I understand we have some folks

50
00:02:44.360 --> 00:02:46.740 A:middle L:90%
from computer science and folks from the business school and

51
00:02:46.740 --> 00:02:50.969 A:middle L:90%
folks from all over the humanities, and we're thrilled

52
00:02:50.969 --> 00:02:53.189 A:middle L:90%
to have an opportunity to talk. And I'm hoping

53
00:02:53.189 --> 00:02:54.639 A:middle L:90%
that it is a chance for you to give me

54
00:02:54.639 --> 00:02:58.069 A:middle L:90%
some feedback and push back a little bit on some

55
00:02:58.069 --> 00:03:00.259 A:middle L:90%
of the things that we're saying. Um, my

56
00:03:00.259 --> 00:03:05.090 A:middle L:90%
topic. I want to start with the title because

57
00:03:05.099 --> 00:03:07.729 A:middle L:90%
we use where I use a word in that title

58
00:03:07.729 --> 00:03:08.930 A:middle L:90%
. That is a rhetorical term of art. The

59
00:03:08.930 --> 00:03:13.219 A:middle L:90%
word topic. Um, it has a special meaning

60
00:03:13.219 --> 00:03:16.169 A:middle L:90%
in, um in rhetorical inquiry that comes from the

61
00:03:16.180 --> 00:03:20.030 A:middle L:90%
Greek word to poi from which it, uh,

62
00:03:20.039 --> 00:03:23.319 A:middle L:90%
inherits. And it isn't quite the same as we

63
00:03:23.319 --> 00:03:27.550 A:middle L:90%
think about when we talk about using the word topic

64
00:03:27.550 --> 00:03:32.759 A:middle L:90%
in everyday speech. And so I'll start there with

65
00:03:34.319 --> 00:03:36.419 A:middle L:90%
some things that you might have heard before I heard

66
00:03:36.419 --> 00:03:38.210 A:middle L:90%
this on the plane. Ladies and gentlemen, the

67
00:03:38.210 --> 00:03:40.150 A:middle L:90%
captain has turned on the fasten seat belt sign.

68
00:03:40.150 --> 00:03:42.879 A:middle L:90%
If you haven't already done so, please start your

69
00:03:42.879 --> 00:03:44.560 A:middle L:90%
carry on luggage underneath the seat in front of you

70
00:03:44.560 --> 00:03:47.449 A:middle L:90%
and an overhead bin. Um, so toe poi

71
00:03:47.460 --> 00:03:53.389 A:middle L:90%
show up not exactly as topics but ways of framing

72
00:03:53.400 --> 00:03:55.080 A:middle L:90%
that can act as a constraint on the on the

73
00:03:55.080 --> 00:04:01.009 A:middle L:90%
speaker but also, um, portend the expectations of

74
00:04:01.009 --> 00:04:03.689 A:middle L:90%
the listener or reader. And so they form a

75
00:04:03.689 --> 00:04:06.729 A:middle L:90%
kind of mini version of a social contract, Um

76
00:04:06.729 --> 00:04:09.759 A:middle L:90%
, and as a result they tend to show up

77
00:04:10.139 --> 00:04:14.020 A:middle L:90%
a lot. And this This is why when you

78
00:04:14.030 --> 00:04:16.220 A:middle L:90%
hear the Greek translation the most common Greek translation of

79
00:04:16.220 --> 00:04:20.029 A:middle L:90%
the word toe boy, it's commonplace. And so

80
00:04:20.040 --> 00:04:23.819 A:middle L:90%
keep that those two words in mind. And that's

81
00:04:23.829 --> 00:04:27.540 A:middle L:90%
a really helpful shorthand. It's a commonplace that we

82
00:04:27.540 --> 00:04:30.959 A:middle L:90%
return to when a situation seems to call for it

83
00:04:30.339 --> 00:04:33.170 A:middle L:90%
. And so if you are sitting on an airplane

84
00:04:33.170 --> 00:04:38.060 A:middle L:90%
and you're about to push back from the jetway,

85
00:04:38.639 --> 00:04:42.639 A:middle L:90%
it not only is commonplace, but there's this commonplace

86
00:04:42.639 --> 00:04:45.350 A:middle L:90%
that occurs in a string of them occur. The

87
00:04:45.350 --> 00:04:46.819 A:middle L:90%
interesting thing about that is that the exact syntax can

88
00:04:46.819 --> 00:04:49.350 A:middle L:90%
vary. Um, the words could be a little

89
00:04:49.350 --> 00:04:53.129 A:middle L:90%
bit different I was on a Southwest flight once.

90
00:04:53.139 --> 00:04:54.680 A:middle L:90%
Um, I think I was going to m l

91
00:04:54.680 --> 00:04:56.439 A:middle L:90%
a. Uh, So it was a writer on

92
00:04:56.439 --> 00:04:58.889 A:middle L:90%
Christmas time, and the crew did the whole thing

93
00:04:58.949 --> 00:05:01.660 A:middle L:90%
in rhyme, Um, uh, to the to

94
00:05:01.660 --> 00:05:04.379 A:middle L:90%
the same rhythm as twas the night before Christmas.

95
00:05:04.389 --> 00:05:08.230 A:middle L:90%
Right? And it was quite entertaining, and it

96
00:05:08.230 --> 00:05:10.790 A:middle L:90%
worked out perfectly fine, because we had we had

97
00:05:10.790 --> 00:05:13.509 A:middle L:90%
the common places in our head. If we were

98
00:05:13.509 --> 00:05:15.459 A:middle L:90%
going to be a little more specific, we could

99
00:05:15.459 --> 00:05:16.769 A:middle L:90%
start describing these. And you can see I have

100
00:05:16.769 --> 00:05:19.610 A:middle L:90%
a shot here from the delta in flight safety video

101
00:05:19.620 --> 00:05:23.370 A:middle L:90%
. Um, that shows that we can kind of

102
00:05:23.370 --> 00:05:27.730 A:middle L:90%
see these commonplaces change. Um, format or change

103
00:05:27.740 --> 00:05:30.319 A:middle L:90%
in this case, uh, mode. We're seeing

104
00:05:30.319 --> 00:05:34.040 A:middle L:90%
the visual correspondence. Um, in technical communication,

105
00:05:34.040 --> 00:05:38.319 A:middle L:90%
we might call these rhetorical moves. And that's the

106
00:05:38.319 --> 00:05:40.560 A:middle L:90%
word I'm going to start. I'm going to use

107
00:05:40.560 --> 00:05:43.259 A:middle L:90%
to stand in for this more fancy pants. Greek

108
00:05:43.259 --> 00:05:46.329 A:middle L:90%
word. Um, And just to step back,

109
00:05:46.329 --> 00:05:48.449 A:middle L:90%
I'll tell you that our computational project is to try

110
00:05:48.449 --> 00:05:53.269 A:middle L:90%
to teach the computer how to find rhetorical moves and

111
00:05:53.269 --> 00:05:55.160 A:middle L:90%
turns out that humans are pretty good at it.

112
00:05:55.540 --> 00:05:58.360 A:middle L:90%
Amazingly good at it. and computers. It's a

113
00:05:58.360 --> 00:06:00.540 A:middle L:90%
hard, hard problem. We had a chance to

114
00:06:00.540 --> 00:06:02.680 A:middle L:90%
talk with group from the business school that Quinn is

115
00:06:02.680 --> 00:06:05.870 A:middle L:90%
part of, and and we had a great talk

116
00:06:05.870 --> 00:06:09.410 A:middle L:90%
about this this morning, and it really gives you

117
00:06:09.410 --> 00:06:13.040 A:middle L:90%
an immense respect for the Hermeneutic ability of a human

118
00:06:13.040 --> 00:06:15.110 A:middle L:90%
being to tease out meaning from a text or even

119
00:06:15.110 --> 00:06:17.920 A:middle L:90%
from a visual artifact. But if I'm talking about

120
00:06:17.930 --> 00:06:21.850 A:middle L:90%
again that toe boy or the topics associated with,

121
00:06:21.860 --> 00:06:26.209 A:middle L:90%
um uh, the the message that you here at

122
00:06:26.209 --> 00:06:28.519 A:middle L:90%
the beginning of a flight, we might call them

123
00:06:28.519 --> 00:06:30.720 A:middle L:90%
part of a common set of user assistance. So

124
00:06:30.720 --> 00:06:34.319 A:middle L:90%
now I'm looking at Carlos, um, uh,

125
00:06:34.329 --> 00:06:38.139 A:middle L:90%
moves that we teach students in a technical communication class

126
00:06:38.139 --> 00:06:40.279 A:middle L:90%
. So one when we're guiding action, we really

127
00:06:40.279 --> 00:06:43.459 A:middle L:90%
make three basic moves that are really important, and

128
00:06:43.459 --> 00:06:45.930 A:middle L:90%
they cut across lots of different situations and so they

129
00:06:45.930 --> 00:06:47.529 A:middle L:90%
can vary. But you want to stage the action

130
00:06:47.529 --> 00:06:49.269 A:middle L:90%
to be performed. You want to tell people about

131
00:06:49.269 --> 00:06:53.459 A:middle L:90%
the prerequisites you want to foreshadow the success conditions.

132
00:06:54.040 --> 00:06:55.779 A:middle L:90%
Then you want to coach them through the steps that

133
00:06:55.779 --> 00:06:58.769 A:middle L:90%
they have to take, and you offer clarifying detail

134
00:06:58.769 --> 00:07:00.069 A:middle L:90%
. Here's a Here's a little one that always impresses

135
00:07:00.069 --> 00:07:03.259 A:middle L:90%
people. UH, we tech com People recognize two

136
00:07:03.259 --> 00:07:06.019 A:middle L:90%
kinds of clarifying detail, and they often show up

137
00:07:06.019 --> 00:07:08.930 A:middle L:90%
his images. One shows you how to do the

138
00:07:08.930 --> 00:07:11.639 A:middle L:90%
step, like it shows you the knob and a

139
00:07:11.649 --> 00:07:13.060 A:middle L:90%
little arrow to turn it, and the other one

140
00:07:13.060 --> 00:07:15.319 A:middle L:90%
shows you what the knob should look like. If

141
00:07:15.319 --> 00:07:17.680 A:middle L:90%
you've done it correctly, that's the process versus outcome

142
00:07:17.680 --> 00:07:21.579 A:middle L:90%
Visual. Often, when you're cursing those Ikea things

143
00:07:21.579 --> 00:07:24.310 A:middle L:90%
, it's because you have one and not the other

144
00:07:24.360 --> 00:07:26.449 A:middle L:90%
, right. Um, the one that you need

145
00:07:27.639 --> 00:07:30.019 A:middle L:90%
. And then we have alerting behaviors, right?

146
00:07:30.019 --> 00:07:31.230 A:middle L:90%
Warning. If you go too far, you'll punch

147
00:07:31.230 --> 00:07:33.629 A:middle L:90%
a hole or be careful you don't want to hurt

148
00:07:33.629 --> 00:07:39.569 A:middle L:90%
yourself. Um, alerting moves now as varied as

149
00:07:39.579 --> 00:07:42.079 A:middle L:90%
writing to guide action is, and you can think

150
00:07:42.079 --> 00:07:45.339 A:middle L:90%
about this manifesting and lots of different specific genres.

151
00:07:45.339 --> 00:07:47.199 A:middle L:90%
Say, baking a cake on the back of a

152
00:07:47.209 --> 00:07:50.029 A:middle L:90%
box of brownies. You'll see these moves recur,

153
00:07:50.029 --> 00:07:53.779 A:middle L:90%
staging, coaching and alerting. And after even with

154
00:07:53.779 --> 00:07:56.459 A:middle L:90%
just this amount of education that I've given you today

155
00:07:56.639 --> 00:07:58.490 A:middle L:90%
, if I pass around a bunch of Duncan Hines

156
00:07:58.490 --> 00:08:00.860 A:middle L:90%
boxes in a highlighter, you could all find them

157
00:08:01.439 --> 00:08:03.110 A:middle L:90%
. You could say, Here's the staging moves.

158
00:08:03.540 --> 00:08:05.339 A:middle L:90%
And here's the coaching moves. And they would say

159
00:08:05.339 --> 00:08:09.540 A:middle L:90%
, Well, how did you know that this picture

160
00:08:09.540 --> 00:08:13.379 A:middle L:90%
of two eggs and a picture of a glass of

161
00:08:13.379 --> 00:08:16.259 A:middle L:90%
water is a staging move right? How do we

162
00:08:16.269 --> 00:08:18.939 A:middle L:90%
get the computer to recognize that? Well, that's

163
00:08:18.939 --> 00:08:22.449 A:middle L:90%
a list of prerequisites, right? It's a visual

164
00:08:22.040 --> 00:08:24.860 A:middle L:90%
, and so these moves can show up in lots

165
00:08:24.860 --> 00:08:26.939 A:middle L:90%
of different ways. Here's the seat back card that

166
00:08:26.939 --> 00:08:28.709 A:middle L:90%
none of you ever look at even though they tell

167
00:08:28.709 --> 00:08:31.379 A:middle L:90%
you to. So if you look at the language

168
00:08:31.379 --> 00:08:35.740 A:middle L:90%
that I told you before about stowing your things in

169
00:08:35.740 --> 00:08:39.799 A:middle L:90%
the overhead bin, visually we see the process and

170
00:08:39.799 --> 00:08:43.210 A:middle L:90%
outcome. Visuals are indicated by arrows, and we

171
00:08:43.210 --> 00:08:46.019 A:middle L:90%
see the right way and the wrong way. Your

172
00:08:46.029 --> 00:08:48.299 A:middle L:90%
tray table must be in the upright and locked position

173
00:08:48.299 --> 00:08:50.990 A:middle L:90%
, and your seat back must be right so we

174
00:08:50.990 --> 00:08:54.279 A:middle L:90%
can reproduce the text and we can see the moves

175
00:08:54.279 --> 00:08:56.129 A:middle L:90%
happening. And we can even recognize them once we

176
00:08:56.129 --> 00:08:58.610 A:middle L:90%
are sensitized to them. So this is really what

177
00:08:58.610 --> 00:09:01.309 A:middle L:90%
we're trying to to think about ways to see if

178
00:09:01.309 --> 00:09:03.320 A:middle L:90%
we could teach the computer to do that, and

179
00:09:03.320 --> 00:09:05.730 A:middle L:90%
I'll talk about why I'll come back and talk about

180
00:09:05.730 --> 00:09:11.889 A:middle L:90%
why that's a useful thing. So this is a

181
00:09:11.889 --> 00:09:16.570 A:middle L:90%
research group in our writing and Digital Environments Research Center

182
00:09:16.570 --> 00:09:20.009 A:middle L:90%
at Michigan State that consists of not just me,

183
00:09:20.009 --> 00:09:22.919 A:middle L:90%
but a close colleague of mine has worked on just

184
00:09:22.919 --> 00:09:26.570 A:middle L:90%
about everything you'll see here today is our post doc

185
00:09:26.570 --> 00:09:31.070 A:middle L:90%
researcher named Ryan O Miso, who earned his PhD

186
00:09:31.070 --> 00:09:31.970 A:middle L:90%
from Ohio State. And it was with us.

187
00:09:31.970 --> 00:09:37.990 A:middle L:90%
We're fortunate for two years, and also my colleagues

188
00:09:37.990 --> 00:09:39.649 A:middle L:90%
, Dean Ray Berger, who is the director of

189
00:09:39.649 --> 00:09:41.600 A:middle L:90%
Matrix. Jeff Grable, Um, who is the

190
00:09:41.600 --> 00:09:45.710 A:middle L:90%
chair of my department and Liza Pots. Um,

191
00:09:45.720 --> 00:09:46.389 A:middle L:90%
so all of the work that I'm going to talk

192
00:09:46.389 --> 00:09:48.460 A:middle L:90%
about here, they get some credit for. I

193
00:09:48.460 --> 00:09:50.850 A:middle L:90%
get all the blame they get all they get to

194
00:09:50.850 --> 00:09:56.100 A:middle L:90%
share the credit. Broadly speaking, the piece that

195
00:09:56.100 --> 00:10:00.029 A:middle L:90%
I'm going to talk about today metaphorically involves our attempts

196
00:10:00.029 --> 00:10:03.149 A:middle L:90%
to create some testing methods, and here the metaphor

197
00:10:03.149 --> 00:10:05.840 A:middle L:90%
I'll invoke is a little more specific. We're thinking

198
00:10:05.840 --> 00:10:09.090 A:middle L:90%
of them of them as essays, Um, which

199
00:10:09.090 --> 00:10:13.450 A:middle L:90%
are certain kind of chemical test to determine what compounds

200
00:10:13.450 --> 00:10:18.399 A:middle L:90%
are in a A particular thing. In our case

201
00:10:18.399 --> 00:10:22.840 A:middle L:90%
we're using the compound is a complex rhetorical text or

202
00:10:22.840 --> 00:10:26.370 A:middle L:90%
a group of texts and the rules of grammar and

203
00:10:26.370 --> 00:10:33.090 A:middle L:90%
syntax provide the markers that we're seeing, what we

204
00:10:33.090 --> 00:10:33.990 A:middle L:90%
need to see. And they also provide these little

205
00:10:33.990 --> 00:10:37.419 A:middle L:90%
hooks upon which we can build subsequent parts of the

206
00:10:37.419 --> 00:10:41.269 A:middle L:90%
analysis. If we're sticking with that s a language

207
00:10:41.269 --> 00:10:45.710 A:middle L:90%
. They are the proteins or other sites that we

208
00:10:45.710 --> 00:10:48.220 A:middle L:90%
can bind to in order to amplify some of the

209
00:10:48.220 --> 00:10:52.460 A:middle L:90%
signals and see if what we're seeing is in there

210
00:10:52.539 --> 00:10:54.559 A:middle L:90%
, and we could dampen down the noise. Um

211
00:10:56.039 --> 00:10:56.779 A:middle L:90%
, and that's a really I know I got all

212
00:10:56.779 --> 00:11:00.690 A:middle L:90%
sciences there, Um, but that's part of what

213
00:11:00.690 --> 00:11:03.389 A:middle L:90%
we're trying to reach for our ways to describe a

214
00:11:03.389 --> 00:11:05.759 A:middle L:90%
kind of analytic procedure that we really haven't experimented with

215
00:11:05.759 --> 00:11:09.340 A:middle L:90%
very much So please at the end. And and

216
00:11:09.340 --> 00:11:11.289 A:middle L:90%
, uh, when as these things occur to you

217
00:11:11.289 --> 00:11:13.610 A:middle L:90%
, jot them down and push back. And I'm

218
00:11:13.610 --> 00:11:16.659 A:middle L:90%
always looking for better ways to describe that. But

219
00:11:16.659 --> 00:11:18.279 A:middle L:90%
as I mentioned, humans are really good at this

220
00:11:18.279 --> 00:11:20.940 A:middle L:90%
end. Computers? Not so. It's not so

221
00:11:20.940 --> 00:11:24.919 A:middle L:90%
much. We had a few specific questions that we've

222
00:11:24.919 --> 00:11:28.740 A:middle L:90%
been working on all year, and two in,

223
00:11:28.750 --> 00:11:31.110 A:middle L:90%
uh, boldface. Here are are ones that I'll

224
00:11:31.110 --> 00:11:33.210 A:middle L:90%
try to address and show you a little bit of

225
00:11:33.210 --> 00:11:37.919 A:middle L:90%
the work that we've done. Um, we started

226
00:11:37.929 --> 00:11:41.889 A:middle L:90%
with a very pragmatic thing. That is the top

227
00:11:41.889 --> 00:11:45.440 A:middle L:90%
question. And that is we had a big data

228
00:11:45.440 --> 00:11:48.120 A:middle L:90%
set that was human coded, um, as part

229
00:11:48.120 --> 00:11:52.090 A:middle L:90%
of a qualitative study that Jeff had led, Um

230
00:11:52.100 --> 00:11:52.860 A:middle L:90%
, that I'll talk about in just a minute.

231
00:11:54.240 --> 00:11:56.120 A:middle L:90%
And it was arduous and painful, and so they

232
00:11:56.120 --> 00:11:58.009 A:middle L:90%
were hoping that we could find some ways to make

233
00:11:58.009 --> 00:12:01.039 A:middle L:90%
it a little easier on them. Um, along

234
00:12:01.039 --> 00:12:03.830 A:middle L:90%
the way, we evolved some other questions that messed

235
00:12:03.830 --> 00:12:07.440 A:middle L:90%
with our group, which isn't, which was slight

236
00:12:07.440 --> 00:12:11.080 A:middle L:90%
words were slightly different, but which this dataset gave

237
00:12:11.080 --> 00:12:13.649 A:middle L:90%
us an occasion to explore. And the one in

238
00:12:15.840 --> 00:12:18.139 A:middle L:90%
the first boldface one is one that we address this

239
00:12:18.139 --> 00:12:20.740 A:middle L:90%
morning a little bit with the research group that you

240
00:12:20.740 --> 00:12:24.129 A:middle L:90%
have here on campus that is looking at postings in

241
00:12:24.129 --> 00:12:31.200 A:middle L:90%
the Motley Fool financial Services website Forum. And they're

242
00:12:31.200 --> 00:12:33.830 A:middle L:90%
doing some analytic procedures on that. And it is

243
00:12:33.830 --> 00:12:37.490 A:middle L:90%
what is the best way to represent a text as

244
00:12:37.490 --> 00:12:39.590 A:middle L:90%
a computational object. If we're going to turn a

245
00:12:39.590 --> 00:12:43.460 A:middle L:90%
text into something you can do math with or on

246
00:12:43.470 --> 00:12:48.330 A:middle L:90%
, or do you use mathematical techniques to transform what

247
00:12:48.330 --> 00:12:50.830 A:middle L:90%
are the best ways to do it? Um,

248
00:12:50.840 --> 00:12:54.450 A:middle L:90%
and of course it depends. But our answer for

249
00:12:54.450 --> 00:12:56.350 A:middle L:90%
today is a graph. And I'll, um,

250
00:12:56.360 --> 00:13:00.570 A:middle L:90%
if you walk away with one strange idea about what

251
00:13:00.570 --> 00:13:03.269 A:middle L:90%
this guy talked about today, it's you talk kind

252
00:13:03.269 --> 00:13:05.820 A:middle L:90%
of about how we can see text as graphs.

253
00:13:07.440 --> 00:13:11.080 A:middle L:90%
Um, we then needed to find some more specific

254
00:13:11.080 --> 00:13:13.539 A:middle L:90%
ways to break down that graph, even if we

255
00:13:13.539 --> 00:13:16.860 A:middle L:90%
could get the transformation to happen and understand what was

256
00:13:16.860 --> 00:13:20.789 A:middle L:90%
going on. But ultimately, the boldface question at

257
00:13:20.789 --> 00:13:22.700 A:middle L:90%
the bottom is what we were after. Can we

258
00:13:22.700 --> 00:13:28.000 A:middle L:90%
identify rhetorical moves? That character characterized the way discussions

259
00:13:28.000 --> 00:13:33.009 A:middle L:90%
carried out in different genres develop over time. And

260
00:13:33.009 --> 00:13:35.429 A:middle L:90%
what we're doing has a lot of, um,

261
00:13:35.440 --> 00:13:39.110 A:middle L:90%
similarity to techniques in digital humanities in the literature area

262
00:13:39.120 --> 00:13:41.559 A:middle L:90%
called Distant Reading. And I don't know if any

263
00:13:41.559 --> 00:13:45.230 A:middle L:90%
of you have read there's a big chronicle piece about

264
00:13:45.230 --> 00:13:48.470 A:middle L:90%
a year ago on distance reading. Distance reading is

265
00:13:48.470 --> 00:13:50.769 A:middle L:90%
a word is kind of a made up term,

266
00:13:50.779 --> 00:13:54.580 A:middle L:90%
um, to deliberately address uh itself as a kind

267
00:13:54.580 --> 00:13:58.460 A:middle L:90%
of opposite to close reading. So if we think

268
00:13:58.460 --> 00:14:01.309 A:middle L:90%
about the ways we generally make meaning with texts we

269
00:14:01.309 --> 00:14:05.370 A:middle L:90%
do so through close reading of the text. A

270
00:14:05.370 --> 00:14:11.190 A:middle L:90%
distance reading is supposed to be a complimentary procedure to

271
00:14:11.200 --> 00:14:15.320 A:middle L:90%
close reading. It doesn't negate the effects of a

272
00:14:15.320 --> 00:14:18.039 A:middle L:90%
close read, but it gives you another tool and

273
00:14:18.039 --> 00:14:20.120 A:middle L:90%
another way to do that. And the idea is

274
00:14:20.159 --> 00:14:22.320 A:middle L:90%
, if we stand back and instead of trying to

275
00:14:22.320 --> 00:14:26.220 A:middle L:90%
derive deep meaning from a single instance or a few

276
00:14:26.220 --> 00:14:28.059 A:middle L:90%
instances, what happens if we step back and look

277
00:14:28.059 --> 00:14:31.259 A:middle L:90%
at millions of instances? What can we learn from

278
00:14:31.259 --> 00:14:33.789 A:middle L:90%
those moments? And, of course, that's what

279
00:14:33.789 --> 00:14:37.700 A:middle L:90%
the digital availability of a lot of these textual materials

280
00:14:37.700 --> 00:14:39.639 A:middle L:90%
now give us the ability to do so, whether

281
00:14:39.639 --> 00:14:45.440 A:middle L:90%
it's the Google Books Project and the Ingraham um uh

282
00:14:45.450 --> 00:14:46.529 A:middle L:90%
, widget that got released and that we all like

283
00:14:46.529 --> 00:14:50.649 A:middle L:90%
to enjoy playing with or you can see patterns and

284
00:14:50.649 --> 00:14:52.559 A:middle L:90%
word adjacent seas across, uh, you know,

285
00:14:52.570 --> 00:14:56.799 A:middle L:90%
500 years or something like that in the Google Books

286
00:14:56.809 --> 00:15:01.299 A:middle L:90%
, uh, happy trust repository. So there's some

287
00:15:01.299 --> 00:15:03.850 A:middle L:90%
really exciting things that are going on in that area

288
00:15:07.039 --> 00:15:11.710 A:middle L:90%
. So our initial dataset Jeff and his team had

289
00:15:11.710 --> 00:15:16.779 A:middle L:90%
been working with some, um, science museums to

290
00:15:16.789 --> 00:15:18.169 A:middle L:90%
two in particular. One is the museum of Life

291
00:15:18.169 --> 00:15:22.700 A:middle L:90%
and Science in Raleigh Durham, North Carolina. It's

292
00:15:22.700 --> 00:15:26.360 A:middle L:90%
actually in Durham and the the Science Museum of Minnesota

293
00:15:26.370 --> 00:15:31.019 A:middle L:90%
in Saint Paul. And they had some money from

294
00:15:31.019 --> 00:15:35.750 A:middle L:90%
the Smells, the Institute for Museum and Library Services

295
00:15:35.139 --> 00:15:39.149 A:middle L:90%
, to do an analysis of both museums. Digital

296
00:15:39.149 --> 00:15:43.809 A:middle L:90%
facilitation strategies to see if they were contributing in a

297
00:15:43.820 --> 00:15:50.039 A:middle L:90%
measurable way to stem learning facilitation is in the two

298
00:15:50.039 --> 00:15:52.669 A:middle L:90%
informal learning, the informal learning world of museums and

299
00:15:52.669 --> 00:15:56.470 A:middle L:90%
science centres. What pedagogy is to folks like us

300
00:15:58.370 --> 00:16:00.559 A:middle L:90%
. These are the things that the museum staff due

301
00:16:00.559 --> 00:16:03.399 A:middle L:90%
to try to encourage, learning to happen in specific

302
00:16:03.399 --> 00:16:07.870 A:middle L:90%
ways. But because it's informal learning, they don't

303
00:16:07.870 --> 00:16:10.480 A:middle L:90%
have all of the tools that we have at our

304
00:16:10.480 --> 00:16:14.059 A:middle L:90%
disposal. They don't have learning objectives and lessons,

305
00:16:14.159 --> 00:16:17.440 A:middle L:90%
and they certainly don't have tests, right? No

306
00:16:17.440 --> 00:16:18.500 A:middle L:90%
one going to Science Museum is going to sit still

307
00:16:18.500 --> 00:16:22.340 A:middle L:90%
for that. So they were really interested in trying

308
00:16:22.340 --> 00:16:25.529 A:middle L:90%
to figure out if the things that they had been

309
00:16:25.529 --> 00:16:30.730 A:middle L:90%
doing an online discussion forums in blogs and message formats

310
00:16:30.730 --> 00:16:33.669 A:middle L:90%
like that on lately on Facebook and Twitter. Um

311
00:16:33.679 --> 00:16:36.440 A:middle L:90%
, if any of those things were helping them to

312
00:16:36.440 --> 00:16:41.429 A:middle L:90%
meet their institutional goals of encouraging more scientific deliberation in

313
00:16:41.429 --> 00:16:44.789 A:middle L:90%
the community, etcetera. And in order to do

314
00:16:44.789 --> 00:16:48.059 A:middle L:90%
that, they had gathered up a bunch of this

315
00:16:48.059 --> 00:16:52.039 A:middle L:90%
material, Um, really years and years worth of

316
00:16:52.039 --> 00:16:55.000 A:middle L:90%
it. And I'll show you some of the actual

317
00:16:55.000 --> 00:16:56.539 A:middle L:90%
data here in a minute. That's kind of fun

318
00:16:56.539 --> 00:17:00.419 A:middle L:90%
to see. Um, And they took a subset

319
00:17:00.419 --> 00:17:03.819 A:middle L:90%
of that that was humanly possible with well fed,

320
00:17:03.829 --> 00:17:07.069 A:middle L:90%
pizza driven graduate students. And they coded it by

321
00:17:07.069 --> 00:17:11.849 A:middle L:90%
hand, um, using some specific categories. And

322
00:17:11.849 --> 00:17:15.829 A:middle L:90%
they were looking for these moves that would indicate that

323
00:17:15.829 --> 00:17:19.089 A:middle L:90%
something was happening that looked like scientific reasoning. And

324
00:17:19.089 --> 00:17:25.369 A:middle L:90%
because these are posts initial posts and then follow up

325
00:17:25.369 --> 00:17:27.309 A:middle L:90%
comments, they happen in a timeline. They their

326
00:17:27.309 --> 00:17:32.000 A:middle L:90%
interactions that start and then proceed. Some of them

327
00:17:32.000 --> 00:17:36.119 A:middle L:90%
are still going on, But they have they develop

328
00:17:36.130 --> 00:17:37.890 A:middle L:90%
over time, and so that that time dimension was

329
00:17:37.890 --> 00:17:41.130 A:middle L:90%
something that they wanted to pay attention to. And

330
00:17:41.130 --> 00:17:45.269 A:middle L:90%
it resulted in these coding sheets that we had in

331
00:17:45.269 --> 00:17:47.690 A:middle L:90%
the on the walls of wide. And I think

332
00:17:47.690 --> 00:17:49.720 A:middle L:90%
when Quinn visited, actually, they're still hanging there

333
00:17:49.890 --> 00:17:53.069 A:middle L:90%
. They look kind of like these PCR tests like

334
00:17:53.079 --> 00:17:56.230 A:middle L:90%
we were doing DNA or something like that, but

335
00:17:56.269 --> 00:17:59.579 A:middle L:90%
we color coded the moves and looked at instances of

336
00:17:59.579 --> 00:18:02.950 A:middle L:90%
them and then try to stand back and see.

337
00:18:02.960 --> 00:18:04.549 A:middle L:90%
All right, is something happening in this in this

338
00:18:04.549 --> 00:18:11.269 A:middle L:90%
thread that trends towards more reasoning, like science scientists

339
00:18:11.269 --> 00:18:15.690 A:middle L:90%
might reason or etcetera, etcetera. And so I

340
00:18:15.700 --> 00:18:18.420 A:middle L:90%
just took a lot of time. And of course

341
00:18:18.420 --> 00:18:19.569 A:middle L:90%
, they could only look at a few posts in

342
00:18:19.579 --> 00:18:23.740 A:middle L:90%
much detail. And so their group came to our

343
00:18:23.740 --> 00:18:26.539 A:middle L:90%
research group and said, Maybe you can help us

344
00:18:26.539 --> 00:18:29.029 A:middle L:90%
with this, and we we saw in that some

345
00:18:29.029 --> 00:18:33.369 A:middle L:90%
interesting opportunities. That's, um Ah, take a

346
00:18:33.369 --> 00:18:37.829 A:middle L:90%
little bit of a look there. Um, here's

347
00:18:37.829 --> 00:18:41.579 A:middle L:90%
what we were obsessed with at the time. So

348
00:18:41.579 --> 00:18:44.529 A:middle L:90%
that's their research problem. Our research problem. It

349
00:18:44.529 --> 00:18:48.390 A:middle L:90%
was slightly different, and that is we were interested

350
00:18:48.400 --> 00:18:53.849 A:middle L:90%
in the possible use of graph theory. Techniques are

351
00:18:53.849 --> 00:19:00.309 A:middle L:90%
sometimes called network analytics to evaluate what was going on

352
00:19:00.369 --> 00:19:03.109 A:middle L:90%
in Texas and in large groups of text and corpus

353
00:19:03.109 --> 00:19:06.029 A:middle L:90%
is text corpus is. But what we were not

354
00:19:06.029 --> 00:19:08.630 A:middle L:90%
interested in is in the ways that, um,

355
00:19:08.640 --> 00:19:12.039 A:middle L:90%
a lot of people were using social network analysis,

356
00:19:12.079 --> 00:19:14.750 A:middle L:90%
um, to do that. And by that I

357
00:19:14.750 --> 00:19:18.519 A:middle L:90%
mean we weren't thinking about the nodes in the graph

358
00:19:18.519 --> 00:19:22.319 A:middle L:90%
the dots being people and the lines are the edges

359
00:19:22.319 --> 00:19:26.329 A:middle L:90%
of the graph being messages of one sort or another

360
00:19:26.339 --> 00:19:29.400 A:middle L:90%
. That's a common way to do it. What

361
00:19:29.400 --> 00:19:30.440 A:middle L:90%
we were thinking of is that the nodes are something

362
00:19:30.440 --> 00:19:34.549 A:middle L:90%
about the text, like pieces like maybe rhetorical moves

363
00:19:34.549 --> 00:19:38.109 A:middle L:90%
eventually or things that are indicators for those. And

364
00:19:38.109 --> 00:19:44.700 A:middle L:90%
the edges represent relationships among those things. So we

365
00:19:44.700 --> 00:19:47.099 A:middle L:90%
had the idea. I'll show you why we had

366
00:19:47.099 --> 00:19:51.210 A:middle L:90%
this idea. But we had not seen very many

367
00:19:52.039 --> 00:19:53.599 A:middle L:90%
, uh, precedents for this because our social science

368
00:19:53.599 --> 00:19:59.049 A:middle L:90%
colleagues were more interested in who are using social network

369
00:19:59.049 --> 00:20:02.029 A:middle L:90%
analysis, were more interested in analyzing what happens among

370
00:20:02.029 --> 00:20:06.470 A:middle L:90%
human agents. And we were really trying to map

371
00:20:06.480 --> 00:20:11.339 A:middle L:90%
these sort of trends in larger scale discourse to find

372
00:20:11.339 --> 00:20:14.000 A:middle L:90%
out what the whole group was doing. And that

373
00:20:14.000 --> 00:20:17.500 A:middle L:90%
was really the key difference for us, so you

374
00:20:17.500 --> 00:20:19.549 A:middle L:90%
can see where our interests started to align. These

375
00:20:19.549 --> 00:20:22.309 A:middle L:90%
were two research groups working independently up until this point

376
00:20:22.309 --> 00:20:25.130 A:middle L:90%
, until we talk to each other and they were

377
00:20:25.130 --> 00:20:26.349 A:middle L:90%
like, Oh, we can help each other.

378
00:20:27.740 --> 00:20:30.289 A:middle L:90%
So here's the chicken thread. I'll talk. I'll

379
00:20:30.289 --> 00:20:33.589 A:middle L:90%
show you what these things look like. So this

380
00:20:33.589 --> 00:20:37.380 A:middle L:90%
is a very typical exchange, only showed you one

381
00:20:37.380 --> 00:20:40.240 A:middle L:90%
part of it, Um, but it's very typical

382
00:20:40.240 --> 00:20:44.079 A:middle L:90%
of the whole thread. It consists of about 857

383
00:20:44.079 --> 00:20:45.720 A:middle L:90%
as of the other day, when I checked and

384
00:20:45.720 --> 00:20:51.089 A:middle L:90%
made the slide posts to a single blog post by

385
00:20:51.089 --> 00:20:55.910 A:middle L:90%
a museum person about backyard chicken raising, which is

386
00:20:55.910 --> 00:20:57.849 A:middle L:90%
a very big deal all over the country. Um

387
00:20:59.240 --> 00:21:02.380 A:middle L:90%
, and in this one, we see anonymous user

388
00:21:02.380 --> 00:21:07.690 A:middle L:90%
Patty asks a question. And then Dr Jackie Jacobs

389
00:21:07.690 --> 00:21:11.819 A:middle L:90%
, University of Minnesota Sota poultry specialist, answers the

390
00:21:11.819 --> 00:21:15.109 A:middle L:90%
question. And I have enormous respect after doing this

391
00:21:15.109 --> 00:21:18.869 A:middle L:90%
project for Dr Jackie Jacobs poultry specialists, because not

392
00:21:18.869 --> 00:21:22.690 A:middle L:90%
only does she know a lot about poultry, but

393
00:21:22.700 --> 00:21:26.059 A:middle L:90%
she is extremely patient, and she answers all of

394
00:21:26.059 --> 00:21:30.529 A:middle L:90%
these questions even when she sees them and variations of

395
00:21:30.529 --> 00:21:33.460 A:middle L:90%
them over and over and over again. Um,

396
00:21:33.940 --> 00:21:36.779 A:middle L:90%
and so she's a wonderful person. She's actually one

397
00:21:36.779 --> 00:21:40.170 A:middle L:90%
of the one of theirs, the science scientist affiliates

398
00:21:40.180 --> 00:21:44.349 A:middle L:90%
at the um at the Science Museum of Minnesota.

399
00:21:45.039 --> 00:21:47.200 A:middle L:90%
So she doesn't work at the museum, but she

400
00:21:47.200 --> 00:21:49.720 A:middle L:90%
has signed up and, uh, monitors the threads

401
00:21:49.720 --> 00:21:53.559 A:middle L:90%
that she's part of. So 857 of these posts

402
00:21:53.559 --> 00:21:56.380 A:middle L:90%
is a lot a ton of text there. This

403
00:21:56.380 --> 00:22:00.589 A:middle L:90%
is actually a rather short exchange, and it's a

404
00:22:00.589 --> 00:22:03.190 A:middle L:90%
short response by Jackie. It's also tends to be

405
00:22:03.190 --> 00:22:06.309 A:middle L:90%
on the shorter side of the kinds of questions or

406
00:22:06.309 --> 00:22:08.509 A:middle L:90%
stories that people tell about their chickens. So you'd

407
00:22:08.509 --> 00:22:11.750 A:middle L:90%
be surprised what people are very passionate about their chickens

408
00:22:11.759 --> 00:22:15.529 A:middle L:90%
. It started in August 3rd 2006, and it's

409
00:22:15.539 --> 00:22:18.140 A:middle L:90%
, um, as of May 5th, 2000 and

410
00:22:18.140 --> 00:22:19.789 A:middle L:90%
12, which is the last time we took a

411
00:22:19.789 --> 00:22:22.480 A:middle L:90%
big snapshot for the For the data set. It

412
00:22:22.480 --> 00:22:26.410 A:middle L:90%
had 857 posts. And the question of what we're

413
00:22:26.410 --> 00:22:30.440 A:middle L:90%
trying to ask across that whole thread is. Is

414
00:22:30.440 --> 00:22:36.529 A:middle L:90%
something happening here that that resolves to people learning to

415
00:22:36.529 --> 00:22:40.069 A:middle L:90%
do science? Or is this helping, uh,

416
00:22:40.079 --> 00:22:44.630 A:middle L:90%
the the Greater Minnesota Metro Minneapolis Metro area, Twin

417
00:22:44.630 --> 00:22:47.500 A:middle L:90%
Cities Metro area? Um, with scientific literacy?

418
00:22:47.500 --> 00:22:49.990 A:middle L:90%
That's the kind of question that that the museum directors

419
00:22:51.000 --> 00:22:52.269 A:middle L:90%
desperately wanted an answer to. And of course,

420
00:22:52.269 --> 00:22:55.490 A:middle L:90%
we couldn't give that broad of an answer. But

421
00:22:55.490 --> 00:22:57.299 A:middle L:90%
maybe we could scale that back and say, Can

422
00:22:57.299 --> 00:23:00.670 A:middle L:90%
we see science happening in this thread? Yeah.

423
00:23:02.839 --> 00:23:06.960 A:middle L:90%
So in order to kind of show you how we

424
00:23:06.960 --> 00:23:07.660 A:middle L:90%
went about what I'm going to talk about is how

425
00:23:07.660 --> 00:23:10.890 A:middle L:90%
we went about answering this. Um, the answer

426
00:23:10.890 --> 00:23:15.920 A:middle L:90%
is our human coders did see some evidence of some

427
00:23:15.930 --> 00:23:18.299 A:middle L:90%
interesting learning going on scientific learning going on in the

428
00:23:18.299 --> 00:23:22.309 A:middle L:90%
chicken threat. And they were pretty confident that it

429
00:23:22.309 --> 00:23:23.559 A:middle L:90%
was It was among all the threads that they looked

430
00:23:23.559 --> 00:23:26.259 A:middle L:90%
at. It was one that was they would hold

431
00:23:26.259 --> 00:23:29.410 A:middle L:90%
up as an example. And so we started with

432
00:23:29.420 --> 00:23:33.359 A:middle L:90%
what the human coders were already recognizing in that way

433
00:23:33.839 --> 00:23:34.769 A:middle L:90%
. But in order to get to a place where

434
00:23:34.769 --> 00:23:40.920 A:middle L:90%
we could evaluate that in a less supervised way using

435
00:23:40.920 --> 00:23:45.730 A:middle L:90%
the computer algorithms, we had do some theorizing about

436
00:23:45.730 --> 00:23:48.670 A:middle L:90%
what was going on and how we might do that

437
00:23:48.640 --> 00:23:52.960 A:middle L:90%
. So again, not networks of humans like this

438
00:23:52.960 --> 00:23:55.259 A:middle L:90%
. So if I were to to take here,

439
00:23:55.259 --> 00:23:59.569 A:middle L:90%
I picked a same same list or same blog.

440
00:23:59.700 --> 00:24:02.740 A:middle L:90%
But I picked a very short thread that has only

441
00:24:02.740 --> 00:24:03.710 A:middle L:90%
about 17 posts. I'm going to show you how

442
00:24:03.710 --> 00:24:07.559 A:middle L:90%
it grows over time. Um, the diagram here

443
00:24:07.940 --> 00:24:11.670 A:middle L:90%
gets built according to the way we think of most

444
00:24:11.670 --> 00:24:17.140 A:middle L:90%
socio grams or social social network analysis. So each

445
00:24:17.140 --> 00:24:18.569 A:middle L:90%
bubble is going to be a response. J.

446
00:24:18.569 --> 00:24:22.269 A:middle L:90%
Gordon made the original post, and we see some

447
00:24:22.269 --> 00:24:26.660 A:middle L:90%
arrows. It's a directed graph in that way responding

448
00:24:26.710 --> 00:24:30.349 A:middle L:90%
, indicating who's responding to who? All right,

449
00:24:30.440 --> 00:24:33.119 A:middle L:90%
So we C. J. Gordon, the museum

450
00:24:33.130 --> 00:24:37.740 A:middle L:90%
expert person and then all these other people start weighing

451
00:24:37.740 --> 00:24:42.480 A:middle L:90%
in pretty soon as soon as these first three posts

452
00:24:42.490 --> 00:24:48.029 A:middle L:90%
after the initial blog, Um, something starts to

453
00:24:48.029 --> 00:24:51.430 A:middle L:90%
happen that are human coders could recognize. That is

454
00:24:51.440 --> 00:24:53.029 A:middle L:90%
what we'll say is a matter of concern arose.

455
00:24:53.240 --> 00:24:56.339 A:middle L:90%
It wasn't a serious matter of concern. This is

456
00:24:56.339 --> 00:24:57.519 A:middle L:90%
not the chicken thread. This is a threat about

457
00:24:57.529 --> 00:25:03.829 A:middle L:90%
some scientists who found a giant squid and, among

458
00:25:03.829 --> 00:25:07.450 A:middle L:90%
other things, in the Giants in the story,

459
00:25:07.180 --> 00:25:14.259 A:middle L:90%
they couldn't resist tasting it. Oh, and the

460
00:25:14.269 --> 00:25:18.430 A:middle L:90%
article specialised, uh, speculates about why that was

461
00:25:18.440 --> 00:25:22.470 A:middle L:90%
. And this led to all manner of stories about

462
00:25:22.470 --> 00:25:23.460 A:middle L:90%
what it might mean to take some of this massive

463
00:25:23.460 --> 00:25:27.140 A:middle L:90%
creature into oneself, and and that they started telling

464
00:25:27.140 --> 00:25:32.599 A:middle L:90%
jokes, reincarnation jokes. So So the thing in

465
00:25:32.599 --> 00:25:34.279 A:middle L:90%
the gray circle is what we're interested in. There

466
00:25:34.289 --> 00:25:37.289 A:middle L:90%
is what's going on in this thread. So people

467
00:25:37.289 --> 00:25:41.859 A:middle L:90%
are just telling jokes about squid, right? Oh

468
00:25:42.240 --> 00:25:45.170 A:middle L:90%
, so I promise you some theories, So bear

469
00:25:45.170 --> 00:25:49.509 A:middle L:90%
with me. This is the piece of the computational

470
00:25:49.509 --> 00:25:53.140 A:middle L:90%
rhetoric side of things that, um, the rhetoric

471
00:25:53.140 --> 00:25:56.069 A:middle L:90%
shins are at the table, I think, in

472
00:25:56.069 --> 00:25:59.809 A:middle L:90%
most cases to provide. And that is So what

473
00:25:59.819 --> 00:26:02.369 A:middle L:90%
do we What would we expect to go on in

474
00:26:02.369 --> 00:26:07.549 A:middle L:90%
a situation like this? Why is it that we

475
00:26:07.549 --> 00:26:10.670 A:middle L:90%
think that we can train a computer or teach a

476
00:26:10.670 --> 00:26:15.460 A:middle L:90%
computer to recognize these things at all? So there's

477
00:26:15.460 --> 00:26:18.529 A:middle L:90%
this guy, Russian, um, theorist named Bactine

478
00:26:18.539 --> 00:26:21.130 A:middle L:90%
. Some of you all have read him, and

479
00:26:21.130 --> 00:26:22.160 A:middle L:90%
some of you will, if you haven't already.

480
00:26:22.539 --> 00:26:26.670 A:middle L:90%
Others won't know who this person is. Um,

481
00:26:27.740 --> 00:26:32.269 A:middle L:90%
but among other things, he's in this particular passage

482
00:26:32.279 --> 00:26:33.890 A:middle L:90%
. It's a famous essay he has called the problem

483
00:26:33.890 --> 00:26:37.119 A:middle L:90%
of speech genres and he says some things in this

484
00:26:37.119 --> 00:26:41.309 A:middle L:90%
passage that, um resolved to what we think of

485
00:26:41.309 --> 00:26:47.420 A:middle L:90%
in writing studies as threshold ideas about genre. What

486
00:26:47.420 --> 00:26:49.700 A:middle L:90%
I mean by that is the way writing studies people

487
00:26:49.700 --> 00:26:53.170 A:middle L:90%
use the word genre and the way normal people who

488
00:26:53.170 --> 00:26:56.059 A:middle L:90%
aren't corrupted like us use the word genre completely different

489
00:26:56.440 --> 00:26:59.500 A:middle L:90%
. And once you cross the threshold, it's hard

490
00:26:59.500 --> 00:27:03.799 A:middle L:90%
to see the other way. Um, our view

491
00:27:03.799 --> 00:27:07.170 A:middle L:90%
of genre in writing studies is something like this.

492
00:27:07.170 --> 00:27:08.849 A:middle L:90%
So I have all these writing states. People here

493
00:27:08.849 --> 00:27:11.299 A:middle L:90%
will throw something at me if I get it wrong

494
00:27:12.039 --> 00:27:15.710 A:middle L:90%
. Um, we see the regularities that arise in

495
00:27:15.710 --> 00:27:19.220 A:middle L:90%
textual forms that we call genres, um, to

496
00:27:19.220 --> 00:27:23.849 A:middle L:90%
be the result of repeated or habituated action. And

497
00:27:23.849 --> 00:27:29.769 A:middle L:90%
these habituated actions are usually responses to recurring social situations

498
00:27:30.539 --> 00:27:34.470 A:middle L:90%
. Where genre conventions these textual forms are most stable

499
00:27:34.480 --> 00:27:38.680 A:middle L:90%
. They are generally reinforced by power. Institutional power

500
00:27:38.690 --> 00:27:44.319 A:middle L:90%
law. Right? Um, sometimes that power is

501
00:27:44.359 --> 00:27:47.650 A:middle L:90%
implicit. Sometimes it's quite explicit. It doesn't get

502
00:27:47.650 --> 00:27:48.690 A:middle L:90%
any more explicit than a form that you fill out

503
00:27:48.690 --> 00:27:53.029 A:middle L:90%
their imposing the structure so that they can normalize the

504
00:27:53.029 --> 00:27:56.660 A:middle L:90%
response right. It's also quite mundane and boring in

505
00:27:56.660 --> 00:28:00.480 A:middle L:90%
on one level. So nobody is imposing genres,

506
00:28:00.480 --> 00:28:04.140 A:middle L:90%
too do much other than maintain social order. And

507
00:28:04.140 --> 00:28:07.569 A:middle L:90%
sometimes that social order is, you know, itself

508
00:28:07.569 --> 00:28:11.079 A:middle L:90%
quite trivial. Sometimes it's not. Um, so

509
00:28:11.079 --> 00:28:17.000 A:middle L:90%
Bactine noticing this is, uh is a kind of

510
00:28:17.000 --> 00:28:21.819 A:middle L:90%
a theory theoretical precursor to this understanding and writing studies

511
00:28:21.819 --> 00:28:23.680 A:middle L:90%
. That is the folks who develop this understanding of

512
00:28:23.680 --> 00:28:27.910 A:middle L:90%
genre and writing. Studies have been reading Bactine among

513
00:28:27.910 --> 00:28:33.829 A:middle L:90%
other people, especially Carolyn Miller and this is the

514
00:28:33.829 --> 00:28:34.740 A:middle L:90%
idea from which that flow. So if we go

515
00:28:34.740 --> 00:28:38.220 A:middle L:90%
back to that idea that Martinez talking about, I'll

516
00:28:38.220 --> 00:28:42.670 A:middle L:90%
just point to the orange text here. Here's what

517
00:28:42.670 --> 00:28:48.059 A:middle L:90%
back Tina saying If I could paraphrase a little bit

518
00:28:48.640 --> 00:28:51.009 A:middle L:90%
, he has an idea that in order to be

519
00:28:51.009 --> 00:28:56.130 A:middle L:90%
understood in order to create mutual understanding among speakers,

520
00:28:56.140 --> 00:28:57.769 A:middle L:90%
we have to repeat ourselves, and we not only

521
00:28:57.769 --> 00:29:00.930 A:middle L:90%
have to repeat what we're saying in the immediate moment

522
00:29:00.940 --> 00:29:03.029 A:middle L:90%
, we have to say what has been said before

523
00:29:03.029 --> 00:29:06.809 A:middle L:90%
in situations like what we've like the one that we're

524
00:29:06.809 --> 00:29:10.690 A:middle L:90%
in. That's how you get recognized to be doing

525
00:29:10.700 --> 00:29:15.410 A:middle L:90%
whatever it is you're up to. And that means

526
00:29:15.410 --> 00:29:17.890 A:middle L:90%
that every word you say has been said before,

527
00:29:17.900 --> 00:29:19.619 A:middle L:90%
except for the very few that aren't and it's surrounded

528
00:29:19.619 --> 00:29:22.130 A:middle L:90%
by lots of things that have been said before.

529
00:29:22.140 --> 00:29:27.529 A:middle L:90%
That's how you are able to disambiguate. What this

530
00:29:27.529 --> 00:29:32.069 A:middle L:90%
means is, um, it's a little dismaying,

531
00:29:32.079 --> 00:29:33.490 A:middle L:90%
I ought to say for especially, we have creative

532
00:29:33.490 --> 00:29:37.119 A:middle L:90%
writers in the room because what it means is that

533
00:29:37.160 --> 00:29:42.539 A:middle L:90%
originality is the kind of exception, whereas re utterance

534
00:29:42.549 --> 00:29:45.910 A:middle L:90%
or reuse is the rule and on a on a

535
00:29:45.910 --> 00:29:48.869 A:middle L:90%
very mundane level, especially at the at the,

536
00:29:48.880 --> 00:29:53.319 A:middle L:90%
uh at the level of everyday interaction everyday discourse.

537
00:29:53.329 --> 00:29:59.140 A:middle L:90%
It's pretty un controversial to think about, but it

538
00:29:59.140 --> 00:30:02.230 A:middle L:90%
does do some interesting things to unsettle how we make

539
00:30:02.230 --> 00:30:03.400 A:middle L:90%
meaning and what it means to write something down.

540
00:30:03.400 --> 00:30:06.779 A:middle L:90%
We have to be recognized first, so we have

541
00:30:06.779 --> 00:30:11.690 A:middle L:90%
to re utter something. Um, the orange text

542
00:30:11.690 --> 00:30:18.250 A:middle L:90%
says this means that any utterance proves to be something

543
00:30:18.250 --> 00:30:21.450 A:middle L:90%
that is best understood, not considered in isolation with

544
00:30:21.450 --> 00:30:23.339 A:middle L:90%
respect to its author. That is not attributed only

545
00:30:23.339 --> 00:30:29.789 A:middle L:90%
to Quinn saying this was a really broody chicken.

546
00:30:29.789 --> 00:30:32.289 A:middle L:90%
Now what am I supposed to do? Jackie Jacobs

547
00:30:33.140 --> 00:30:37.339 A:middle L:90%
, but in response to other kinds of utterances of

548
00:30:37.339 --> 00:30:38.910 A:middle L:90%
that type as a link in a chain of speech

549
00:30:38.910 --> 00:30:44.579 A:middle L:90%
communication with respect to other related utterances so that all

550
00:30:44.579 --> 00:30:48.460 A:middle L:90%
those things share some things in common. And again

551
00:30:48.839 --> 00:30:52.730 A:middle L:90%
, um, this isn't terribly, um, controversial

552
00:30:52.730 --> 00:30:56.890 A:middle L:90%
if we step back and think about scientific discourse as

553
00:30:56.890 --> 00:30:59.150 A:middle L:90%
a whole. So if we think about the genre

554
00:30:59.160 --> 00:31:02.880 A:middle L:90%
and the relative stability of the article of the scientific

555
00:31:02.880 --> 00:31:06.150 A:middle L:90%
article in Science, um, this makes a ton

556
00:31:06.150 --> 00:31:08.940 A:middle L:90%
of sense. We want We want scientific articles,

557
00:31:08.960 --> 00:31:14.000 A:middle L:90%
particularly within disciplines, to be very stable. So

558
00:31:14.000 --> 00:31:15.900 A:middle L:90%
that we can use them in particular predictable ways.

559
00:31:15.900 --> 00:31:18.259 A:middle L:90%
We want to be able to We expect something of

560
00:31:18.259 --> 00:31:21.470 A:middle L:90%
the abstract. We expect something of the methods we

561
00:31:21.470 --> 00:31:23.460 A:middle L:90%
expect, and those things have relationships to one another

562
00:31:25.339 --> 00:31:27.420 A:middle L:90%
. And those get reiterated every time you publish one

563
00:31:27.430 --> 00:31:30.579 A:middle L:90%
. There's in that case institutional authority, keeping that

564
00:31:30.579 --> 00:31:34.849 A:middle L:90%
genre together, right, and you can take little

565
00:31:34.859 --> 00:31:37.950 A:middle L:90%
little deviations from it. But if you go too

566
00:31:37.950 --> 00:31:41.940 A:middle L:90%
far, you get smacked. If you go way

567
00:31:41.940 --> 00:31:42.829 A:middle L:90%
too far, you just it'll be okay. You're

568
00:31:42.829 --> 00:31:45.720 A:middle L:90%
just not writing that genre. You're doing something else

569
00:31:48.039 --> 00:31:52.190 A:middle L:90%
, all right, so that's the idea, right

570
00:31:52.190 --> 00:31:53.940 A:middle L:90%
? And that's an abstraction that I've sort of lived

571
00:31:53.940 --> 00:31:57.180 A:middle L:90%
with and came come to grips with. And since

572
00:31:57.180 --> 00:32:00.420 A:middle L:90%
I was a graduate student, since I was dismayed

573
00:32:00.420 --> 00:32:02.829 A:middle L:90%
that not everything coming out of my mouth was my

574
00:32:02.829 --> 00:32:05.269 A:middle L:90%
idea. But that was part of this sort of

575
00:32:05.269 --> 00:32:08.809 A:middle L:90%
social fabric that I was in. But it was

576
00:32:08.809 --> 00:32:12.819 A:middle L:90%
that link in the chain of speech communication with respect

577
00:32:12.819 --> 00:32:15.099 A:middle L:90%
to other related utterances that really jumped out when we

578
00:32:15.099 --> 00:32:19.849 A:middle L:90%
started rethinking what we were doing were like That means

579
00:32:19.849 --> 00:32:22.549 A:middle L:90%
that every text is a graph. It's a network

580
00:32:23.539 --> 00:32:28.200 A:middle L:90%
, but not each text within itself. But the

581
00:32:28.200 --> 00:32:31.660 A:middle L:90%
text pieces in one utterance and in another utterance and

582
00:32:31.660 --> 00:32:34.740 A:middle L:90%
in another utterance and another utterance. That's what he's

583
00:32:34.740 --> 00:32:37.410 A:middle L:90%
telling us. That's the best way to understand them

584
00:32:37.420 --> 00:32:39.130 A:middle L:90%
. And I'm like, Well, maybe, But

585
00:32:39.130 --> 00:32:46.859 A:middle L:90%
let's maybe let's try it. So let's go back

586
00:32:46.859 --> 00:32:50.259 A:middle L:90%
to our thread and I'll we'll get to where we

587
00:32:50.269 --> 00:32:52.960 A:middle L:90%
eventually got to. We're now a little bit further

588
00:32:52.960 --> 00:32:54.769 A:middle L:90%
down the line in the giant squid thread, and

589
00:32:54.769 --> 00:32:59.140 A:middle L:90%
we could see another concern arises, which is.

590
00:32:59.150 --> 00:33:01.470 A:middle L:90%
People keep telling reincarnation jokes on the right hand side

591
00:33:01.470 --> 00:33:06.609 A:middle L:90%
there, and people like what? That's the next

592
00:33:06.609 --> 00:33:12.480 A:middle L:90%
one, but it predominates. More reincarnation jokes.

593
00:33:12.539 --> 00:33:14.650 A:middle L:90%
A few people try to break in, but by

594
00:33:14.650 --> 00:33:21.000 A:middle L:90%
that time there is momentum. One guy, the

595
00:33:21.000 --> 00:33:24.700 A:middle L:90%
guy, The thing at the top. Mhm Masonic

596
00:33:24.710 --> 00:33:30.549 A:middle L:90%
co toothless. It's like the next highest size squid

597
00:33:30.559 --> 00:33:32.670 A:middle L:90%
. Not quite giant but super big. I don't

598
00:33:32.670 --> 00:33:36.410 A:middle L:90%
know what the size squid it is, but so

599
00:33:36.410 --> 00:33:37.859 A:middle L:90%
this was This is one of the only on topic

600
00:33:37.859 --> 00:33:42.119 A:middle L:90%
posts we might say that comes in and it says

601
00:33:42.119 --> 00:33:44.259 A:middle L:90%
, Hey, I work at this other science museum

602
00:33:44.640 --> 00:33:46.789 A:middle L:90%
. We have a video of some people catching a

603
00:33:46.789 --> 00:33:52.099 A:middle L:90%
squid like this I think it's the Humboldt Bay squid

604
00:33:52.099 --> 00:33:53.359 A:middle L:90%
. If you've ever seen one of those guys,

605
00:33:53.839 --> 00:34:00.380 A:middle L:90%
um, they ignore him, they just keep on

606
00:34:00.380 --> 00:34:05.650 A:middle L:90%
with themselves. Yes. So what we what we

607
00:34:05.650 --> 00:34:08.260 A:middle L:90%
wanted to try to do is could we again,

608
00:34:08.260 --> 00:34:10.860 A:middle L:90%
this isn't hard for a human to map. We

609
00:34:10.860 --> 00:34:14.019 A:middle L:90%
can kind of you could read it down and kind

610
00:34:14.019 --> 00:34:15.820 A:middle L:90%
of see what's going on. And what we wanted

611
00:34:15.820 --> 00:34:21.090 A:middle L:90%
to see is could we teach the computer to see

612
00:34:21.280 --> 00:34:25.239 A:middle L:90%
something, uh, to teach the computer to find

613
00:34:25.239 --> 00:34:29.239 A:middle L:90%
the grey circles? That is what What? What

614
00:34:29.239 --> 00:34:31.039 A:middle L:90%
is it about those bubbles in the gray circle on

615
00:34:31.039 --> 00:34:32.409 A:middle L:90%
the right and the grace work on the left?

616
00:34:32.409 --> 00:34:36.670 A:middle L:90%
That makes us put them in those circles. Um

617
00:34:37.239 --> 00:34:39.110 A:middle L:90%
, and And how can we How can we see

618
00:34:39.110 --> 00:34:44.139 A:middle L:90%
that? So this is where we turn to the

619
00:34:44.139 --> 00:34:47.250 A:middle L:90%
idea of some network metrics. This is a very

620
00:34:47.250 --> 00:34:52.429 A:middle L:90%
simple picture of a network, and I'm gonna probably

621
00:34:52.429 --> 00:34:55.070 A:middle L:90%
offend the math people in the room by kind of

622
00:34:55.070 --> 00:34:58.710 A:middle L:90%
going through and talking a little bit about what we

623
00:34:58.710 --> 00:35:00.610 A:middle L:90%
mean when we evaluate a graph. So a graph

624
00:35:00.610 --> 00:35:05.599 A:middle L:90%
has a computational object because it's a well understood structure

625
00:35:05.599 --> 00:35:07.369 A:middle L:90%
that we can turn into a set of numbers,

626
00:35:07.369 --> 00:35:10.650 A:middle L:90%
and we can perform a set of mathematical operations to

627
00:35:10.650 --> 00:35:15.929 A:middle L:90%
derive and also to analyse. And it's actually not

628
00:35:15.929 --> 00:35:19.889 A:middle L:90%
terribly complex, and it's really interesting once you know

629
00:35:19.900 --> 00:35:22.190 A:middle L:90%
how it works. So if every circle in the

630
00:35:22.190 --> 00:35:25.239 A:middle L:90%
graph is what we call a node and every line

631
00:35:25.239 --> 00:35:28.440 A:middle L:90%
is an edge, one of the ways that we

632
00:35:28.440 --> 00:35:31.860 A:middle L:90%
can start to evaluate what the graph structure is is

633
00:35:32.039 --> 00:35:37.500 A:middle L:90%
the amount of connections that anyone node has, and

634
00:35:37.500 --> 00:35:39.960 A:middle L:90%
that's called its degree. So if you look at

635
00:35:39.960 --> 00:35:42.969 A:middle L:90%
the note in the middle of the graph, it

636
00:35:42.969 --> 00:35:46.539 A:middle L:90%
has a degree of six, right, And this

637
00:35:46.539 --> 00:35:50.619 A:middle L:90%
is a non directed graph, meaning that we're not

638
00:35:50.630 --> 00:35:54.070 A:middle L:90%
caring about the the lines having arrows going in or

639
00:35:54.070 --> 00:35:58.079 A:middle L:90%
out. But in some cases, we might be

640
00:35:58.079 --> 00:36:00.510 A:middle L:90%
like if we're setting up a communication network, we

641
00:36:00.510 --> 00:36:04.119 A:middle L:90%
might be interested in whether the messages traveling from me

642
00:36:04.119 --> 00:36:06.420 A:middle L:90%
to you or from you to be, and in

643
00:36:06.420 --> 00:36:07.960 A:middle L:90%
that case, we would measure two kinds of degree

644
00:36:07.969 --> 00:36:13.039 A:middle L:90%
in degree and out degree. But in each case

645
00:36:13.039 --> 00:36:15.130 A:middle L:90%
, we reflect that value as a property of the

646
00:36:15.130 --> 00:36:19.179 A:middle L:90%
node. The node in the middle has degree of

647
00:36:19.179 --> 00:36:22.239 A:middle L:90%
six and on the edges degree of three, right

648
00:36:23.329 --> 00:36:27.840 A:middle L:90%
? Nothing has a has a degree of less than

649
00:36:27.840 --> 00:36:30.739 A:middle L:90%
two, so that's one way to think about it

650
00:36:30.750 --> 00:36:34.039 A:middle L:90%
, the other the other now. So let me

651
00:36:34.039 --> 00:36:37.519 A:middle L:90%
point out something that's interesting about that. That isn't

652
00:36:37.530 --> 00:36:40.510 A:middle L:90%
irrespective of the shape of the network. So we

653
00:36:40.510 --> 00:36:43.980 A:middle L:90%
could move. We could turn things around, and

654
00:36:43.980 --> 00:36:46.710 A:middle L:90%
we could. I could mash it up and draw

655
00:36:46.710 --> 00:36:51.139 A:middle L:90%
it slightly differently and still maintain the degree values here

656
00:36:51.329 --> 00:36:54.280 A:middle L:90%
. But there's something else unique about the node that

657
00:36:54.280 --> 00:36:58.900 A:middle L:90%
says D six in the middle, and that is

658
00:36:59.139 --> 00:37:01.039 A:middle L:90%
its position in the graph relative to the other.

659
00:37:01.429 --> 00:37:05.489 A:middle L:90%
Other notes. It occupies a position that we call

660
00:37:05.489 --> 00:37:08.670 A:middle L:90%
the brokerage position, and that's because it sits between

661
00:37:08.670 --> 00:37:12.510 A:middle L:90%
one group of nodes on the left and another group

662
00:37:12.510 --> 00:37:15.559 A:middle L:90%
of nodes on the right. It's the middle man

663
00:37:15.739 --> 00:37:19.199 A:middle L:90%
, right? It's a powerful note. Because of

664
00:37:19.199 --> 00:37:22.059 A:middle L:90%
that, in some kinds of networks, right,

665
00:37:22.170 --> 00:37:27.170 A:middle L:90%
it can become the switchboard by which messages traveling from

666
00:37:27.170 --> 00:37:29.869 A:middle L:90%
one side must pass to the other. So if

667
00:37:29.869 --> 00:37:32.760 A:middle L:90%
you own that node, you could charge for traffic

668
00:37:32.760 --> 00:37:36.539 A:middle L:90%
across it. It's like a little bridge. Um

669
00:37:37.130 --> 00:37:38.489 A:middle L:90%
, there are whole countries that have built their economy

670
00:37:38.489 --> 00:37:44.059 A:middle L:90%
on being that node. Uh, and, uh

671
00:37:44.070 --> 00:37:45.840 A:middle L:90%
, by the same token, it has a little

672
00:37:45.840 --> 00:37:50.090 A:middle L:90%
bit of a privilege if we think about traffic over

673
00:37:50.090 --> 00:37:52.650 A:middle L:90%
that network. So everything that happens over the network

674
00:37:52.650 --> 00:37:54.639 A:middle L:90%
, if it crosses from one side to the other

675
00:37:54.639 --> 00:38:00.409 A:middle L:90%
, has to go through No d right. So

676
00:38:00.420 --> 00:38:01.849 A:middle L:90%
that means that we have a property for that too

677
00:38:01.849 --> 00:38:05.309 A:middle L:90%
. That means that no d has something called high

678
00:38:05.309 --> 00:38:08.239 A:middle L:90%
between the centrality. So those are just two of

679
00:38:08.239 --> 00:38:12.059 A:middle L:90%
the network metrics that we have, and they are

680
00:38:12.070 --> 00:38:15.909 A:middle L:90%
not invented by me. They're part of a section

681
00:38:15.920 --> 00:38:20.019 A:middle L:90%
of mathematics called graph theory. And hundreds of years

682
00:38:20.019 --> 00:38:21.869 A:middle L:90%
ago, people started working on all of this,

683
00:38:21.869 --> 00:38:24.159 A:middle L:90%
and they not only, um, taught us ways

684
00:38:24.159 --> 00:38:29.440 A:middle L:90%
to give us some metrics for understanding how network structure

685
00:38:29.440 --> 00:38:31.119 A:middle L:90%
looks in a static picture like this. But they

686
00:38:31.119 --> 00:38:35.659 A:middle L:90%
also have worked on networks that grow and change over

687
00:38:35.659 --> 00:38:37.969 A:middle L:90%
time so that you can model, for example,

688
00:38:38.070 --> 00:38:43.260 A:middle L:90%
what the impact on a network will be if you

689
00:38:43.260 --> 00:38:45.789 A:middle L:90%
add X more nodes and where do you need to

690
00:38:45.789 --> 00:38:49.369 A:middle L:90%
add edges now? We're all grateful for this because

691
00:38:49.380 --> 00:38:52.219 A:middle L:90%
otherwise our cable TV would work a lot less well

692
00:38:52.219 --> 00:38:54.829 A:middle L:90%
if people weren't able to analyze the structure of the

693
00:38:54.829 --> 00:38:58.030 A:middle L:90%
network and find out where you need to put repeaters

694
00:38:58.320 --> 00:39:00.829 A:middle L:90%
and where they need to run more lines and all

695
00:39:00.829 --> 00:39:02.699 A:middle L:90%
that kind of stuff. Um, it also has

696
00:39:04.070 --> 00:39:06.849 A:middle L:90%
all manner of other kinds of things. Like if

697
00:39:06.849 --> 00:39:09.900 A:middle L:90%
you imagine a big matrix like this with a lot

698
00:39:09.900 --> 00:39:15.599 A:middle L:90%
less regularized space between nodes, et cetera, You

699
00:39:15.599 --> 00:39:19.719 A:middle L:90%
get algorithms, for example, for filtration. Like

700
00:39:19.719 --> 00:39:22.719 A:middle L:90%
, how long does it take for a particular soil

701
00:39:22.719 --> 00:39:24.539 A:middle L:90%
density for water to filter through it? So that

702
00:39:24.539 --> 00:39:28.179 A:middle L:90%
tells you how much water you need to put on

703
00:39:28.179 --> 00:39:30.539 A:middle L:90%
certain kinds of soils in order to achieve irrigation.

704
00:39:30.550 --> 00:39:35.489 A:middle L:90%
So they're all manner of practical reasons to think about

705
00:39:35.500 --> 00:39:39.150 A:middle L:90%
networks and lattices and matrices and to apply these analytics

706
00:39:39.150 --> 00:39:45.380 A:middle L:90%
to them. And we saw some potential value because

707
00:39:45.380 --> 00:39:49.010 A:middle L:90%
of the back Tinian theory conjecture of doing that with

708
00:39:49.010 --> 00:39:54.809 A:middle L:90%
text. So here's the squid thread. This is

709
00:39:54.809 --> 00:39:57.949 A:middle L:90%
something. This is a picture of something called topic

710
00:39:57.949 --> 00:40:01.380 A:middle L:90%
mapping, using a really simple, um, set

711
00:40:01.389 --> 00:40:06.389 A:middle L:90%
of natural language processing techniques without working too hard,

712
00:40:06.539 --> 00:40:10.230 A:middle L:90%
we strip away, um, all but the stems

713
00:40:10.230 --> 00:40:15.280 A:middle L:90%
of words. And then we look at, um

714
00:40:15.289 --> 00:40:19.030 A:middle L:90%
how often they appear and their adjacent seas to other

715
00:40:19.030 --> 00:40:22.199 A:middle L:90%
words. And what we can do is we can

716
00:40:22.199 --> 00:40:25.590 A:middle L:90%
start to see what topics characterize what's being talked about

717
00:40:25.599 --> 00:40:29.929 A:middle L:90%
. So without any human reader looking at that thread

718
00:40:29.929 --> 00:40:31.659 A:middle L:90%
I just showed you, we can say with pretty

719
00:40:31.670 --> 00:40:37.659 A:middle L:90%
pretty good confidence that this is a threat about squid

720
00:40:37.309 --> 00:40:40.739 A:middle L:90%
. And if we wanted to search for that,

721
00:40:42.210 --> 00:40:44.130 A:middle L:90%
this is a way that we could do it.

722
00:40:44.610 --> 00:40:47.829 A:middle L:90%
Some of you might recognize between the centrality and Eigen

723
00:40:47.829 --> 00:40:52.940 A:middle L:90%
vector centrality associated with another popular algorithm that most of

724
00:40:52.940 --> 00:40:54.730 A:middle L:90%
us use all the time, which is the Google

725
00:40:54.730 --> 00:40:58.170 A:middle L:90%
page rank algorithm. That's the way we return.

726
00:40:58.170 --> 00:41:00.250 A:middle L:90%
Valid search results, um, is that they evaluate

727
00:41:00.480 --> 00:41:05.239 A:middle L:90%
the structure of networks um, caused by hyper linking

728
00:41:05.809 --> 00:41:07.809 A:middle L:90%
, and that helps them return to us a little

729
00:41:07.809 --> 00:41:12.039 A:middle L:90%
bit more valid Search results when we search for specific

730
00:41:12.039 --> 00:41:15.099 A:middle L:90%
terms connected to or on a particular page. So

731
00:41:15.099 --> 00:41:17.980 A:middle L:90%
they're doing something like this. Um, but if

732
00:41:17.980 --> 00:41:22.110 A:middle L:90%
you look just below squid, something else was happening

733
00:41:22.110 --> 00:41:25.280 A:middle L:90%
that we were really excited about. Now, as

734
00:41:25.280 --> 00:41:30.530 A:middle L:90%
reparations, we were were promiscuous and in our use

735
00:41:30.530 --> 00:41:31.449 A:middle L:90%
of these methods will use all of them and we

736
00:41:31.449 --> 00:41:35.550 A:middle L:90%
don't know what we're really doing. Um, and

737
00:41:35.550 --> 00:41:37.510 A:middle L:90%
one of the things we were doing is leaving words

738
00:41:37.510 --> 00:41:42.420 A:middle L:90%
in that typical stop word lists, uh, would

739
00:41:42.420 --> 00:41:49.280 A:middle L:90%
throw out like words like if squid if and think

740
00:41:49.280 --> 00:41:54.199 A:middle L:90%
were interesting to us because those were indicators that something

741
00:41:54.210 --> 00:41:57.320 A:middle L:90%
interesting was going on. And in this case,

742
00:41:57.710 --> 00:42:00.090 A:middle L:90%
um, when you go back and you see people

743
00:42:00.090 --> 00:42:04.309 A:middle L:90%
framing conditional statements with if and saying think a lot

744
00:42:04.309 --> 00:42:07.530 A:middle L:90%
like I think, um, they are indicators of

745
00:42:07.530 --> 00:42:08.909 A:middle L:90%
some hedging that was going on. But it was

746
00:42:08.920 --> 00:42:14.699 A:middle L:90%
interpersonal hedging, right? They were sort of making

747
00:42:14.699 --> 00:42:19.219 A:middle L:90%
jokes around this idea of the thing of the eating

748
00:42:19.219 --> 00:42:22.110 A:middle L:90%
squid and what what the Squid's power could do to

749
00:42:22.110 --> 00:42:23.090 A:middle L:90%
the scientists. But this gave us an idea.

750
00:42:23.090 --> 00:42:25.789 A:middle L:90%
It was completely like, Oh, I wonder if

751
00:42:25.789 --> 00:42:29.380 A:middle L:90%
we could find other examples of this because hedging behavior

752
00:42:29.380 --> 00:42:30.960 A:middle L:90%
is something that we're actually interested in. So we

753
00:42:30.960 --> 00:42:35.809 A:middle L:90%
ran the same thing on the chicken thread, and

754
00:42:35.820 --> 00:42:38.489 A:middle L:90%
we zeroed in on words that we thought had a

755
00:42:38.489 --> 00:42:43.659 A:middle L:90%
high correspondence to these hedge moves. Now, when

756
00:42:43.659 --> 00:42:46.090 A:middle L:90%
we say heads, we mean things that scientists do

757
00:42:46.119 --> 00:42:52.389 A:middle L:90%
to to deliberately, uh, limit the the audacity

758
00:42:52.389 --> 00:42:54.800 A:middle L:90%
of their claims when the evidence is less than convincing

759
00:42:54.800 --> 00:42:57.510 A:middle L:90%
, right? It's, uh it's what you're you're

760
00:42:57.510 --> 00:42:59.849 A:middle L:90%
being careful about what we say. I'm doing a

761
00:42:59.849 --> 00:43:01.420 A:middle L:90%
lot of it right now. Um, so in

762
00:43:01.420 --> 00:43:05.980 A:middle L:90%
this case, we we were looking at some interesting

763
00:43:05.980 --> 00:43:09.510 A:middle L:90%
patterns that we saw in the degree distribution. I'm

764
00:43:09.510 --> 00:43:14.210 A:middle L:90%
sorry. Only so these are just the links going

765
00:43:14.210 --> 00:43:19.090 A:middle L:90%
in in our network diagram turning every word in the

766
00:43:19.090 --> 00:43:22.920 A:middle L:90%
text into a node and using, um And where'd

767
00:43:22.920 --> 00:43:27.030 A:middle L:90%
adjacency is to see which words were most connected to

768
00:43:27.030 --> 00:43:30.329 A:middle L:90%
other words using Just agree. We saw one thing

769
00:43:30.329 --> 00:43:31.690 A:middle L:90%
that we expected to see from based on other research

770
00:43:31.690 --> 00:43:35.840 A:middle L:90%
. Is that all the way at the top and

771
00:43:35.840 --> 00:43:38.119 A:middle L:90%
something that looks a lot like a power law distribution

772
00:43:38.119 --> 00:43:40.730 A:middle L:90%
for those math people in the room is what we

773
00:43:40.730 --> 00:43:45.460 A:middle L:90%
would expect to see is the topic word. And

774
00:43:45.460 --> 00:43:46.519 A:middle L:90%
that way. Up the top 7. 50.

775
00:43:47.800 --> 00:43:51.429 A:middle L:90%
All right, I'm sorry. Down here is the

776
00:43:51.429 --> 00:43:55.210 A:middle L:90%
word egg. So that wasn't too surprising in the

777
00:43:55.219 --> 00:43:59.099 A:middle L:90%
in this thread about chickens and eggs, that egg

778
00:43:59.110 --> 00:44:02.960 A:middle L:90%
is the outlier word. But it was relatively surprising

779
00:44:02.960 --> 00:44:08.409 A:middle L:90%
to see clustered here with, um, all of

780
00:44:08.409 --> 00:44:13.920 A:middle L:90%
these modal verbs would should may, Will, Might

781
00:44:13.929 --> 00:44:16.320 A:middle L:90%
, can and not too far away from all of

782
00:44:16.320 --> 00:44:21.190 A:middle L:90%
those were other conditions which I think was there again

783
00:44:21.190 --> 00:44:23.369 A:middle L:90%
. Like I think if you do this, you

784
00:44:23.369 --> 00:44:27.809 A:middle L:90%
might And we're like, Oh, this is good

785
00:44:27.809 --> 00:44:34.239 A:middle L:90%
. This is good. So when we come back

786
00:44:34.239 --> 00:44:37.119 A:middle L:90%
to seeing rhetorical patterns, we were like, Well

787
00:44:37.119 --> 00:44:40.900 A:middle L:90%
, this this this degree thing is just not giving

788
00:44:40.900 --> 00:44:45.659 A:middle L:90%
us enough sensitivity. So we we wanted to see

789
00:44:45.619 --> 00:44:50.809 A:middle L:90%
if there was a way to more precisely evaluate what

790
00:44:50.809 --> 00:44:54.579 A:middle L:90%
role these topic words and models or indicators of hedging

791
00:44:54.579 --> 00:44:58.699 A:middle L:90%
were playing. Because we had a hunch now that

792
00:44:58.699 --> 00:45:04.070 A:middle L:90%
there were little networks inside here that corresponded to some

793
00:45:04.070 --> 00:45:07.610 A:middle L:90%
behavior that we that we perhaps we're hoping, uh

794
00:45:07.619 --> 00:45:08.119 A:middle L:90%
, to see. And and in fact, we

795
00:45:08.119 --> 00:45:12.590 A:middle L:90%
recall that in the science, um, are in

796
00:45:12.590 --> 00:45:16.769 A:middle L:90%
the hand coated edition that are human Raiders we're seeing

797
00:45:16.769 --> 00:45:20.710 A:middle L:90%
. They were indicating that some science learning was going

798
00:45:20.710 --> 00:45:27.449 A:middle L:90%
on. So what we did next is we took

799
00:45:27.980 --> 00:45:32.539 A:middle L:90%
each instance of the word egg, right, And

800
00:45:32.550 --> 00:45:36.969 A:middle L:90%
we looked at, and I told this story earlier

801
00:45:36.969 --> 00:45:40.380 A:middle L:90%
today. This is another kind of discovery story we're

802
00:45:40.380 --> 00:45:45.159 A:middle L:90%
looking at. How often it Rikers? So it

803
00:45:45.159 --> 00:45:49.030 A:middle L:90%
was. There were in across all of those posts

804
00:45:49.030 --> 00:45:52.039 A:middle L:90%
That giant data set that I showed you it showed

805
00:45:52.039 --> 00:45:57.150 A:middle L:90%
up more than 400 times the word egg, and

806
00:45:57.150 --> 00:45:59.949 A:middle L:90%
it would sometimes show up, closer together and sometimes

807
00:45:59.949 --> 00:46:02.340 A:middle L:90%
further apart. But when we look at the distribution

808
00:46:02.340 --> 00:46:07.500 A:middle L:90%
of how long it would it would, how many

809
00:46:07.500 --> 00:46:09.989 A:middle L:90%
tokens would be in between each recurrence? We saw

810
00:46:09.989 --> 00:46:15.440 A:middle L:90%
that there was an interesting stabilisation point that is,

811
00:46:15.440 --> 00:46:21.110 A:middle L:90%
there was a big chunk of of these intervals that

812
00:46:21.110 --> 00:46:25.690 A:middle L:90%
were about the same size about 17. And we

813
00:46:25.699 --> 00:46:28.409 A:middle L:90%
were like, What is going on there? So

814
00:46:28.409 --> 00:46:30.480 A:middle L:90%
we didn't actually understand that very well. We could

815
00:46:30.480 --> 00:46:35.679 A:middle L:90%
always toggle back and forth between the clear text and

816
00:46:35.679 --> 00:46:39.119 A:middle L:90%
see the actual natural language text thread and what we

817
00:46:39.119 --> 00:46:43.409 A:middle L:90%
were seeing in the, uh in the analysis.

818
00:46:43.880 --> 00:46:46.099 A:middle L:90%
And what we were pretty confident of is that when

819
00:46:46.110 --> 00:46:50.309 A:middle L:90%
the cycles were long, when eggs showed up and

820
00:46:50.309 --> 00:46:53.340 A:middle L:90%
there were, like 300 tokens before another iteration of

821
00:46:53.340 --> 00:46:55.480 A:middle L:90%
eggs showed up. That was when Jackie Jacobs was

822
00:46:55.480 --> 00:46:59.869 A:middle L:90%
talking our poultry expert because we could see that happening

823
00:46:59.880 --> 00:47:02.960 A:middle L:90%
quite a lot. But in the other times,

824
00:47:02.969 --> 00:47:06.269 A:middle L:90%
um, it looked like it was more genuine,

825
00:47:06.269 --> 00:47:07.469 A:middle L:90%
genuinely dia logic, which is a good sign to

826
00:47:07.469 --> 00:47:09.760 A:middle L:90%
us. And it looked like there was more of

827
00:47:09.760 --> 00:47:15.519 A:middle L:90%
this kind of questioning and proffering a potential, uh

828
00:47:15.530 --> 00:47:20.320 A:middle L:90%
, idea. And then there was, uh,

829
00:47:20.329 --> 00:47:24.070 A:middle L:90%
answering or modifying in other words, more deliberative activity

830
00:47:24.070 --> 00:47:27.570 A:middle L:90%
of the sort that we thought that our human Raiders

831
00:47:27.570 --> 00:47:30.469 A:middle L:90%
were looking at. So, on a whim,

832
00:47:30.469 --> 00:47:31.940 A:middle L:90%
we said Yes, but what does that? How

833
00:47:31.940 --> 00:47:35.500 A:middle L:90%
does that actually show up? And so we started

834
00:47:35.500 --> 00:47:39.480 A:middle L:90%
pulling those out as individual pieces of the graph and

835
00:47:39.480 --> 00:47:43.119 A:middle L:90%
throwing them into a software package called Jeffy, which

836
00:47:43.119 --> 00:47:45.349 A:middle L:90%
helps us visualize some of these networks. And that's

837
00:47:45.349 --> 00:47:50.750 A:middle L:90%
what you're seeing here. So here's one instance of

838
00:47:50.750 --> 00:47:54.590 A:middle L:90%
egg and we walk around here until we get to

839
00:47:54.590 --> 00:48:05.269 A:middle L:90%
the next one and then another one starts. And

840
00:48:05.269 --> 00:48:07.989 A:middle L:90%
so what we were seeing is when we zoomed in

841
00:48:07.059 --> 00:48:10.110 A:middle L:90%
from the big picture of the network, our graph

842
00:48:10.110 --> 00:48:14.829 A:middle L:90%
was a series of cycles. Now I'm gonna come

843
00:48:14.829 --> 00:48:19.300 A:middle L:90%
back to my title Common places. So this is

844
00:48:19.300 --> 00:48:22.280 A:middle L:90%
where we were like, Oh, interesting. Okay

845
00:48:22.869 --> 00:48:24.289 A:middle L:90%
, these are places that group is returning to.

846
00:48:25.550 --> 00:48:30.530 A:middle L:90%
In other words, what holds this deliberative discourse together

847
00:48:30.530 --> 00:48:36.400 A:middle L:90%
is a mention of the word egg and some pattern

848
00:48:36.400 --> 00:48:37.519 A:middle L:90%
, and it turns out in here. You'll see

849
00:48:37.519 --> 00:48:43.119 A:middle L:90%
some of these. I believe there's something this one

850
00:48:43.130 --> 00:48:50.500 A:middle L:90%
. If there's one of hedge value words, so

851
00:48:50.500 --> 00:48:51.980 A:middle L:90%
here's what it kind of looks like. This is

852
00:48:51.980 --> 00:49:00.010 A:middle L:90%
a typical cycle. So here's our high value,

853
00:49:00.010 --> 00:49:02.849 A:middle L:90%
high centrality value token egg. You can see the

854
00:49:02.849 --> 00:49:06.119 A:middle L:90%
number is quite a lot bigger than all the others

855
00:49:07.469 --> 00:49:09.230 A:middle L:90%
. And then there's an if then then there say

856
00:49:09.230 --> 00:49:14.590 A:middle L:90%
, not Can will all of our models right?

857
00:49:14.760 --> 00:49:19.659 A:middle L:90%
And all of these are, um, roughly commonly

858
00:49:19.659 --> 00:49:22.000 A:middle L:90%
represented in these kinds of cycles that we're talking about

859
00:49:22.570 --> 00:49:25.210 A:middle L:90%
. So these are all the way across. This

860
00:49:25.210 --> 00:49:29.150 A:middle L:90%
is what what we started thinking of is our egg

861
00:49:29.150 --> 00:49:37.820 A:middle L:90%
cycles, or are rhetorical cycles mhm? Let me

862
00:49:37.820 --> 00:49:39.590 A:middle L:90%
try to help you understand what we think that means

863
00:49:40.369 --> 00:49:43.179 A:middle L:90%
. So I'm gonna walk back. I promise I

864
00:49:43.179 --> 00:49:45.329 A:middle L:90%
would do this for the research team today. Here's

865
00:49:45.340 --> 00:49:47.929 A:middle L:90%
here's how we prepared the sample and here's here's the

866
00:49:47.929 --> 00:49:50.900 A:middle L:90%
I won't go into all the technical parts, but

867
00:49:51.269 --> 00:49:53.460 A:middle L:90%
how we got to a picture like that. So

868
00:49:53.460 --> 00:49:57.130 A:middle L:90%
first we take all the post and we stem them

869
00:49:57.130 --> 00:49:58.980 A:middle L:90%
so that we're seeing just the stems of the words

870
00:49:58.980 --> 00:50:01.480 A:middle L:90%
Take off the endings and we calculate the frequency distributions

871
00:50:01.480 --> 00:50:06.880 A:middle L:90%
for the tokens. Then we process the tokens as

872
00:50:06.880 --> 00:50:09.349 A:middle L:90%
diagrams to start making drawing edges. And then,

873
00:50:09.400 --> 00:50:12.980 A:middle L:90%
according to when they were posted, we build the

874
00:50:12.980 --> 00:50:15.719 A:middle L:90%
graph. That is, the more tokens that are

875
00:50:15.719 --> 00:50:20.219 A:middle L:90%
added. We get more links, right? As

876
00:50:20.219 --> 00:50:25.840 A:middle L:90%
people post to the comments to the threat, this

877
00:50:25.840 --> 00:50:30.190 A:middle L:90%
effectively converts the whole text into a big graph.

878
00:50:30.559 --> 00:50:34.329 A:middle L:90%
At this point we're in Bactine land. We no

879
00:50:34.329 --> 00:50:37.539 A:middle L:90%
longer have any way to trace back. Specifically,

880
00:50:37.550 --> 00:50:39.449 A:middle L:90%
um, in the in this view, who said

881
00:50:39.449 --> 00:50:45.400 A:middle L:90%
what to who. So we're not thinking about individual

882
00:50:45.400 --> 00:50:49.820 A:middle L:90%
speakers or individual agents at all. We're just looking

883
00:50:49.820 --> 00:50:53.429 A:middle L:90%
at the structure of the whole. Then we calculate

884
00:50:53.429 --> 00:50:58.039 A:middle L:90%
the between the centrality and we order the tokens high

885
00:50:58.039 --> 00:51:00.329 A:middle L:90%
to low. And that gives us our candidates for

886
00:51:00.329 --> 00:51:06.440 A:middle L:90%
these walks or cycles. Then the high between these

887
00:51:06.440 --> 00:51:08.489 A:middle L:90%
terms are used as the anchors and we plot those

888
00:51:08.860 --> 00:51:12.099 A:middle L:90%
. And then we code the walks for the presence

889
00:51:12.099 --> 00:51:15.760 A:middle L:90%
of these high between this tokens that indicate hedging and

890
00:51:15.760 --> 00:51:20.210 A:middle L:90%
that's what we call our hedge finder. If we

891
00:51:20.210 --> 00:51:22.530 A:middle L:90%
see enough of those, we think that we can

892
00:51:22.530 --> 00:51:24.440 A:middle L:90%
do something interesting. Which is to say in this

893
00:51:24.440 --> 00:51:30.489 A:middle L:90%
thread science is happening or it's more science c maybe

894
00:51:30.860 --> 00:51:37.289 A:middle L:90%
right, that's a hedge. Let's see what these

895
00:51:37.289 --> 00:51:38.130 A:middle L:90%
look like so that you can give a sense of

896
00:51:38.130 --> 00:51:40.219 A:middle L:90%
what I'm talking about. So I pulled this one

897
00:51:40.219 --> 00:51:44.610 A:middle L:90%
off of our test set. Um, because as

898
00:51:44.619 --> 00:51:49.429 A:middle L:90%
we couldn't just go by, we had to evaluate

899
00:51:49.429 --> 00:51:54.250 A:middle L:90%
our little hedge finding analysis here. Um, so

900
00:51:54.250 --> 00:51:57.719 A:middle L:90%
this is the heads that comes from a scientific article

901
00:51:57.719 --> 00:52:00.619 A:middle L:90%
. It's actually a Cochrane review of the use of

902
00:52:00.630 --> 00:52:06.719 A:middle L:90%
electronic devices to monitor, um and help type two

903
00:52:06.719 --> 00:52:10.659 A:middle L:90%
diabetes patients. Um, Cochrane Reviews are really reliable

904
00:52:10.670 --> 00:52:13.789 A:middle L:90%
sources of us, uh, for us, for

905
00:52:13.800 --> 00:52:15.409 A:middle L:90%
hedges, because they try to translate the results of

906
00:52:15.409 --> 00:52:20.440 A:middle L:90%
research for practitioners. And they have this passage in

907
00:52:20.440 --> 00:52:22.769 A:middle L:90%
them. That is a what's called a plain language

908
00:52:22.769 --> 00:52:24.449 A:middle L:90%
summary of the science. And they say, Okay

909
00:52:24.449 --> 00:52:27.849 A:middle L:90%
, primary care doctors. Here's what the research says

910
00:52:27.860 --> 00:52:30.230 A:middle L:90%
. And now here's what you should do. And

911
00:52:30.230 --> 00:52:31.849 A:middle L:90%
so it's super hedged all the time, it says

912
00:52:31.900 --> 00:52:35.300 A:middle L:90%
, appear to have right based on the strength of

913
00:52:35.300 --> 00:52:38.250 A:middle L:90%
the evidence. So these are the hedges that were

914
00:52:38.250 --> 00:52:40.699 A:middle L:90%
kind of dealing with. What we did is we

915
00:52:40.699 --> 00:52:47.969 A:middle L:90%
built a test set that included about 900 sentences that

916
00:52:47.969 --> 00:52:52.159 A:middle L:90%
were hedged hedges and 900 that weren't. And all

917
00:52:52.159 --> 00:52:54.719 A:middle L:90%
of these came from scientific articles so that we could

918
00:52:54.719 --> 00:52:59.150 A:middle L:90%
have a at least a prima facie argument that this

919
00:52:59.150 --> 00:53:07.219 A:middle L:90%
is science, science, people doing science. And

920
00:53:07.219 --> 00:53:08.710 A:middle L:90%
then we set to work trying to see if our

921
00:53:08.710 --> 00:53:15.360 A:middle L:90%
heads finder could detect in a random string of these

922
00:53:15.360 --> 00:53:19.010 A:middle L:90%
sentences what the humans could do. Now we're a

923
00:53:19.010 --> 00:53:21.420 A:middle L:90%
lot more narrow. We're just trying to see if

924
00:53:21.429 --> 00:53:22.440 A:middle L:90%
you can see a hedge and the computer can see

925
00:53:22.440 --> 00:53:27.420 A:middle L:90%
a hedge. The rules more or less go like

926
00:53:27.420 --> 00:53:29.630 A:middle L:90%
this. So when we're building the algorithm, this

927
00:53:29.630 --> 00:53:32.469 A:middle L:90%
is kind of our, uh our parameters, if

928
00:53:32.469 --> 00:53:37.139 A:middle L:90%
you will, were trying to tune walks of in

929
00:53:37.139 --> 00:53:38.769 A:middle L:90%
token link. That repeat seemed to act as indicators

930
00:53:38.769 --> 00:53:42.019 A:middle L:90%
for types of genres. And so in. In

931
00:53:42.019 --> 00:53:45.460 A:middle L:90%
this case, as I mentioned before, Jackie Jacobs

932
00:53:45.469 --> 00:53:49.539 A:middle L:90%
, long explanations were kind of out of genre for

933
00:53:49.539 --> 00:53:52.130 A:middle L:90%
what we were looking for, which were these more

934
00:53:52.130 --> 00:53:54.969 A:middle L:90%
deliberative moments. So they were more like an essay

935
00:53:54.980 --> 00:54:01.949 A:middle L:90%
on chickens rather than reasoning about chickens. So within

936
00:54:01.949 --> 00:54:07.050 A:middle L:90%
that narrow range that was relatively stable, Um,

937
00:54:07.440 --> 00:54:10.269 A:middle L:90%
we were seeing that head cycles are structures similar to

938
00:54:10.269 --> 00:54:15.090 A:middle L:90%
those greatly weakened utterance boundaries, right. The permeable

939
00:54:15.090 --> 00:54:20.940 A:middle L:90%
to the author's expression ideas. Um, so we

940
00:54:20.940 --> 00:54:24.139 A:middle L:90%
mark those statements as hedges or non hedges in the

941
00:54:24.139 --> 00:54:29.239 A:middle L:90%
same way we think that humans are seeing them in

942
00:54:29.239 --> 00:54:30.690 A:middle L:90%
that way. That is that they're picking up the

943
00:54:30.690 --> 00:54:38.710 A:middle L:90%
same signal, if you will. So if we

944
00:54:38.710 --> 00:54:42.610 A:middle L:90%
showed these to a human, we would just say

945
00:54:42.610 --> 00:54:44.550 A:middle L:90%
, Is that a hedges in a non hedge?

946
00:54:45.239 --> 00:54:50.079 A:middle L:90%
Right? And, uh, yeah, I'll just

947
00:54:50.079 --> 00:54:51.929 A:middle L:90%
skip this one part. Let me let me show

948
00:54:51.929 --> 00:54:57.849 A:middle L:90%
you how we did that. Now we've tried to

949
00:54:57.849 --> 00:55:00.900 A:middle L:90%
be careful about how we're training, um, our

950
00:55:00.900 --> 00:55:04.559 A:middle L:90%
little algorithm to do this because again, we do

951
00:55:04.559 --> 00:55:06.679 A:middle L:90%
not want, um, we do not want to

952
00:55:06.679 --> 00:55:09.460 A:middle L:90%
have the machine returning, a result that a person

953
00:55:09.469 --> 00:55:14.889 A:middle L:90%
would also see. So in this case, sentences

954
00:55:14.889 --> 00:55:17.329 A:middle L:90%
that admit to incomplete our limitations and knowledge. That's

955
00:55:17.329 --> 00:55:22.329 A:middle L:90%
probably our clearest example. And then there are some

956
00:55:22.329 --> 00:55:23.699 A:middle L:90%
other kinds of hedges here that are a little bit

957
00:55:23.699 --> 00:55:29.440 A:middle L:90%
more subtle. So if you look at the third

958
00:55:29.440 --> 00:55:30.710 A:middle L:90%
one, they're the ones that predict outcomes which are

959
00:55:30.710 --> 00:55:35.980 A:middle L:90%
less than definitive or speculative. So you see may

960
00:55:35.989 --> 00:55:38.820 A:middle L:90%
or might in those kinds of things, note that

961
00:55:38.829 --> 00:55:43.170 A:middle L:90%
In every case, the hedge is more than just

962
00:55:43.739 --> 00:55:47.059 A:middle L:90%
one word or the co presence of two stable tokens

963
00:55:47.059 --> 00:55:51.820 A:middle L:90%
. It is a relationship among one or more tokens

964
00:55:51.869 --> 00:55:54.099 A:middle L:90%
that might vary but which are predictable from a particular

965
00:55:54.099 --> 00:56:00.780 A:middle L:90%
list. And so it is probably a stable co

966
00:56:00.780 --> 00:56:05.960 A:middle L:90%
variant relationship at some point, that's what That's the

967
00:56:05.960 --> 00:56:09.420 A:middle L:90%
assumption that we were looking at. So we use

968
00:56:09.420 --> 00:56:15.579 A:middle L:90%
these 900 heads 909 hedge sentences tested it to train

969
00:56:15.579 --> 00:56:20.469 A:middle L:90%
the classifier. We tested it against 155 hand coated

970
00:56:20.769 --> 00:56:25.230 A:middle L:90%
155 non hedge sentences. And then we did that

971
00:56:25.239 --> 00:56:29.369 A:middle L:90%
. We repeated that, um, randomized, the

972
00:56:29.369 --> 00:56:32.059 A:middle L:90%
order in which they happen in each pass relied on

973
00:56:32.070 --> 00:56:37.460 A:middle L:90%
that randomized training and about 80% of the time,

974
00:56:37.829 --> 00:56:42.659 A:middle L:90%
um, the computer marks it, right. So

975
00:56:43.230 --> 00:56:45.110 A:middle L:90%
, um, the way it looks is decidedly non

976
00:56:45.110 --> 00:56:49.570 A:middle L:90%
exciting. It spits out an excel CSP Excel spreadsheet

977
00:56:49.570 --> 00:56:52.769 A:middle L:90%
, and and it has the sentence and it says

978
00:56:52.780 --> 00:56:55.039 A:middle L:90%
hedge or non hedge, just like you would in

979
00:56:55.039 --> 00:56:59.230 A:middle L:90%
the in the data set that I showed you before

980
00:56:59.880 --> 00:57:01.389 A:middle L:90%
. And it still gets it wrong about 20% of

981
00:57:01.389 --> 00:57:06.590 A:middle L:90%
the time. But it has some possibility. Has

982
00:57:06.590 --> 00:57:14.590 A:middle L:90%
some has some some thoughts. So we're starting to

983
00:57:14.599 --> 00:57:16.360 A:middle L:90%
think of what? What is this thing that we're

984
00:57:16.369 --> 00:57:17.719 A:middle L:90%
that we're doing here? Like, we have to

985
00:57:17.719 --> 00:57:22.230 A:middle L:90%
give it some names? Um, and we realized

986
00:57:22.230 --> 00:57:24.099 A:middle L:90%
that as we've gone along, and as I've presented

987
00:57:24.099 --> 00:57:27.960 A:middle L:90%
it for you here today, we're getting really much

988
00:57:27.960 --> 00:57:29.639 A:middle L:90%
more narrow and more narrow and more narrow. And

989
00:57:29.639 --> 00:57:31.809 A:middle L:90%
that's what I was saying before about how, really

990
00:57:31.820 --> 00:57:36.210 A:middle L:90%
, uh, excited. You get about humans ability

991
00:57:36.210 --> 00:57:39.050 A:middle L:90%
to interpret text because the signals that we're picking up

992
00:57:39.050 --> 00:57:43.210 A:middle L:90%
when we read all of these things are really ridiculously

993
00:57:43.210 --> 00:57:47.530 A:middle L:90%
difficult to train a non human to do. And

994
00:57:47.530 --> 00:57:50.480 A:middle L:90%
I think we're not in any trouble at any time

995
00:57:50.480 --> 00:57:53.409 A:middle L:90%
soon in that regard. Um, but we also

996
00:57:53.409 --> 00:57:59.429 A:middle L:90%
think that there are some reliable indicators for lots of

997
00:57:59.429 --> 00:58:00.940 A:middle L:90%
different kinds of genres that could be incredibly useful.

998
00:58:00.949 --> 00:58:05.989 A:middle L:90%
So what is the science museum do with something like

999
00:58:05.989 --> 00:58:07.380 A:middle L:90%
this? Well, here's what we hope to have

1000
00:58:07.380 --> 00:58:09.510 A:middle L:90%
them doing, maybe as soon as the end of

1001
00:58:09.510 --> 00:58:15.179 A:middle L:90%
this year. And that is in an unsupervised way

1002
00:58:15.190 --> 00:58:20.019 A:middle L:90%
using the hedge finder to see which threads are doing

1003
00:58:20.019 --> 00:58:22.570 A:middle L:90%
the kind of work to facilitate science, learning that

1004
00:58:22.570 --> 00:58:24.380 A:middle L:90%
they hope that they would do and which threads are

1005
00:58:24.380 --> 00:58:28.630 A:middle L:90%
not. So that then that we can start thinking

1006
00:58:28.630 --> 00:58:31.969 A:middle L:90%
about well, which facilitation strategies might have some impact

1007
00:58:31.969 --> 00:58:36.460 A:middle L:90%
on those things? Is it, Um, And

1008
00:58:36.460 --> 00:58:37.920 A:middle L:90%
so that's That's the key question, right? The

1009
00:58:37.929 --> 00:58:42.440 A:middle L:90%
what happens in order to create that that pattern?

1010
00:58:42.820 --> 00:58:45.860 A:middle L:90%
Because we have a lot of cases threads that go

1011
00:58:45.860 --> 00:58:46.579 A:middle L:90%
on for a long time. We see them move

1012
00:58:46.579 --> 00:58:52.030 A:middle L:90%
in and out of the desired head zone, for

1013
00:58:52.030 --> 00:58:57.900 A:middle L:90%
example, for this particular test, and one of

1014
00:58:57.900 --> 00:59:00.820 A:middle L:90%
the things that we're able to see is questions have

1015
00:59:00.820 --> 00:59:05.889 A:middle L:90%
a remarkable ability to bring people back into that 17

1016
00:59:05.889 --> 00:59:08.460 A:middle L:90%
or 18 token cycle. That seems to be the

1017
00:59:08.460 --> 00:59:12.880 A:middle L:90%
sweet spot for people reasoning in interesting ways. And

1018
00:59:12.880 --> 00:59:15.030 A:middle L:90%
that's that was a great that was grateful to me

1019
00:59:15.039 --> 00:59:16.880 A:middle L:90%
because that's what teachers do all the time. And

1020
00:59:16.880 --> 00:59:20.190 A:middle L:90%
that's what scientists do all the time as they ask

1021
00:59:20.190 --> 00:59:22.159 A:middle L:90%
questions. And it makes sense on its face to

1022
00:59:22.159 --> 00:59:27.150 A:middle L:90%
say, um, if somebody's going on and on

1023
00:59:27.519 --> 00:59:30.429 A:middle L:90%
a question might bring the group back. Uh,

1024
00:59:30.440 --> 00:59:34.440 A:middle L:90%
like I'm going on and on right now, Uh

1025
00:59:35.820 --> 00:59:37.550 A:middle L:90%
, mhm. It leaves open, though a fair

1026
00:59:37.550 --> 00:59:39.320 A:middle L:90%
amount of questions, and I want to just kind

1027
00:59:39.320 --> 00:59:42.150 A:middle L:90%
of throw out a few. And as I mentioned

1028
00:59:42.150 --> 00:59:45.079 A:middle L:90%
, this is all very experimental work. We think

1029
00:59:45.079 --> 00:59:47.769 A:middle L:90%
that the Heads Finder is one very narrow test linked

1030
00:59:47.769 --> 00:59:52.530 A:middle L:90%
to one very specific kind of discourse that runs on

1031
00:59:52.570 --> 00:59:57.949 A:middle L:90%
a fairly specific kind of medium or mode of text

1032
00:59:57.960 --> 01:00:00.889 A:middle L:90%
. Textual interaction, which is a list serve discussions

1033
01:00:00.889 --> 01:00:02.929 A:middle L:90%
and online discussions. Does it work? If you

1034
01:00:02.940 --> 01:00:08.059 A:middle L:90%
if the utterances are scientific articles of scientists talking back

1035
01:00:08.059 --> 01:00:09.610 A:middle L:90%
and forth for each other, does it work?

1036
01:00:09.610 --> 01:00:15.039 A:middle L:90%
If there, um, newspaper articles, we couldn't

1037
01:00:15.039 --> 01:00:17.159 A:middle L:90%
tell you, but I suspect it would be a

1038
01:00:17.159 --> 01:00:22.030 A:middle L:90%
lot noisier, and Messier does more Heggie equal more

1039
01:00:22.030 --> 01:00:27.019 A:middle L:90%
science C We think so, and it can be

1040
01:00:27.019 --> 01:00:30.139 A:middle L:90%
a useful tool. But in finding uh, sorting

1041
01:00:30.139 --> 01:00:34.199 A:middle L:90%
through millions of threads to find out which ones are

1042
01:00:34.210 --> 01:00:37.710 A:middle L:90%
worth taking a closer look at, however, there

1043
01:00:37.710 --> 01:00:40.099 A:middle L:90%
are probably some real limits there. For example,

1044
01:00:40.099 --> 01:00:43.440 A:middle L:90%
we were talking about some today that, um,

1045
01:00:43.449 --> 01:00:46.289 A:middle L:90%
there's a lot of interpersonal hedging going on in these

1046
01:00:46.289 --> 01:00:50.570 A:middle L:90%
threads because they're humans dealing with other humans. So

1047
01:00:50.570 --> 01:00:52.010 A:middle L:90%
some of these hedges are not about the subject matter

1048
01:00:52.010 --> 01:00:54.190 A:middle L:90%
there about Well, I don't want to offend this

1049
01:00:54.190 --> 01:01:00.869 A:middle L:90%
person, so they're treading lightly instead, so we

1050
01:01:00.869 --> 01:01:02.949 A:middle L:90%
aren't necessarily seeing a 1 to 1. Correspondence with

1051
01:01:02.949 --> 01:01:07.349 A:middle L:90%
more heads, behavior and more scientific reasoning, um

1052
01:01:07.360 --> 01:01:09.739 A:middle L:90%
, are are special topics common. That's a slightly

1053
01:01:09.739 --> 01:01:14.739 A:middle L:90%
different version of my question that I asked special topics

1054
01:01:15.110 --> 01:01:19.400 A:middle L:90%
our topics that are key to a particular kind of

1055
01:01:19.400 --> 01:01:23.199 A:middle L:90%
discourse or genre. Here we will feel pretty confident

1056
01:01:23.199 --> 01:01:25.639 A:middle L:90%
in saying they appear to be, and that is

1057
01:01:27.110 --> 01:01:30.619 A:middle L:90%
, hedging is common. If you're if it's scientist

1058
01:01:30.619 --> 01:01:34.690 A:middle L:90%
talking to other scientists, or if it's a scientist

1059
01:01:34.690 --> 01:01:37.949 A:middle L:90%
talking to a layperson about a scientific topic, hedging

1060
01:01:37.949 --> 01:01:39.969 A:middle L:90%
is there. And so that's what we would expect

1061
01:01:39.969 --> 01:01:46.289 A:middle L:90%
to see right. Our common topics common so common

1062
01:01:46.289 --> 01:01:50.929 A:middle L:90%
topics are topics that are a little bit more broad

1063
01:01:50.940 --> 01:01:53.099 A:middle L:90%
things like cause and effect. Do we see that

1064
01:01:53.099 --> 01:01:57.239 A:middle L:90%
all the time when people are doing deliberative? A

1065
01:01:57.250 --> 01:02:00.989 A:middle L:90%
reasoning of any sort? Well, Aristotle encourages us

1066
01:02:00.000 --> 01:02:01.849 A:middle L:90%
to teach all of our students that that's how it

1067
01:02:01.849 --> 01:02:06.289 A:middle L:90%
always happens. But we don't really know that they're

1068
01:02:06.289 --> 01:02:09.829 A:middle L:90%
that common are common sense approach, and our close

1069
01:02:09.829 --> 01:02:15.559 A:middle L:90%
reading will say that in many instances, cause and

1070
01:02:15.559 --> 01:02:17.510 A:middle L:90%
effect is an effective way to reason. But we

1071
01:02:17.510 --> 01:02:20.829 A:middle L:90%
don't know, but now we might have a shot

1072
01:02:20.840 --> 01:02:24.219 A:middle L:90%
at starting to find out if those common places that

1073
01:02:24.219 --> 01:02:29.320 A:middle L:90%
we think about when we associate good communication or effective

1074
01:02:29.320 --> 01:02:32.210 A:middle L:90%
communication in one area can actually be validated against large

1075
01:02:32.210 --> 01:02:37.469 A:middle L:90%
examples of multiple situations of those things. So I'm

1076
01:02:37.469 --> 01:02:40.739 A:middle L:90%
gonna stop there in case I, uh, have

1077
01:02:40.739 --> 01:02:44.530 A:middle L:90%
any or in case you would like to ask any

1078
01:02:44.530 --> 01:02:45.380 A:middle L:90%
questions. I know it can be a little bewildering

1079
01:02:45.380 --> 01:02:47.239 A:middle L:90%
, but I appreciate so much you're listening to me

1080
01:02:47.239 --> 01:02:59.059 A:middle L:90%
today and welcome any questions you might have. I've

1081
01:02:59.070 --> 01:03:00.469 A:middle L:90%
got a microphone since we're recording this session, If

1082
01:03:00.469 --> 01:03:04.409 A:middle L:90%
you wouldn't mind raising hand, I'll bring the mic

1083
01:03:04.420 --> 01:03:10.559 A:middle L:90%
to you so you can ask the questions. Building

1084
01:03:10.559 --> 01:03:16.440 A:middle L:90%
a question that, yeah, we talked to worry

1085
01:03:16.440 --> 01:03:20.090 A:middle L:90%
about the medical research that we're doing in this school

1086
01:03:20.099 --> 01:03:22.300 A:middle L:90%
And, um, for us, we're looking at

1087
01:03:22.300 --> 01:03:25.679 A:middle L:90%
how experts speak versus well known experts speak. So

1088
01:03:25.690 --> 01:03:30.929 A:middle L:90%
for us, hedging actually in debt indicates non expertise

1089
01:03:30.300 --> 01:03:32.489 A:middle L:90%
. I'm not sure the stock's gonna go up,

1090
01:03:32.500 --> 01:03:36.429 A:middle L:90%
but I think it might versus somebody who's an expert

1091
01:03:36.429 --> 01:03:38.940 A:middle L:90%
says Apple's going out by now something like that.

1092
01:03:38.949 --> 01:03:44.000 A:middle L:90%
So I think it's interesting that in your testing data

1093
01:03:44.010 --> 01:03:45.670 A:middle L:90%
, when you're looking at those science articles and grabbing

1094
01:03:45.679 --> 01:03:52.090 A:middle L:90%
hedging words as evidence of science that that last section

1095
01:03:52.090 --> 01:03:53.309 A:middle L:90%
is looking at how they're kind of translating it for

1096
01:03:53.309 --> 01:03:57.280 A:middle L:90%
the practitioner and later. And so I think that

1097
01:03:57.280 --> 01:04:00.719 A:middle L:90%
that may be something that carries on in the future

1098
01:04:00.719 --> 01:04:02.460 A:middle L:90%
of your research. As far as this relationship of

1099
01:04:02.469 --> 01:04:06.730 A:middle L:90%
hedging indicating science, which seems contradictory to what we're

1100
01:04:06.730 --> 01:04:10.829 A:middle L:90%
doing because I don't think scientists as expert but hedging

1101
01:04:10.840 --> 01:04:13.949 A:middle L:90%
for us as a non expert, Um, if

1102
01:04:13.949 --> 01:04:16.340 A:middle L:90%
a scientist is talking to a scientist versus a scientist

1103
01:04:16.340 --> 01:04:19.500 A:middle L:90%
taught me to lay audience, how does their vocabulary

1104
01:04:19.510 --> 01:04:23.650 A:middle L:90%
change and or how does their hedges? Yeah,

1105
01:04:23.650 --> 01:04:28.860 A:middle L:90%
that that's the interesting thing about this and the thing

1106
01:04:28.860 --> 01:04:30.150 A:middle L:90%
that we would have to be careful about. We

1107
01:04:30.150 --> 01:04:32.119 A:middle L:90%
talked a little bit about what is the likely outcome

1108
01:04:32.119 --> 01:04:34.780 A:middle L:90%
in the near term of this line of research is

1109
01:04:34.780 --> 01:04:39.369 A:middle L:90%
that we'll have a little piece of analytic code that

1110
01:04:39.369 --> 01:04:42.340 A:middle L:90%
we could actually give your research group, and you

1111
01:04:42.340 --> 01:04:45.920 A:middle L:90%
could run to find hedges in your corpus. But

1112
01:04:45.920 --> 01:04:48.579 A:middle L:90%
what those hedges mean in your corpus, our technique

1113
01:04:48.579 --> 01:04:50.820 A:middle L:90%
could be silent about it would be like buying a

1114
01:04:50.820 --> 01:04:54.000 A:middle L:90%
chemical test from, you know, one of those

1115
01:04:54.000 --> 01:04:57.349 A:middle L:90%
scientific companies where you can find out the salinity of

1116
01:04:57.349 --> 01:05:01.130 A:middle L:90%
something. Unless you know what that is An indicator

1117
01:05:01.130 --> 01:05:04.630 A:middle L:90%
for it won't do a lot of good. So

1118
01:05:04.630 --> 01:05:06.909 A:middle L:90%
this is as close as we get in the humanities

1119
01:05:06.909 --> 01:05:10.679 A:middle L:90%
. Two basic science. And that's part of why

1120
01:05:10.679 --> 01:05:12.630 A:middle L:90%
it's a little dry. I apologize for that.

1121
01:05:12.639 --> 01:05:15.050 A:middle L:90%
But, um, I think what we would definitely

1122
01:05:15.050 --> 01:05:19.710 A:middle L:90%
need to do in, um in response to a

1123
01:05:19.719 --> 01:05:24.099 A:middle L:90%
problem like the one that you're pointing out is one

1124
01:05:24.099 --> 01:05:26.900 A:middle L:90%
is have more kinds of tests, right? So

1125
01:05:26.900 --> 01:05:29.210 A:middle L:90%
if we can do if we can develop a hedge

1126
01:05:29.210 --> 01:05:33.110 A:middle L:90%
finder, odds are we could find other word proteins

1127
01:05:33.119 --> 01:05:38.900 A:middle L:90%
and genre characteristics to develop other tests. Then we

1128
01:05:38.900 --> 01:05:42.280 A:middle L:90%
could use those, perhaps in combination and with some

1129
01:05:42.280 --> 01:05:45.650 A:middle L:90%
nuance, um, to do some more interesting work

1130
01:05:45.650 --> 01:05:46.780 A:middle L:90%
and maybe those. That is where we start to

1131
01:05:46.780 --> 01:05:59.050 A:middle L:90%
cross some boundaries. Perhaps. Uh huh. I

1132
01:05:59.050 --> 01:06:00.730 A:middle L:90%
practiced My question was maybe not fully understanding everything,

1133
01:06:00.730 --> 01:06:03.989 A:middle L:90%
but I just have 111 question. The populations that

1134
01:06:03.989 --> 01:06:06.920 A:middle L:90%
you look at the exchanges where the groups basically homogeneous

1135
01:06:06.929 --> 01:06:11.130 A:middle L:90%
and because if you're looking for, um accepting certain

1136
01:06:11.130 --> 01:06:13.380 A:middle L:90%
kinds of proteins from things, it's going to depend

1137
01:06:13.380 --> 01:06:15.530 A:middle L:90%
on the culture and the people within that group.

1138
01:06:15.530 --> 01:06:16.800 A:middle L:90%
So where if you were to do this in another

1139
01:06:16.800 --> 01:06:19.800 A:middle L:90%
group or another strand of, you know, say

1140
01:06:20.179 --> 01:06:24.469 A:middle L:90%
posts, it may be very different. And so

1141
01:06:24.469 --> 01:06:27.480 A:middle L:90%
we may find other proteins. Uh, yeah,

1142
01:06:27.480 --> 01:06:29.710 A:middle L:90%
I guess it would depend on. That's a great

1143
01:06:29.719 --> 01:06:31.369 A:middle L:90%
, great question and a great point. I would

1144
01:06:31.369 --> 01:06:35.780 A:middle L:90%
say it depend on the criteria for homogeneity, but

1145
01:06:35.780 --> 01:06:40.400 A:middle L:90%
I suspect that they are very homogeneous. They're the

1146
01:06:40.400 --> 01:06:45.219 A:middle L:90%
kind of people who go to science blogs, which

1147
01:06:45.219 --> 01:06:46.800 A:middle L:90%
means there are a lot of kids. Um,

1148
01:06:47.380 --> 01:06:53.719 A:middle L:90%
and there were a lot of folks who are engaged

1149
01:06:53.719 --> 01:06:56.530 A:middle L:90%
, I guess, in there in in the museum

1150
01:06:56.530 --> 01:06:59.960 A:middle L:90%
itself, in many cases, but also in issues

1151
01:06:59.960 --> 01:07:03.139 A:middle L:90%
related to whatever the science is talking about. Now

1152
01:07:03.150 --> 01:07:05.750 A:middle L:90%
, that wasn't always the case. We did see

1153
01:07:05.750 --> 01:07:10.340 A:middle L:90%
a pretty wide range of literacy skills in the posts

1154
01:07:10.619 --> 01:07:11.880 A:middle L:90%
. That could have been a function of age.

1155
01:07:11.880 --> 01:07:14.619 A:middle L:90%
It also could have been a function of, um

1156
01:07:14.630 --> 01:07:16.840 A:middle L:90%
where the, uh you know where the folks are

1157
01:07:16.840 --> 01:07:19.579 A:middle L:90%
coming from. They weren't always. You could kind

1158
01:07:19.579 --> 01:07:21.679 A:middle L:90%
of tell that they weren't always, um, first

1159
01:07:21.679 --> 01:07:26.760 A:middle L:90%
language as English speakers. Um, but we also

1160
01:07:26.760 --> 01:07:29.719 A:middle L:90%
didn't have a lot of demographic information because many of

1161
01:07:29.719 --> 01:07:31.670 A:middle L:90%
especially about the, uh, the folks who are

1162
01:07:31.670 --> 01:07:36.190 A:middle L:90%
coming in as guests, they are posting with anonymous

1163
01:07:36.199 --> 01:07:39.739 A:middle L:90%
things. I think I had another one in here

1164
01:07:39.750 --> 01:07:42.170 A:middle L:90%
. Yeah, here's another one. Here's a Here's

1165
01:07:42.170 --> 01:07:54.070 A:middle L:90%
an example. So we're guessing this is someone in

1166
01:07:54.070 --> 01:08:08.570 A:middle L:90%
the UK five pence coin all English speakers to That's

1167
01:08:08.579 --> 01:08:16.479 A:middle L:90%
pretty. That's another limiting factor here. Uh huh

1168
01:08:16.869 --> 01:08:19.689 A:middle L:90%
. In this methodology, you are stemming the words

1169
01:08:19.689 --> 01:08:24.140 A:middle L:90%
before you analyze them, which means not looking at

1170
01:08:24.140 --> 01:08:26.649 A:middle L:90%
them in terms of parts of speech, Whether this

1171
01:08:26.649 --> 01:08:28.899 A:middle L:90%
is a noun or this is a verb, what

1172
01:08:28.899 --> 01:08:30.949 A:middle L:90%
sort of losses and gains would be involved with running

1173
01:08:30.960 --> 01:08:34.340 A:middle L:90%
this sort of graphing through part of speech analysis,

1174
01:08:34.710 --> 01:08:38.500 A:middle L:90%
perhaps the ratio of announced the verbs in a in

1175
01:08:38.500 --> 01:08:40.550 A:middle L:90%
an utterance give you a clue as well as to

1176
01:08:40.550 --> 01:08:44.399 A:middle L:90%
whether this is science or not, it might mhm

1177
01:08:45.869 --> 01:08:48.930 A:middle L:90%
we we had early on, we we we.

1178
01:08:48.939 --> 01:08:51.109 A:middle L:90%
So the easy way to say that is we were

1179
01:08:51.109 --> 01:08:57.100 A:middle L:90%
aware of part of speech tagging as a possible preparation

1180
01:08:57.100 --> 01:09:00.369 A:middle L:90%
step. But in working with the with the questions

1181
01:09:00.369 --> 01:09:01.760 A:middle L:90%
that we are working with, we never got to

1182
01:09:01.760 --> 01:09:03.869 A:middle L:90%
a place where it made a whole lot of sense

1183
01:09:03.880 --> 01:09:09.399 A:middle L:90%
to us to do to run that particular procedure,

1184
01:09:09.770 --> 01:09:14.140 A:middle L:90%
but we certainly wouldn't rule it out. And I

1185
01:09:14.140 --> 01:09:17.869 A:middle L:90%
don't know. I can't think off hand. I'm

1186
01:09:17.869 --> 01:09:19.390 A:middle L:90%
pretty good at thinking on my feet. I can't

1187
01:09:19.390 --> 01:09:21.699 A:middle L:90%
think of a way. It would change what we

1188
01:09:21.699 --> 01:09:25.689 A:middle L:90%
have. What I talked about today, Um,

1189
01:09:27.270 --> 01:09:30.779 A:middle L:90%
in part because what were often seeing related to the

1190
01:09:30.789 --> 01:09:35.850 A:middle L:90%
high centrality content words is either it's either a concrete

1191
01:09:35.850 --> 01:09:39.899 A:middle L:90%
or an abstract noun as the topic, Um,

1192
01:09:40.470 --> 01:09:45.000 A:middle L:90%
even if it's Jared. So even if it's fishing

1193
01:09:45.680 --> 01:09:48.189 A:middle L:90%
as opposed to have fish there, they've identified it

1194
01:09:48.189 --> 01:09:51.779 A:middle L:90%
in order to talk about it. And so that's

1195
01:09:51.779 --> 01:09:56.659 A:middle L:90%
a good question. I don't know. Maybe it's

1196
01:09:56.659 --> 01:10:01.470 A:middle L:90%
time for one more question. This is This is

1197
01:10:01.470 --> 01:10:08.770 A:middle L:90%
a big picture question. Your associate graduate school folks

1198
01:10:08.770 --> 01:10:11.859 A:middle L:90%
in humanities, uh, on the writing studies and

1199
01:10:11.859 --> 01:10:13.979 A:middle L:90%
one of the community when the Slams against writing studies

1200
01:10:13.979 --> 01:10:15.170 A:middle L:90%
is that it keeps me part of the towards social

1201
01:10:15.170 --> 01:10:18.840 A:middle L:90%
science and away from the humanities background. And now

1202
01:10:18.840 --> 01:10:21.920 A:middle L:90%
we're talking about computation models of basic science and English

1203
01:10:21.920 --> 01:10:25.319 A:middle L:90%
studies. I wonder if you talk about the pluses

1204
01:10:25.319 --> 01:10:27.699 A:middle L:90%
and minuses in the big picture. Ideologically, politically

1205
01:10:27.710 --> 01:10:30.869 A:middle L:90%
, institutionally, about this sort of move where English

1206
01:10:30.260 --> 01:10:33.720 A:middle L:90%
studies textual studies, one person at a book come

1207
01:10:33.720 --> 01:10:38.920 A:middle L:90%
close reading, moving towards what most would consider science

1208
01:10:39.310 --> 01:10:45.039 A:middle L:90%
for universities. Ideologically, politically, economically. Yeah

1209
01:10:45.050 --> 01:10:46.229 A:middle L:90%
, and we thought I have thought a lot about

1210
01:10:46.229 --> 01:10:51.840 A:middle L:90%
those things. So I'll talk about one thing that

1211
01:10:51.840 --> 01:10:55.289 A:middle L:90%
I see as an unqualified benefit. And that is

1212
01:10:56.060 --> 01:11:00.130 A:middle L:90%
this project has been able to bring to the table

1213
01:11:00.479 --> 01:11:04.130 A:middle L:90%
more diverse, Um, disciplinary perspectives and almost any

1214
01:11:04.130 --> 01:11:08.180 A:middle L:90%
other project that I've that I've been engaged in.

1215
01:11:08.310 --> 01:11:11.560 A:middle L:90%
We can get computer science people. We can get

1216
01:11:11.569 --> 01:11:14.130 A:middle L:90%
who do natural language processing. We can get linguists

1217
01:11:14.130 --> 01:11:15.090 A:middle L:90%
. We can get literature, Folks, we have

1218
01:11:15.090 --> 01:11:19.979 A:middle L:90%
someone at Michigan State who does is interested in,

1219
01:11:20.359 --> 01:11:24.170 A:middle L:90%
uh, what's it called? The style? Um

1220
01:11:24.170 --> 01:11:26.500 A:middle L:90%
, a tree. So author attribution, which is

1221
01:11:26.500 --> 01:11:30.289 A:middle L:90%
kind of the reverse of what we're after, Uh

1222
01:11:30.300 --> 01:11:34.079 A:middle L:90%
, and we there's a there's because it, um

1223
01:11:34.960 --> 01:11:39.560 A:middle L:90%
because it centers around the text in in an interesting

1224
01:11:39.560 --> 01:11:44.100 A:middle L:90%
way. What is satisfying about the way those discussions

1225
01:11:44.100 --> 01:11:48.899 A:middle L:90%
come about is those disciplinary interests are still all sort

1226
01:11:48.899 --> 01:11:53.000 A:middle L:90%
of returning to Well, how would a human makes

1227
01:11:53.000 --> 01:11:55.689 A:middle L:90%
sense of this? Um, the other thing that

1228
01:11:55.689 --> 01:11:58.479 A:middle L:90%
we've tried really hard to do is well, we're

1229
01:11:58.479 --> 01:12:00.289 A:middle L:90%
not necessarily trying to replace the human in all of

1230
01:12:00.289 --> 01:12:03.439 A:middle L:90%
this equation. What we're trying to do is supplement

1231
01:12:03.439 --> 01:12:06.659 A:middle L:90%
what is really impossible for a person to do.

1232
01:12:08.050 --> 01:12:11.670 A:middle L:90%
Uh, but yet which these science museum folks and

1233
01:12:11.670 --> 01:12:15.539 A:middle L:90%
others are feeling increasingly, uh, responsible for doing

1234
01:12:15.539 --> 01:12:16.760 A:middle L:90%
, which is what's going on on all your list

1235
01:12:16.760 --> 01:12:20.069 A:middle L:90%
. Serve. We paid all this money to blog

1236
01:12:20.449 --> 01:12:24.729 A:middle L:90%
. What is the blog doing? And to know

1237
01:12:24.729 --> 01:12:28.119 A:middle L:90%
what the blog is doing is ridiculous on the scale

1238
01:12:28.119 --> 01:12:31.430 A:middle L:90%
of all of these posts, right? So I

1239
01:12:31.439 --> 01:12:33.630 A:middle L:90%
think we talked a little bit about this today.

1240
01:12:33.630 --> 01:12:38.729 A:middle L:90%
Is that the rhetorical question is driving the rest of

1241
01:12:38.729 --> 01:12:41.189 A:middle L:90%
it, and that isn't That hasn't always been the

1242
01:12:41.189 --> 01:12:43.439 A:middle L:90%
case when I've gone to interdisciplinary meetings, but it's

1243
01:12:43.439 --> 01:12:46.220 A:middle L:90%
very exciting for us because, um, the computer

1244
01:12:46.220 --> 01:12:48.090 A:middle L:90%
science folks are able to say, Well, what

1245
01:12:48.090 --> 01:12:50.180 A:middle L:90%
would you What would the theory expect to happen?

1246
01:12:50.180 --> 01:12:54.939 A:middle L:90%
And if I said that the curve looks like this

1247
01:12:54.949 --> 01:12:57.100 A:middle L:90%
, I'll explain what that means. But is that

1248
01:12:57.100 --> 01:12:59.090 A:middle L:90%
the curve you would expect to see? And I

1249
01:12:59.090 --> 01:13:00.159 A:middle L:90%
go, I don't wait. All right, so

1250
01:13:00.649 --> 01:13:02.939 A:middle L:90%
I have to kind of figure that out, but

1251
01:13:02.939 --> 01:13:08.250 A:middle L:90%
we're really trying to draw a shape that is conforming

1252
01:13:08.250 --> 01:13:11.630 A:middle L:90%
to an interpretation of a text. We're just trying

1253
01:13:11.630 --> 01:13:13.340 A:middle L:90%
to do it with lots and lots and lots of

1254
01:13:13.340 --> 01:13:16.770 A:middle L:90%
examples. So I think in that way, um

1255
01:13:17.250 --> 01:13:21.100 A:middle L:90%
, I don't see any threat to any of those

1256
01:13:21.350 --> 01:13:27.489 A:middle L:90%
seats at the table In terms of where to their

1257
01:13:27.489 --> 01:13:30.319 A:middle L:90%
disciplinary expertise. It does occasion a very different way

1258
01:13:30.319 --> 01:13:33.420 A:middle L:90%
of working. Um, we couldn't do this project

1259
01:13:33.420 --> 01:13:35.380 A:middle L:90%
outside of the research center. I don't think,

1260
01:13:35.850 --> 01:13:40.989 A:middle L:90%
um, not because it requires a lot of resources

1261
01:13:40.989 --> 01:13:42.880 A:middle L:90%
to do. It was relatively cheap to do,

1262
01:13:42.880 --> 01:13:44.880 A:middle L:90%
except that we had to. Somebody had to pay

1263
01:13:44.880 --> 01:13:47.479 A:middle L:90%
all those graduate students to find 900 head sentences.

1264
01:13:47.949 --> 01:13:51.380 A:middle L:90%
You didn't think I did that, right? Um

1265
01:13:51.850 --> 01:13:56.689 A:middle L:90%
, okay. But apart from that, these are

1266
01:13:56.689 --> 01:13:58.840 A:middle L:90%
free tools out on the internet, and we're stringing

1267
01:13:58.840 --> 01:14:00.050 A:middle L:90%
together a little python script here and a little.

1268
01:14:00.050 --> 01:14:01.960 A:middle L:90%
Our script here, I forgot, are we did

1269
01:14:01.960 --> 01:14:08.369 A:middle L:90%
some more, um, and so that it wasn't

1270
01:14:08.380 --> 01:14:10.569 A:middle L:90%
the technical infrastructure of the center, but it was

1271
01:14:10.569 --> 01:14:13.119 A:middle L:90%
the social infrastructure of the research center that allows us

1272
01:14:13.119 --> 01:14:15.939 A:middle L:90%
to have these have these projects that go forward.

1273
01:14:15.939 --> 01:14:23.029 A:middle L:90%
I think Bill has about 15 more minutes before we

1274
01:14:23.029 --> 01:14:24.710 A:middle L:90%
need to drag him off to his next meeting.

1275
01:14:24.710 --> 01:14:25.979 A:middle L:90%
I know that some of you need to go,

1276
01:14:25.979 --> 01:14:27.409 A:middle L:90%
but if you would like to stick around and talk

1277
01:14:27.409 --> 01:14:29.189 A:middle L:90%
with him a little bit more, we'd be happy

1278
01:14:29.189 --> 01:14:30.569 A:middle L:90%
to have you do that. We've got additional refreshments

1279
01:14:30.579 --> 01:14:32.489 A:middle L:90%
. We hope to see many of you back here

1280
01:14:32.500 --> 01:14:36.500 A:middle L:90%
on October 2nd. Wednesday at one PM for next

1281
01:14:36.510 --> 01:14:41.619 A:middle L:90%
digital discussions, conversation will be on the big data

1282
01:14:41.630 --> 01:14:46.210 A:middle L:90%
project on the Spanish 1918 Spanish influenza project That's happening

1283
01:14:46.220 --> 01:14:48.189 A:middle L:90%
right here at Virginia Tech. So once again,

1284
01:14:48.199 --> 01:14:49.689 A:middle L:90%
thank you to Bill Hart Davidson.