1 00:00:15,220 --> 00:00:19,340 PATRICK WINSTON: We might wonder a little bit about the 2 00:00:19,340 --> 00:00:24,540 nature of human intelligence, and we might reflect a little 3 00:00:24,540 --> 00:00:26,230 bit on the kind of intelligence we've been 4 00:00:26,230 --> 00:00:28,310 talking about in the past few weeks. 5 00:00:28,310 --> 00:00:31,160 It's been an intelligence of sorts. 6 00:00:31,160 --> 00:00:33,980 Those programs, support vector machines, boosting, they can 7 00:00:33,980 --> 00:00:36,030 do really smart things. 8 00:00:36,030 --> 00:00:38,710 But the peculiar thing about systems that use those methods 9 00:00:38,710 --> 00:00:43,770 is that those systems don't have any idea about what 10 00:00:43,770 --> 00:00:45,260 they're doing. 11 00:00:45,260 --> 00:00:47,380 They don't know anything. 12 00:00:47,380 --> 00:00:49,810 So they don't give us very much of an insight into the 13 00:00:49,810 --> 00:00:52,720 nature of human intelligence. 14 00:00:52,720 --> 00:00:54,500 And, after all, I'd like to have a model of human 15 00:00:54,500 --> 00:00:58,000 intelligence because, let's face it, we're the smartest 16 00:00:58,000 --> 00:00:59,830 things around. 17 00:00:59,830 --> 00:01:02,760 So there are lots of ways to approach that question. 18 00:01:02,760 --> 00:01:08,090 And we'll approach that question first from a 19 00:01:08,090 --> 00:01:10,560 evolutionary point of view. 20 00:01:10,560 --> 00:01:16,990 Some scientists believe, me for instance, that we have a 21 00:01:16,990 --> 00:01:26,280 family tree that looks about like this. 22 00:01:30,110 --> 00:01:33,970 Too small to see much of that, but the main point is that we 23 00:01:33,970 --> 00:01:36,360 haven't been around very long. 24 00:01:36,360 --> 00:01:40,470 We humans have been around maybe 200,000 years, and the 25 00:01:40,470 --> 00:01:44,030 dinosaurs died out 60 million years ago. 26 00:01:44,030 --> 00:01:47,890 So in the blink of an eye, we seem to have, more or less, 27 00:01:47,890 --> 00:01:50,500 taken over. 28 00:01:50,500 --> 00:01:54,590 When you look at that family tree on a scale where you can 29 00:01:54,590 --> 00:01:57,960 see it, one of the characteristics is increasing 30 00:01:57,960 --> 00:01:59,210 brain size. 31 00:02:02,380 --> 00:02:06,980 There we are on the left, chimpanzee on the right. 32 00:02:06,980 --> 00:02:11,300 Clearly, mostly mouth, not to much brain in there. 33 00:02:11,300 --> 00:02:15,040 That one down below is a reconstruction of one of those 34 00:02:15,040 --> 00:02:20,380 pithecus type bipedal apes from about 4 million 35 00:02:20,380 --> 00:02:22,680 years ago or so. 36 00:02:22,680 --> 00:02:25,770 So we became bipedal a long time before we 37 00:02:25,770 --> 00:02:27,810 had much of a brain. 38 00:02:27,810 --> 00:02:30,980 So you might think, well, maybe brain size has a lot to 39 00:02:30,980 --> 00:02:34,079 do with it, and I suppose it does. 40 00:02:34,079 --> 00:02:40,700 So we can plot brain volume of our ancestors versus time. 41 00:02:40,700 --> 00:02:42,950 So the picture I just showed you was from about 3 million 42 00:02:42,950 --> 00:02:45,740 years ago, I guess. 43 00:02:45,740 --> 00:02:48,530 And then on the upper right hand corner, oh, that's not 44 00:02:48,530 --> 00:02:51,500 just us, that's also the Neanderthals. 45 00:02:51,500 --> 00:02:54,750 Their brains might have been slightly bigger than ours. 46 00:02:54,750 --> 00:02:58,020 So it isn't just brain size. 47 00:02:58,020 --> 00:03:00,660 Here's what that guy looks like. 48 00:03:00,660 --> 00:03:04,930 That's a Neanderthal, of course, on the left. 49 00:03:04,930 --> 00:03:06,680 And that's one of us on the right. 50 00:03:06,680 --> 00:03:12,630 Some conspicuous differences, big heads, the rib cage is 51 00:03:12,630 --> 00:03:17,130 kind of conical in shape, a large pelvis. 52 00:03:17,130 --> 00:03:19,670 People like to make a lot of speculations about how they 53 00:03:19,670 --> 00:03:22,031 must have moved around. 54 00:03:22,031 --> 00:03:26,062 But one thing is plain, they didn't amount to much. 55 00:03:26,062 --> 00:03:29,300 They could make stone tools, but those stone tools didn't 56 00:03:29,300 --> 00:03:33,170 change much over tens of thousands of years. 57 00:03:33,170 --> 00:03:36,240 And that was pretty much the story with us too, until 58 00:03:36,240 --> 00:03:40,400 something happened, probably in southern Africa, probably 59 00:03:40,400 --> 00:03:45,500 in a group of individuals, maybe less than 1,000. 60 00:03:45,500 --> 00:03:46,520 And what's the evidence for that? 61 00:03:46,520 --> 00:03:48,840 The evidence for that is mostly-- 62 00:03:48,840 --> 00:03:54,420 comes from DNA studies with a lot of probabilistic 63 00:03:54,420 --> 00:03:57,850 assumptions, and Monte Carlo simulations, but it seems 64 00:03:57,850 --> 00:04:00,640 among the competing hypothesis for how we came to populate 65 00:04:00,640 --> 00:04:03,290 the world, it seems that there was a group of us, homo 66 00:04:03,290 --> 00:04:06,820 sapiens, in Southern Africa that got something that nobody 67 00:04:06,820 --> 00:04:11,920 else had and the highest probability scenario is that 68 00:04:11,920 --> 00:04:14,870 we quickly took over. 69 00:04:14,870 --> 00:04:18,560 That population of homo sapiens dominated the rest, 70 00:04:18,560 --> 00:04:21,970 went out of Africa, and within the blink of an eye did that 71 00:04:21,970 --> 00:04:23,220 sort of stuff. 72 00:04:25,300 --> 00:04:28,330 What's that sort of stuff? 73 00:04:28,330 --> 00:04:31,120 Those two paintings are from Lascaux about 74 00:04:31,120 --> 00:04:35,190 25,000 years ago. 75 00:04:35,190 --> 00:04:37,870 Paleoanthropologists, like Tattersall, take that as plain 76 00:04:37,870 --> 00:04:40,190 evidence that there was symbolic thought on the people 77 00:04:40,190 --> 00:04:43,710 who were around at that time, us homo sapiens. 78 00:04:43,710 --> 00:04:49,120 The head is a carving of a mastodon tusk of a woman 79 00:04:49,120 --> 00:04:51,110 25,000 years ago. 80 00:04:51,110 --> 00:04:56,750 Also, plainly symbolic, people are making a lot of jewelry 81 00:04:56,750 --> 00:04:57,860 and doing self adornment. 82 00:04:57,860 --> 00:05:00,130 The Neanderthals never seemed to do that. 83 00:05:00,130 --> 00:05:03,180 That jewelry making seems to have gone back to Southern 84 00:05:03,180 --> 00:05:06,410 Africa, maybe 70,000 years ago. 85 00:05:06,410 --> 00:05:09,060 People were puncturing seashells and using them as 86 00:05:09,060 --> 00:05:11,990 necklaces, apparently. 87 00:05:11,990 --> 00:05:15,350 So something happened, and the paleoanthropologists, who 88 00:05:15,350 --> 00:05:18,110 write fascinating stuff, don't quite know how to talk about 89 00:05:18,110 --> 00:05:19,360 it other than to say it looks like we 90 00:05:19,360 --> 00:05:21,820 became somehow symbolic. 91 00:05:21,820 --> 00:05:25,780 And somehow that has something to do with language. 92 00:05:25,780 --> 00:05:30,280 So if you talk to Noam Chomsky he will say-- 93 00:05:30,280 --> 00:05:32,020 let me get this precise, this is near his 94 00:05:32,020 --> 00:05:34,450 quotation as I can get. 95 00:05:34,450 --> 00:05:37,920 He thinks it was the ability to take two concepts and put 96 00:05:37,920 --> 00:05:41,400 them together to form a third concept without disturbing the 97 00:05:41,400 --> 00:05:44,850 original concepts and without limit, and each 98 00:05:44,850 --> 00:05:47,030 part of that's important. 99 00:05:47,030 --> 00:05:50,100 The without limit part is what separates us from species that 100 00:05:50,100 --> 00:05:53,659 might be able to do that a little bit, but we can do it 101 00:05:53,659 --> 00:05:57,210 without any apparent limit. 102 00:05:57,210 --> 00:05:58,760 So that's a linguist speaking. 103 00:05:58,760 --> 00:06:00,910 He talks a lot about the merge operation, and 104 00:06:00,910 --> 00:06:03,080 combinators and language-- 105 00:06:03,080 --> 00:06:04,680 using terms foreign to us. 106 00:06:04,680 --> 00:06:06,550 Better not use the term "combinator," it's kind of a 107 00:06:06,550 --> 00:06:08,810 computer science term. 108 00:06:08,810 --> 00:06:13,210 But whatever it is, it seemed to happen about that time. 109 00:06:13,210 --> 00:06:16,980 It didn't happen slowly and proportion to brain size, it 110 00:06:16,980 --> 00:06:20,460 seemed to happen all of a sudden in consequence of a 111 00:06:20,460 --> 00:06:24,520 brain that had grown big enough to be an enablement but 112 00:06:24,520 --> 00:06:27,480 the capability was not what pulled 113 00:06:27,480 --> 00:06:29,710 evolution in that direction. 114 00:06:29,710 --> 00:06:35,000 So, I believe, that whatever that was, that capability, 115 00:06:35,000 --> 00:06:41,540 enabled humans, us humans, to tell and understand stories, 116 00:06:41,540 --> 00:06:45,150 and that's what separates us from the other primates. 117 00:06:45,150 --> 00:06:46,300 That ability to-- 118 00:06:46,300 --> 00:06:51,280 The symbolic ability, whatever it is, enabled storytelling 119 00:06:51,280 --> 00:06:52,070 and understanding. 120 00:06:52,070 --> 00:06:54,770 And that's what all education is about and that's why our 121 00:06:54,770 --> 00:06:56,880 species is special. 122 00:06:56,880 --> 00:07:01,830 So what we're going to talk about today is something you 123 00:07:01,830 --> 00:07:03,720 might think of as an instantiation of that 124 00:07:03,720 --> 00:07:06,740 hypothesis, one way of thinking about it. 125 00:07:06,740 --> 00:07:08,920 And it's a way of thinking about what the linguists would 126 00:07:08,920 --> 00:07:11,090 call the inner language. 127 00:07:11,090 --> 00:07:14,240 It's not the language with which we communicate, it's the 128 00:07:14,240 --> 00:07:18,660 language with which we think, which is closely related to 129 00:07:18,660 --> 00:07:21,840 the language with which we communicate, but may not be 130 00:07:21,840 --> 00:07:24,090 quite the same thing. 131 00:07:24,090 --> 00:07:26,900 So many of you are bilingual. 132 00:07:26,900 --> 00:07:28,200 Chris is bilingual. 133 00:07:28,200 --> 00:07:30,640 Chris, have you ever had the experience of remembering that 134 00:07:30,640 --> 00:07:33,860 someone said something to you, but not remembering what 135 00:07:33,860 --> 00:07:36,658 language they used? 136 00:07:36,658 --> 00:07:37,908 STUDENT: [INAUDIBLE]. 137 00:07:42,450 --> 00:07:43,250 PATRICK WINSTON: How about [? you, Sian, ?] have you had 138 00:07:43,250 --> 00:07:44,677 that experience? 139 00:07:44,677 --> 00:07:45,870 STUDENT: [INAUDIBLE]. 140 00:07:45,870 --> 00:07:46,420 PATRICK WINSTON: What? 141 00:07:46,420 --> 00:07:47,670 STUDENT: I always [INAUDIBLE]. 142 00:07:50,121 --> 00:07:52,490 PATRICK WINSTON: [INAUDIBLE] experience of having-- 143 00:07:52,490 --> 00:07:55,010 David, have you ever had that experience of remembering some 144 00:07:55,010 --> 00:07:57,140 conversation but not remembering the language in 145 00:07:57,140 --> 00:07:58,659 which it was cast? 146 00:07:58,659 --> 00:08:00,046 STUDENT: Well, you remember something you usually don't 147 00:08:00,046 --> 00:08:01,894 know which language it was in anyway. 148 00:08:01,894 --> 00:08:03,880 PATRICK WINSTON: You usually don't remember. 149 00:08:03,880 --> 00:08:05,480 That's a common view. 150 00:08:05,480 --> 00:08:07,110 You remember something was said, that there was a 151 00:08:07,110 --> 00:08:11,640 conversation that had some content, but if it's with a 152 00:08:11,640 --> 00:08:14,250 speaker of your own language and your embedded in another 153 00:08:14,250 --> 00:08:18,340 place, you often don't remember what language the 154 00:08:18,340 --> 00:08:20,110 conversation was in. 155 00:08:20,110 --> 00:08:22,500 Is that right, [? Wana, ?] you remember things like that? 156 00:08:22,500 --> 00:08:23,912 Not sure? 157 00:08:23,912 --> 00:08:25,941 STUDENT: People who [INAUDIBLE] speak in a certain 158 00:08:25,941 --> 00:08:27,520 language, so-- 159 00:08:27,520 --> 00:08:29,060 PATRICK WINSTON: Sometimes you don't have that confusion, she 160 00:08:29,060 --> 00:08:32,400 says, because you always speak to particular people in a 161 00:08:32,400 --> 00:08:33,330 particular language. 162 00:08:33,330 --> 00:08:36,169 But many people report that they have that experience of 163 00:08:36,169 --> 00:08:40,289 not remembering which language something was said in. 164 00:08:40,289 --> 00:08:42,590 Well, OK, so what are we going to do? 165 00:08:42,590 --> 00:08:45,610 We need an inner language, and maybe we can start just by 166 00:08:45,610 --> 00:08:48,690 saying, let's have something that looks sort 167 00:08:48,690 --> 00:08:49,980 of familiar to us. 168 00:08:49,980 --> 00:08:55,290 We have an object and it's supported by some other 169 00:08:55,290 --> 00:08:57,960 objects, so those are support relations. 170 00:09:02,680 --> 00:09:05,850 That's one example of what we might call a semantic net. 171 00:09:05,850 --> 00:09:10,940 It's a network that's got nodes and links, it's got-- 172 00:09:10,940 --> 00:09:12,640 it has some meaning. 173 00:09:12,640 --> 00:09:14,790 That's where the word "semantic" comes from. 174 00:09:14,790 --> 00:09:18,760 Well, we might have another example that looks like this. 175 00:09:18,760 --> 00:09:30,640 There's Macbeth, there's Duncan, and Macbeth murders 176 00:09:30,640 --> 00:09:39,870 Duncan, then we also know, somehow, there's a kill 177 00:09:39,870 --> 00:09:43,720 involved as a consequence, and then ultimately, 178 00:09:43,720 --> 00:09:46,300 Duncan has a property. 179 00:09:53,500 --> 00:09:55,870 And that property is the property of being dead. 180 00:09:58,460 --> 00:10:01,530 So there's another semantic net recording something that 181 00:10:01,530 --> 00:10:06,030 happens in Shakespeare's Macbeth plot. 182 00:10:06,030 --> 00:10:07,880 Now, we can decorate that a little bit. 183 00:10:07,880 --> 00:10:11,200 So as to get a couple of other concepts in play. 184 00:10:11,200 --> 00:10:14,090 First of all, the thing we've got already is we've got 185 00:10:14,090 --> 00:10:15,340 combinators. 186 00:10:22,520 --> 00:10:23,770 Then combinators. 187 00:10:26,570 --> 00:10:30,590 Well, a fancy name for those links that connect the nodes. 188 00:10:30,590 --> 00:10:34,410 Another thing we've got is an opportunity for connecting the 189 00:10:34,410 --> 00:10:36,370 links themselves. 190 00:10:36,370 --> 00:10:41,920 So the murder sort of implies the kill, and the kill leads 191 00:10:41,920 --> 00:10:45,170 us to conclude that the victim is dead. 192 00:10:45,170 --> 00:10:50,720 So that is treating the links themselves as objects that can 193 00:10:50,720 --> 00:10:54,240 be the subject or object of other links. 194 00:10:54,240 --> 00:11:04,630 So we call that process "reification." 195 00:11:04,630 --> 00:11:08,100 Now, in artificial intelligence, semantic nets 196 00:11:08,100 --> 00:11:11,460 we're all over the early work, but if you have a big network 197 00:11:11,460 --> 00:11:14,780 that covers the wall you need some way of putting a 198 00:11:14,780 --> 00:11:17,810 spotlight on some pieces of it. 199 00:11:17,810 --> 00:11:22,580 So Marvin Minsky put a lot of technical content into that 200 00:11:22,580 --> 00:11:26,830 idea and created the notion of-- 201 00:11:26,830 --> 00:11:30,210 a notion that deserves another color here-- 202 00:11:30,210 --> 00:11:33,850 he suggested that we need a localization process, so we 203 00:11:33,850 --> 00:11:38,110 have frames, so-called frames or templates. 204 00:11:38,110 --> 00:11:41,340 And a frame for this murder action might be that there's a 205 00:11:41,340 --> 00:11:51,450 murder frame that has an agent and has a victim, and the 206 00:11:51,450 --> 00:11:59,650 agent is Macbeth, and the victim is Duncan. 207 00:12:02,950 --> 00:12:13,780 So that's a way of putting a localization layer on top of 208 00:12:13,780 --> 00:12:16,230 what we've got so far. 209 00:12:16,230 --> 00:12:18,900 Later on, I'll add sequence to that list. 210 00:12:18,900 --> 00:12:21,820 So this is where it rested for a long time, and in some 211 00:12:21,820 --> 00:12:25,300 sense, still rests there because as soon as you've got 212 00:12:25,300 --> 00:12:27,110 combinators, you've got something 213 00:12:27,110 --> 00:12:30,690 that's pretty much universal. 214 00:12:30,690 --> 00:12:33,690 You can do anything with it. 215 00:12:33,690 --> 00:12:37,060 The trouble is it's sort of down at the bit level. 216 00:12:37,060 --> 00:12:39,390 It's like assembly code. 217 00:12:39,390 --> 00:12:44,230 It doesn't have, as a concept, enough organization to help 218 00:12:44,230 --> 00:12:48,360 you go to the next level of achievement. 219 00:12:48,360 --> 00:12:53,760 There's also a little problem here that deserves also, some 220 00:12:53,760 --> 00:13:00,770 mention, and that is that we have, over this whole thing, 221 00:13:00,770 --> 00:13:02,265 the problem of parasitic semantics. 222 00:13:05,910 --> 00:13:08,695 Parasitic semantics. 223 00:13:15,318 --> 00:13:18,320 A kind of ugliness that surrounds this whole concept 224 00:13:18,320 --> 00:13:21,360 because when we look at a diagram like that, and we say, 225 00:13:21,360 --> 00:13:23,260 oh, Macbeth murdered Duncan, that means 226 00:13:23,260 --> 00:13:24,880 Duncan's the victim. 227 00:13:24,880 --> 00:13:27,720 We know there must have been a motive, maybe Macbeth wanted 228 00:13:27,720 --> 00:13:28,190 to be king. 229 00:13:28,190 --> 00:13:31,810 Well, we know all that stuff, and there's a tendency to 230 00:13:31,810 --> 00:13:35,530 project that knowing into the machine. 231 00:13:35,530 --> 00:13:38,920 If you're going to play with your telephone, please leave. 232 00:13:41,670 --> 00:13:51,870 So if we project meaning into that that's our understanding 233 00:13:51,870 --> 00:13:53,750 that's not the machine's understanding. 234 00:13:53,750 --> 00:13:55,815 So much of the meaning can be said to be parasitic. 235 00:13:58,700 --> 00:14:00,740 We're the parasite, and we're projecting 236 00:14:00,740 --> 00:14:02,325 the into that thing. 237 00:14:02,325 --> 00:14:05,110 Putting that diagram into some machine form doesn't mean the 238 00:14:05,110 --> 00:14:07,710 machine knows anything. 239 00:14:07,710 --> 00:14:10,140 It might be able to conclude some things, but it's 240 00:14:10,140 --> 00:14:13,300 understanding is not grounded in any kind of contact with 241 00:14:13,300 --> 00:14:14,880 the physical world. 242 00:14:14,880 --> 00:14:18,310 So we have to worry a lot about that and where 243 00:14:18,310 --> 00:14:20,460 philosophers would stop there and go off and write a few 244 00:14:20,460 --> 00:14:22,340 books on the subject. 245 00:14:22,340 --> 00:14:24,470 But we're not philosophers so we're going to just mention 246 00:14:24,470 --> 00:14:27,700 the problem and go barreling ahead. 247 00:14:27,700 --> 00:14:31,220 So we need to use this notion of semantic net, and we have 248 00:14:31,220 --> 00:14:33,200 to ask ourselves some questions about what elements 249 00:14:33,200 --> 00:14:35,660 of the inner language are most useful and yet it might be 250 00:14:35,660 --> 00:14:39,960 very complicated, but here's usefulness number one. 251 00:14:39,960 --> 00:14:43,330 The notion of classification. 252 00:14:43,330 --> 00:14:46,660 So we know about stuff, and we know about, 253 00:14:46,660 --> 00:14:48,055 for example, pianos. 254 00:14:52,620 --> 00:14:54,055 And we know about tools. 255 00:14:57,160 --> 00:15:02,300 And we know about maps. 256 00:15:02,300 --> 00:15:06,010 But we know about those things on different levels. 257 00:15:06,010 --> 00:15:11,120 So say I'm thinking about a tool, do you have a very good 258 00:15:11,120 --> 00:15:12,960 image of what I'm talking about? 259 00:15:12,960 --> 00:15:15,690 The answer has to be no because the notion of a tool 260 00:15:15,690 --> 00:15:18,960 is very vague, so it's hard to form a picture of 261 00:15:18,960 --> 00:15:21,360 what that's all about. 262 00:15:21,360 --> 00:15:24,720 On the other hand, if I said I'm thinking about a mac. 263 00:15:24,720 --> 00:15:26,990 Well, this is interesting because there's lexical 264 00:15:26,990 --> 00:15:27,670 ambiguity there. 265 00:15:27,670 --> 00:15:31,970 You don't know if I'm talking about the Apple type Mac or 266 00:15:31,970 --> 00:15:35,400 the apple type Mac, or should I say the 267 00:15:35,400 --> 00:15:38,800 fruit or the computer? 268 00:15:38,800 --> 00:15:43,660 So there's lexical ambiguity there at two levels or more. 269 00:15:43,660 --> 00:15:47,260 But let's fill this in a little bit. 270 00:15:47,260 --> 00:15:49,010 If I know I'm talking about a piano, you can form a picture 271 00:15:49,010 --> 00:15:51,100 of that so that seems to be on a more detailed level where 272 00:15:51,100 --> 00:15:52,750 you can do hallucination. 273 00:15:52,750 --> 00:15:54,720 At a higher level you have just a musical instrument. 274 00:15:58,730 --> 00:16:01,210 And I can give you a tool to think about by writing hammer. 275 00:16:06,260 --> 00:16:08,060 And if I'm going to have a mac, it's 276 00:16:08,060 --> 00:16:10,890 going to be an apple. 277 00:16:10,890 --> 00:16:15,480 And in this case, I want you to think about a fruit. 278 00:16:15,480 --> 00:16:17,650 And down here, I can be more specific about 279 00:16:17,650 --> 00:16:19,850 these things too. 280 00:16:19,850 --> 00:16:24,330 I can add a slight refinement of detail and say, I'm 281 00:16:24,330 --> 00:16:25,723 thinking about one of these. 282 00:16:25,723 --> 00:16:28,014 Do you know what this is? 283 00:16:28,014 --> 00:16:29,490 STUDENT: [INAUDIBLE]. 284 00:16:29,490 --> 00:16:32,070 PATRICK WINSTON: No, it's not a mere hammer, it's a 285 00:16:32,070 --> 00:16:34,590 ball-peen hammer. 286 00:16:34,590 --> 00:16:36,290 In some circles, it's called a ladies hammer. 287 00:16:36,290 --> 00:16:38,600 I don't know why. 288 00:16:38,600 --> 00:16:40,050 But it's-- 289 00:16:40,050 --> 00:16:42,580 What's it for? 290 00:16:42,580 --> 00:16:44,420 Most people buy it mostly because it's 291 00:16:44,420 --> 00:16:46,060 small and light weight. 292 00:16:46,060 --> 00:16:48,570 But, in fact, it's for metal working. 293 00:16:48,570 --> 00:16:50,810 It's for taking a piece of sheet metal and pounding it 294 00:16:50,810 --> 00:16:52,630 out into an ashtray or something 295 00:16:52,630 --> 00:16:53,870 or for seating rivets. 296 00:16:53,870 --> 00:16:56,530 It's a metal worker's hammer. 297 00:16:56,530 --> 00:16:58,450 So you might not have known about that before but now, at 298 00:16:58,450 --> 00:17:00,300 least, you have a word to hang that knowledge on. 299 00:17:00,300 --> 00:17:01,620 It's a ball peen hammer. 300 00:17:06,010 --> 00:17:10,059 So we have various levels here, going from very specific 301 00:17:10,059 --> 00:17:11,800 to very general. 302 00:17:11,800 --> 00:17:15,520 And we can even go to a level of specificity for pianos by 303 00:17:15,520 --> 00:17:16,770 saying we've got a Bosendorfer. 304 00:17:23,430 --> 00:17:24,680 Why is a Bosendorfer special? 305 00:17:27,950 --> 00:17:31,240 I mean, is it like a Baldwin? 306 00:17:31,240 --> 00:17:33,010 Something's [INAUDIBLE], Yoka-- 307 00:17:37,150 --> 00:17:38,710 Lots of piano types, what's special? 308 00:17:38,710 --> 00:17:42,110 You see, you don't know because unless you play the 309 00:17:42,110 --> 00:17:45,290 piano, and probably unless you're a serious piano player 310 00:17:45,290 --> 00:17:47,082 you don't know that a Bosendorfer-- 311 00:17:47,082 --> 00:17:48,658 Ariel, you know. 312 00:17:48,658 --> 00:17:51,950 STUDENT: I think its supposed to have an extra octave at the 313 00:17:51,950 --> 00:17:52,610 bottom, black keys. 314 00:17:52,610 --> 00:17:53,600 Pretty cool. 315 00:17:53,600 --> 00:17:56,580 PATRICK WINSTON: It's got some extra keys at the bottom. 316 00:17:56,580 --> 00:17:58,240 And most people don't know that unless they're serious 317 00:17:58,240 --> 00:17:59,350 about the piano. 318 00:17:59,350 --> 00:18:01,160 Some professional piano players, when they're 319 00:18:01,160 --> 00:18:04,110 confronted with a Bosendorfer have to have someone cover 320 00:18:04,110 --> 00:18:07,790 those keys because it screws up their peripheral vision, 321 00:18:07,790 --> 00:18:09,420 and hit the wrong key. 322 00:18:09,420 --> 00:18:10,410 Because they're not used to having those 323 00:18:10,410 --> 00:18:12,780 extra keys at the bottom. 324 00:18:12,780 --> 00:18:15,365 So that's a little detail but the Bosendorfer. 325 00:18:15,365 --> 00:18:19,470 So you can make a kind of graph, and you can say, let's 326 00:18:19,470 --> 00:18:31,140 go from low, very general, to a basic level, 327 00:18:31,140 --> 00:18:32,580 to a specific level. 328 00:18:36,170 --> 00:18:41,550 So it is the case in human knowledge that that graph has 329 00:18:41,550 --> 00:18:43,920 a tendency to look sort of like this. 330 00:18:49,030 --> 00:18:53,650 So here's tool, here's hammer, here's ball peen. 331 00:18:53,650 --> 00:18:57,540 So that level, where you have a big jump, that's the general 332 00:18:57,540 --> 00:18:58,840 to basic level of transition. 333 00:19:01,580 --> 00:19:04,070 So that basic level is probably there because that's 334 00:19:04,070 --> 00:19:08,090 the level on which we hang a huge amount of our knowledge. 335 00:19:08,090 --> 00:19:12,200 We know a lot about pianos, and it all seems to be hanging 336 00:19:12,200 --> 00:19:14,070 on that word piano, which gives us 337 00:19:14,070 --> 00:19:16,980 power with the concept. 338 00:19:16,980 --> 00:19:20,680 So that's example number one of an element 339 00:19:20,680 --> 00:19:21,600 of our inner language. 340 00:19:21,600 --> 00:19:25,860 The ability to assemble things into hierarchies like that, 341 00:19:25,860 --> 00:19:30,740 and hang knowledge about those objects on the elements in 342 00:19:30,740 --> 00:19:31,990 that hierarchy. 343 00:19:35,220 --> 00:19:37,790 Well, given that you have elements in the hierarchy, how 344 00:19:37,790 --> 00:19:38,650 do you talk about them? 345 00:19:38,650 --> 00:19:41,740 Well, I like to consider the possibility, just for the sake 346 00:19:41,740 --> 00:19:44,750 of illustration, that you're thinking about a car crashing 347 00:19:44,750 --> 00:19:46,750 into a wall. 348 00:19:46,750 --> 00:19:50,220 So you've got things to think about like the speed of the 349 00:19:50,220 --> 00:20:04,705 car, the distance to the wall, and the condition of the car. 350 00:20:08,790 --> 00:20:12,110 And you've got the period before the crash, during the 351 00:20:12,110 --> 00:20:14,650 crash, and after the crash. 352 00:20:14,650 --> 00:20:18,390 So you might want to think about how to talk about those 353 00:20:18,390 --> 00:20:20,725 objects in those three time periods. 354 00:20:23,790 --> 00:20:30,410 So we can do that with a vocabulary of change and we do 355 00:20:30,410 --> 00:20:34,400 that because we believe that most of human thinking is 356 00:20:34,400 --> 00:20:37,360 thinking about change causing change. 357 00:20:37,360 --> 00:20:41,070 And that flies in the face of what we learned as engineers. 358 00:20:41,070 --> 00:20:44,995 Because in engineering, we learn about state, and once 359 00:20:44,995 --> 00:20:47,900 you know the state of the system, you know everything 360 00:20:47,900 --> 00:20:50,020 you need to know in order to predict the future. 361 00:20:50,020 --> 00:20:53,640 The trouble is, in our heads, thinking about everything 362 00:20:53,640 --> 00:20:56,007 there is in the world, including the current phase of 363 00:20:56,007 --> 00:20:58,530 the moon, is too much stuff. 364 00:20:58,530 --> 00:21:03,950 So mostly, our thinking, we think, is hinging on the idea 365 00:21:03,950 --> 00:21:06,530 that change leads to change. 366 00:21:06,530 --> 00:21:09,620 So that's why we have a vocabulary of change. 367 00:21:09,620 --> 00:21:14,460 So in the period before the crash, the speed of the car is 368 00:21:14,460 --> 00:21:15,090 not changing. 369 00:21:15,090 --> 00:21:18,600 There's a little notation for not change, no delta. 370 00:21:18,600 --> 00:21:22,790 The distance to the wall, that's decreasing. 371 00:21:22,790 --> 00:21:27,370 The condition of the car, that's not changing. 372 00:21:27,370 --> 00:21:30,210 Then, the car hits the wall. 373 00:21:30,210 --> 00:21:33,810 So the speed of the car disappears, the distance to 374 00:21:33,810 --> 00:21:37,950 the wall disappears, and the condition of the car will 375 00:21:37,950 --> 00:21:40,010 change dramatically. 376 00:21:40,010 --> 00:21:44,110 Finally, after the crash is over, the speed of the car 377 00:21:44,110 --> 00:21:45,740 does not appear. 378 00:21:45,740 --> 00:21:50,520 The distance to the wall does not change, and the condition 379 00:21:50,520 --> 00:21:53,550 of the car also does not change. 380 00:21:53,550 --> 00:21:57,020 So that's hinting at a vocabulary of change, and its 381 00:21:57,020 --> 00:22:00,470 use, which will be the second element in our development of 382 00:22:00,470 --> 00:22:04,950 a vocabulary of ways in which we might have constructed are 383 00:22:04,950 --> 00:22:06,700 inner language. 384 00:22:06,700 --> 00:22:10,800 So this a particular idea, that's classification. 385 00:22:10,800 --> 00:22:17,930 This is transition, and a system that purports to 386 00:22:17,930 --> 00:22:20,850 understand stories with a heavy emphasis on this notion 387 00:22:20,850 --> 00:22:25,450 of transition and we believe, that is to say, I think, that 388 00:22:25,450 --> 00:22:31,880 that vocabulary has to have decrease, increase, change, 389 00:22:31,880 --> 00:22:35,000 appear, and disappear. 390 00:22:35,000 --> 00:22:39,910 So there are 10 things you can have in such diagrams. 391 00:22:39,910 --> 00:22:41,070 I've done five. 392 00:22:41,070 --> 00:22:47,110 That's because, for each of those, there's a not 393 00:22:47,110 --> 00:22:50,070 variation on that. 394 00:22:50,070 --> 00:22:52,270 So with a vocabulary of 10 things can go a long way 395 00:22:52,270 --> 00:22:55,250 toward helping to describe things that are in process of 396 00:22:55,250 --> 00:22:56,980 change and making transition. 397 00:22:56,980 --> 00:22:59,200 And we have a lot of those words in our vocabulary. 398 00:22:59,200 --> 00:23:00,680 We use those words a lot in our vocabulary. 399 00:23:00,680 --> 00:23:03,380 They seem heavily connected with vision. 400 00:23:03,380 --> 00:23:05,700 Our friend appeared. 401 00:23:05,700 --> 00:23:07,580 The cat disappeared. 402 00:23:07,580 --> 00:23:09,910 The speed increased. 403 00:23:09,910 --> 00:23:11,675 So this is a description of a crash. 404 00:23:16,090 --> 00:23:20,430 In terms of those kinds of elements, now, I say to you, 405 00:23:20,430 --> 00:23:23,410 how does a camera work? 406 00:23:23,410 --> 00:23:31,160 Well, I could say the camera works because a photon crashes 407 00:23:31,160 --> 00:23:32,410 into a photo receptor. 408 00:23:34,700 --> 00:23:37,450 So when I say a photon crashes into a photo receptors, why am 409 00:23:37,450 --> 00:23:40,230 I saying that, and how does it help? 410 00:23:40,230 --> 00:23:42,890 I'm saying that and it helps because it's the same pattern 411 00:23:42,890 --> 00:23:46,390 of change you already know about when you talk about a 412 00:23:46,390 --> 00:23:48,560 car crashing into a wall. 413 00:23:48,560 --> 00:23:50,180 How does that work? 414 00:23:50,180 --> 00:24:03,790 The speed of the photon, the distance of the photon to the 415 00:24:03,790 --> 00:24:09,470 receptor, and the condition of the receptor. 416 00:24:13,680 --> 00:24:16,720 So analogies like that are very much of the core of what 417 00:24:16,720 --> 00:24:18,700 we think about all the time. 418 00:24:18,700 --> 00:24:23,630 Really then, there is representation number two. 419 00:24:23,630 --> 00:24:27,040 Number one is class, number two is transition, and now, 420 00:24:27,040 --> 00:24:33,600 you're ready for number three, which is trajectory. 421 00:24:33,600 --> 00:24:37,960 Linguists, who study sentences, often talk in terms 422 00:24:37,960 --> 00:24:41,430 of fundamental patterns that seemed to be in a lot of what 423 00:24:41,430 --> 00:24:45,670 we say, and a lot of what we say is about objects moving 424 00:24:45,670 --> 00:24:46,920 along trajectories. 425 00:24:49,520 --> 00:24:51,375 So we can talk about a trajectory frame. 426 00:25:00,680 --> 00:25:02,650 And a trajectory frame will have elements like this. 427 00:25:02,650 --> 00:25:05,750 It has an object moving on a trajectory that ends up at a 428 00:25:05,750 --> 00:25:07,000 destination. 429 00:25:11,000 --> 00:25:12,715 You might start out at a source. 430 00:25:16,450 --> 00:25:21,970 It's probably been arranged by some kind of agent, and the 431 00:25:21,970 --> 00:25:26,840 agent may assist himself making the motion happen with 432 00:25:26,840 --> 00:25:28,090 some kind of instrument. 433 00:25:33,700 --> 00:25:35,300 There might be somebody helping out 434 00:25:35,300 --> 00:25:37,030 over here, a co-agent. 435 00:25:41,490 --> 00:25:42,620 Well, what else can we have? 436 00:25:42,620 --> 00:25:47,235 A beneficiary, someone is helped out by the action. 437 00:25:55,570 --> 00:25:59,240 Sometimes, the motion is arranged by a conveyance. 438 00:26:05,450 --> 00:26:12,180 So these are a lot of slots, finite slots, and descriptions 439 00:26:12,180 --> 00:26:14,460 of actions, many of which involve motion on a 440 00:26:14,460 --> 00:26:16,300 trajectory. 441 00:26:16,300 --> 00:26:19,650 We have a tendency in language to decorate these things in 442 00:26:19,650 --> 00:26:22,540 one way or another, depending on the language. 443 00:26:22,540 --> 00:26:24,690 And so in many languages, the decoration is by way of 444 00:26:24,690 --> 00:26:26,350 position in the sentence. 445 00:26:26,350 --> 00:26:29,400 In English, it's often by way of a preposition. 446 00:26:29,400 --> 00:26:39,430 It's used to help zero in on a particular role of an object 447 00:26:39,430 --> 00:26:42,550 in the trajectory scenario. 448 00:26:42,550 --> 00:26:49,110 So if I say, I baked a cake with a friend, there's a with 449 00:26:49,110 --> 00:26:51,040 preposition. 450 00:26:51,040 --> 00:26:54,750 If I baked a cake for a friend, the friend is the 451 00:26:54,750 --> 00:26:56,160 beneficiary. 452 00:26:56,160 --> 00:27:01,390 If I baked a cake with an oven, that's an instrument. 453 00:27:01,390 --> 00:27:08,480 The object may be moving to a destination from a source, and 454 00:27:08,480 --> 00:27:11,460 if I'm going to New York by train, I put a 455 00:27:11,460 --> 00:27:14,080 by on top of that. 456 00:27:14,080 --> 00:27:16,290 If the agent isn't in subject position, I would say 457 00:27:16,290 --> 00:27:18,700 something like, oh, I see all the work 458 00:27:18,700 --> 00:27:21,520 was done by a student. 459 00:27:21,520 --> 00:27:23,950 So those prepositions have a tendency to help us zoom in on 460 00:27:23,950 --> 00:27:27,130 the actual role of particular objects in this whole package, 461 00:27:27,130 --> 00:27:29,630 the whole frame. 462 00:27:29,630 --> 00:27:31,300 So this is number three. 463 00:27:34,320 --> 00:27:36,490 There's a variation on this in which there's no actual 464 00:27:36,490 --> 00:27:39,930 trajectory in which case we'll just call that a role frame. 465 00:27:39,930 --> 00:27:41,780 Because if there's no trajectory, we can still have 466 00:27:41,780 --> 00:27:44,915 things such as an instrument, a co-agent, and a beneficiary. 467 00:27:47,450 --> 00:27:50,120 So now, we've got three representations. 468 00:27:50,120 --> 00:27:51,410 You might say, well, what good are they? 469 00:27:51,410 --> 00:27:53,680 And you can determine what good they are these days 470 00:27:53,680 --> 00:27:57,460 because it's easier to go over established corpuses, and say, 471 00:27:57,460 --> 00:28:01,700 what fraction of those, of the sentences, in such a corpus 472 00:28:01,700 --> 00:28:07,660 involve classification or a transition or a trajectory. 473 00:28:07,660 --> 00:28:10,170 The most well known of these is the so-called Wall Street 474 00:28:10,170 --> 00:28:10,810 Journal Corpus. 475 00:28:10,810 --> 00:28:14,790 It has 50,000 sentences in it, drawn from some period of 476 00:28:14,790 --> 00:28:22,700 time, all the syntactical language types work with that 477 00:28:22,700 --> 00:28:24,530 corpus a great deal. 478 00:28:24,530 --> 00:28:26,570 And we worked with it a little bit too, to see what fraction 479 00:28:26,570 --> 00:28:31,210 of the sentences or what the density of trajectories and 480 00:28:31,210 --> 00:28:34,170 transitions are in those sentences. 481 00:28:34,170 --> 00:28:38,010 So I have to say that a little more carefully, because the 482 00:28:38,010 --> 00:28:48,730 finding is that in 100 sentences you'll find about 25 483 00:28:48,730 --> 00:28:52,120 transitions or trajectories. 484 00:28:52,120 --> 00:28:54,590 So they're very densely represented. 485 00:28:54,590 --> 00:28:56,270 They're often very abstract. 486 00:28:56,270 --> 00:29:05,470 Prices rose, the economy went to someplace, but there are 487 00:29:05,470 --> 00:29:08,100 still words that denote transition or trajectory. 488 00:29:10,660 --> 00:29:14,800 Of course, once you have all this stuff, that you have then 489 00:29:14,800 --> 00:29:17,110 have a desire to put it together. 490 00:29:17,110 --> 00:29:22,810 So the next thing we need to talk about is story sequences. 491 00:29:22,810 --> 00:29:27,200 So a story sequence can be a single sentence, and I want to 492 00:29:27,200 --> 00:29:30,135 illustrate that with one of my favorites. 493 00:29:35,050 --> 00:29:36,300 Here's the sentences. 494 00:29:48,220 --> 00:29:50,915 I think I've chosen a gender neutral name so as not getting 495 00:29:50,915 --> 00:29:52,270 in any trouble. 496 00:29:52,270 --> 00:29:56,600 So Pat, but I don't call myself Pat because I decided 497 00:29:56,600 --> 00:29:58,570 when I was 18 years old that pat is a unit 498 00:29:58,570 --> 00:30:00,470 of measure for butter. 499 00:30:00,470 --> 00:30:02,550 In any case, with Pat comforted, Mary, do you have 500 00:30:02,550 --> 00:30:06,100 an image of what happened? 501 00:30:06,100 --> 00:30:08,940 Probably not a very firm image. 502 00:30:08,940 --> 00:30:13,520 You know that Pat did something, but you don't know 503 00:30:13,520 --> 00:30:15,790 exactly what. 504 00:30:15,790 --> 00:30:19,150 Nevertheless, when Pat comforted Chris, you can 505 00:30:19,150 --> 00:30:26,000 construct something that looks like a role frame. 506 00:30:26,000 --> 00:30:32,930 Because the role frame for that would have an agent, and 507 00:30:32,930 --> 00:30:35,840 that would be Pat. 508 00:30:35,840 --> 00:30:37,090 There's an action. 509 00:30:39,740 --> 00:30:41,760 We're going to put a question mark in there because we don't 510 00:30:41,760 --> 00:30:44,760 have a very firm image of what the action is. 511 00:30:44,760 --> 00:30:51,190 Then again, we're building a wall frame, like so. 512 00:30:51,190 --> 00:30:52,440 The object is Chris. 513 00:30:56,230 --> 00:30:59,140 Oh, you know that is the object. 514 00:31:08,160 --> 00:31:09,710 Is there anything else we can say? 515 00:31:09,710 --> 00:31:12,820 Oh, yes, we can probably say something more. 516 00:31:12,820 --> 00:31:15,630 Something else comes to mind when you see 517 00:31:15,630 --> 00:31:17,910 Pat comforted Chris. 518 00:31:17,910 --> 00:31:23,120 There's a sort of result, and the result is 519 00:31:23,120 --> 00:31:24,370 a transition frame. 520 00:31:28,515 --> 00:31:36,370 And the transition frame involves an object, which is 521 00:31:36,370 --> 00:31:50,570 Chris, and Chris has a mood, which presumably, is improved. 522 00:31:50,570 --> 00:31:53,460 It goes up. 523 00:31:53,460 --> 00:31:54,855 Did you have something, Elliott? 524 00:31:54,855 --> 00:31:58,320 STUDENT: Could you, I guess, analogize the Pat comforted 525 00:31:58,320 --> 00:32:04,260 Chris to something like Pat gave comfort to Chris as Chris 526 00:32:04,260 --> 00:32:06,240 is the destination? 527 00:32:06,240 --> 00:32:08,715 And couldn't this be [INAUDIBLE]? 528 00:32:08,715 --> 00:32:11,550 PATRICK WINSTON: Elliott is wandering into a very 529 00:32:11,550 --> 00:32:14,250 interesting area having to do with, couldn't you think about 530 00:32:14,250 --> 00:32:17,860 this in another way and think of it as something moving 531 00:32:17,860 --> 00:32:19,230 along a trajectory. 532 00:32:19,230 --> 00:32:24,600 Comfort is moving, if not from Pat, at least to Chris. 533 00:32:24,600 --> 00:32:26,610 And that's a very important kind of observation because 534 00:32:26,610 --> 00:32:30,220 what it would suggest is there can be a utility in thinking 535 00:32:30,220 --> 00:32:33,300 of things in multiple ways, multiple representations. 536 00:32:33,300 --> 00:32:37,040 Marvin Minsky has a wonderful aphoristic phrase, which is, 537 00:32:37,040 --> 00:32:40,700 if you can only think about something in one way, you have 538 00:32:40,700 --> 00:32:43,160 no recourse when you get stuck. 539 00:32:43,160 --> 00:32:46,630 So multiple representations mean you have multiple ways of 540 00:32:46,630 --> 00:32:50,340 gathering regularity from the world, and collecting it and 541 00:32:50,340 --> 00:32:52,250 therefore that'll make you smarter. 542 00:32:52,250 --> 00:32:56,080 So yes, you could do that, and that would be a compliment to 543 00:32:56,080 --> 00:32:58,020 what I'm doing now. 544 00:32:58,020 --> 00:32:59,490 Let me continue what I'm doing now. 545 00:32:59,490 --> 00:33:01,780 So what have I done? 546 00:33:01,780 --> 00:33:05,720 I've got a role frame and a tradition frame, and the 547 00:33:05,720 --> 00:33:09,550 transition frame is the target of the result 548 00:33:09,550 --> 00:33:12,350 slot in the role frame. 549 00:33:15,110 --> 00:33:17,230 Now, we can modify this a little bit. 550 00:33:20,440 --> 00:33:24,060 And maybe want to say, instead of comforted, terrorized. 551 00:33:30,840 --> 00:33:33,490 And how would that change things? 552 00:33:33,490 --> 00:33:37,490 We don't know exactly what Pat did so the 553 00:33:37,490 --> 00:33:40,180 action remains unknown. 554 00:33:40,180 --> 00:33:42,670 The agent and the object are the same. 555 00:33:42,670 --> 00:33:45,010 But the result here is presumably that 556 00:33:45,010 --> 00:33:49,590 the mood went down. 557 00:33:49,590 --> 00:33:51,550 With just what we've got so far, we can answer a lot of 558 00:33:51,550 --> 00:33:52,460 questions, by the way. 559 00:33:52,460 --> 00:33:55,010 Once we've got the sentence understood in these terms, we 560 00:33:55,010 --> 00:33:57,210 can say who did the thing? 561 00:33:57,210 --> 00:33:58,030 What's this all about? 562 00:33:58,030 --> 00:33:59,490 And the answer is Pat. 563 00:33:59,490 --> 00:34:00,330 What did Pat do? 564 00:34:00,330 --> 00:34:02,130 Comfort, terrorized. 565 00:34:02,130 --> 00:34:03,910 Who do they do it too? 566 00:34:03,910 --> 00:34:05,310 Who was the object? 567 00:34:05,310 --> 00:34:06,780 It's Chris. 568 00:34:06,780 --> 00:34:08,350 What was the result? 569 00:34:08,350 --> 00:34:09,949 Chris felt better. 570 00:34:09,949 --> 00:34:12,260 Chris felt worse. 571 00:34:12,260 --> 00:34:14,639 So these representations already give us a question and 572 00:34:14,639 --> 00:34:22,219 answering capability that makes for an understanding of 573 00:34:22,219 --> 00:34:24,760 the sentence. 574 00:34:24,760 --> 00:34:26,920 But still, we haven't been very specific, so our next 575 00:34:26,920 --> 00:34:31,310 step takes this same sentence in a more specific direction. 576 00:34:31,310 --> 00:34:32,770 So here's a way that goes. 577 00:34:35,300 --> 00:34:40,469 Kissed Chris. 578 00:34:40,469 --> 00:34:44,530 Now, you begin to get a sense of what's going on, you form a 579 00:34:44,530 --> 00:34:47,629 mental image, so you could have a hallucination. 580 00:34:53,920 --> 00:34:57,870 And that hallucinations will also be a kind of frame, but, 581 00:34:57,870 --> 00:35:00,570 in this case, it'll be a trajectory frame. 582 00:35:05,570 --> 00:35:11,475 And the object would be, I don't know, Pat's lips. 583 00:35:16,230 --> 00:35:23,450 And the destination will be Chris' lips. 584 00:35:26,670 --> 00:35:29,490 I don't know, is that right? 585 00:35:29,490 --> 00:35:30,400 It depends. 586 00:35:30,400 --> 00:35:31,350 It depends. 587 00:35:31,350 --> 00:35:34,800 Have you all formed a picture of what's going on here? 588 00:35:34,800 --> 00:35:40,080 So it will be different depending on whether Chris is 589 00:35:40,080 --> 00:35:47,970 Pat's girlfriend or if Chris is Pat's daughter or if Chris 590 00:35:47,970 --> 00:35:53,830 is a frog and Pat is a prince -ess, I guess, the way the 591 00:35:53,830 --> 00:35:55,080 story usually goes. 592 00:35:58,080 --> 00:36:01,080 So somehow we have, in our heads, all kinds of libraries 593 00:36:01,080 --> 00:36:03,490 that help us to form mental pictures of things when we see 594 00:36:03,490 --> 00:36:06,470 things like kissed. 595 00:36:06,470 --> 00:36:10,870 So one final one just to show the variety. 596 00:36:10,870 --> 00:36:13,920 We could say that Pat stabbed Chris. 597 00:36:20,670 --> 00:36:23,410 What changes? 598 00:36:23,410 --> 00:36:28,550 Let's see, in the case of kissed, the mood is going up. 599 00:36:28,550 --> 00:36:31,760 In the case of stabbed, the mood is going down. 600 00:36:31,760 --> 00:36:36,710 You can also probably say that the health is going down. 601 00:36:36,710 --> 00:36:38,750 And the destination is Chris' body. 602 00:36:41,310 --> 00:36:43,700 And the object that's moving is Pat's knife. 603 00:36:47,660 --> 00:36:51,460 We get the same pattern with both of those sentences. 604 00:36:51,460 --> 00:36:55,060 So both of them involve a sequence that starts off with 605 00:36:55,060 --> 00:36:58,080 the action, moves to a transition and a trajectory 606 00:36:58,080 --> 00:37:00,140 and those are all arranged in a line. 607 00:37:00,140 --> 00:37:06,790 And that line is something that gives us a lot of power 608 00:37:06,790 --> 00:37:10,460 over the situation relative to a semantic net. 609 00:37:10,460 --> 00:37:14,360 So I'm going to decorate that one more time, and say that 610 00:37:14,360 --> 00:37:16,580 another element we get out of our 611 00:37:16,580 --> 00:37:19,160 internal language is sequence. 612 00:37:19,160 --> 00:37:24,110 An element we need in order to have anything that looks like 613 00:37:24,110 --> 00:37:27,510 an account of an inner language is sequence. 614 00:37:27,510 --> 00:37:30,370 Because if you think about things being arrayed in a vast 615 00:37:30,370 --> 00:37:32,520 spreading network, it's hard to deal with them. 616 00:37:32,520 --> 00:37:36,320 But if you thing about things being arrayed along a line, in 617 00:37:36,320 --> 00:37:42,490 a sequence of actions or events, like so, then that 618 00:37:42,490 --> 00:37:45,530 imposes enough constraint to get a handle on 619 00:37:45,530 --> 00:37:47,330 what's going on. 620 00:37:47,330 --> 00:37:52,050 So what we're going to call this is the representation, 621 00:37:52,050 --> 00:37:56,100 and I guess I'm up to one, two, three, four. 622 00:37:56,100 --> 00:37:58,130 This is a representation of story sequence. 623 00:38:06,660 --> 00:38:09,420 so even though that's a kind of micro-story, it's still an 624 00:38:09,420 --> 00:38:13,510 example of a story sequence because we get the power out 625 00:38:13,510 --> 00:38:17,406 of it by arranging everything in a line. 626 00:38:17,406 --> 00:38:20,400 You have a sense, I think, especially if you play a 627 00:38:20,400 --> 00:38:24,160 musical instrument, on how dependent we are on sequence. 628 00:38:24,160 --> 00:38:26,580 So if you play a musical instrument, you probably know 629 00:38:26,580 --> 00:38:31,270 how difficult it is to start replaying a piece of music 630 00:38:31,270 --> 00:38:33,380 from the middle of a measure. 631 00:38:33,380 --> 00:38:35,440 You have to go back to at least the beginning of the 632 00:38:35,440 --> 00:38:37,545 measure, and probably to the beginning of the phrase, if 633 00:38:37,545 --> 00:38:39,330 not the beginning of the piece. 634 00:38:39,330 --> 00:38:43,880 So our memory seems to be, at least in music, very rooted in 635 00:38:43,880 --> 00:38:46,250 the idea of sequences. 636 00:38:46,250 --> 00:38:48,475 And that's often true of storytelling too. 637 00:38:48,475 --> 00:38:52,100 We have to go back to the beginning of at least a scene, 638 00:38:52,100 --> 00:38:56,260 because somehow these things are arranged in sequences that 639 00:38:56,260 --> 00:39:03,070 form, somehow, usefulness out of their sequentialness. 640 00:39:03,070 --> 00:39:05,280 So there's one more thing we can talk about and that is the 641 00:39:05,280 --> 00:39:11,920 idea of not just the idea of how these sequences are 642 00:39:11,920 --> 00:39:14,840 constructed and what they're constructed out of, but we can 643 00:39:14,840 --> 00:39:19,300 also talk a little bit in terms of libraries of stories. 644 00:39:27,640 --> 00:39:29,840 And when we talk about libraries of stories, we can 645 00:39:29,840 --> 00:39:33,350 think about kind of the sort of standard stories that we 646 00:39:33,350 --> 00:39:36,290 have and how they're arranged, and how we can know a lot 647 00:39:36,290 --> 00:39:39,840 about something by what it's super class is. 648 00:39:39,840 --> 00:39:43,770 So it's a variation on the theme of learning stuff from 649 00:39:43,770 --> 00:39:45,520 the super classes. 650 00:39:45,520 --> 00:39:47,840 So here's an event frame. 651 00:39:50,780 --> 00:39:54,990 And then, in addition to event frames, 652 00:39:54,990 --> 00:39:56,260 there's disaster frames. 653 00:39:59,700 --> 00:40:00,950 And then there are party frames. 654 00:40:04,320 --> 00:40:11,010 And parties and disasters are both events. 655 00:40:11,010 --> 00:40:13,520 And when we talk about disasters, we know, in turn, 656 00:40:13,520 --> 00:40:15,630 they break up in a variety of things. 657 00:40:15,630 --> 00:40:22,805 We have earthquake disasters, and maybe we have hurricanes. 658 00:40:27,890 --> 00:40:30,440 And in the party world we have, I don't know, birthday 659 00:40:30,440 --> 00:40:36,300 parties, and we have weddings. 660 00:40:41,310 --> 00:40:44,920 And each of these types of frames invites us to fill in 661 00:40:44,920 --> 00:40:46,050 particular slots. 662 00:40:46,050 --> 00:40:47,910 So if we're reading a newspaper story about a 663 00:40:47,910 --> 00:40:51,730 wedding, we know that we're going to be learning the same 664 00:40:51,730 --> 00:40:54,870 sorts of things we will learn about any party except there 665 00:40:54,870 --> 00:40:58,790 is the additional information that we expect that says 666 00:40:58,790 --> 00:41:01,040 something about the bride and the groom. 667 00:41:10,220 --> 00:41:13,150 If we have a raw event and we don't know anything more about 668 00:41:13,150 --> 00:41:14,630 it than that, there's a time and place. 669 00:41:20,480 --> 00:41:23,800 If it's a disaster frame, it gets a grizzly over here, but 670 00:41:23,800 --> 00:41:31,080 if there's a disaster frame, we might have the fatalities, 671 00:41:31,080 --> 00:41:34,800 and how much it cost. 672 00:41:34,800 --> 00:41:40,800 If it's an earthquake frame, we need to know the magnitude 673 00:41:40,800 --> 00:41:42,920 of the quake, and the name of the fault. 674 00:41:46,540 --> 00:41:53,050 If it's a hurricane, we have the category and the name. 675 00:41:56,140 --> 00:42:01,040 So each of these things up there can be viewed, not just 676 00:42:01,040 --> 00:42:05,690 an example of something, but as a new frame all by itself. 677 00:42:05,690 --> 00:42:10,500 As we mature, we have these first four things as building 678 00:42:10,500 --> 00:42:14,470 blocks, and then, we educate ourselves, and we get all 679 00:42:14,470 --> 00:42:16,370 those kinds of frames that help us to 680 00:42:16,370 --> 00:42:19,290 understand the world. 681 00:42:19,290 --> 00:42:24,150 But how to fill those in from a newspaper story can be 682 00:42:24,150 --> 00:42:25,780 sometimes, quite a challenge. 683 00:42:25,780 --> 00:42:29,910 Actually the worst thing to understand is children's 684 00:42:29,910 --> 00:42:36,820 stories because, and this was determined experimentally when 685 00:42:36,820 --> 00:42:39,350 people tried to understand children's stories, because it 686 00:42:39,350 --> 00:42:45,450 turns out that children's stories are not simpler than 687 00:42:45,450 --> 00:42:46,540 the stories we write for adults. 688 00:42:46,540 --> 00:42:50,550 In many cases, they're harder. 689 00:42:50,550 --> 00:42:52,350 If you read about Shakespearean plots, it's all 690 00:42:52,350 --> 00:42:56,360 about intrigue, murder, jealousy, greed, but when 691 00:42:56,360 --> 00:42:58,540 you're trying to write a children's story it can be 692 00:42:58,540 --> 00:43:01,300 about anything. 693 00:43:01,300 --> 00:43:08,200 And worse yet, the children's story often raises problems 694 00:43:08,200 --> 00:43:09,830 that you don't see in newspaper stories. 695 00:43:09,830 --> 00:43:11,080 Let me illustrate that for you. 696 00:43:19,310 --> 00:43:20,560 You want to read that story? 697 00:43:25,080 --> 00:43:27,670 You have no trouble figuring it out, of course, but think 698 00:43:27,670 --> 00:43:30,090 about a poor machine. 699 00:43:30,090 --> 00:43:32,570 It's struggling to understand anything. 700 00:43:32,570 --> 00:43:33,820 What's going to be the problem? 701 00:43:36,010 --> 00:43:37,700 It's going to have trouble figuring out those pronoun 702 00:43:37,700 --> 00:43:39,480 antecedents. 703 00:43:39,480 --> 00:43:39,990 Look at them. 704 00:43:39,990 --> 00:43:41,240 They're complicated. 705 00:43:43,620 --> 00:43:45,750 [MUSIC PLAYING] 706 00:43:45,750 --> 00:43:47,000 PATRICK WINSTON: Shut [INAUDIBLE]. 707 00:43:51,755 --> 00:43:53,005 Oh, that's the way. 708 00:43:56,960 --> 00:43:58,210 [INAUDIBLE]. 709 00:44:00,640 --> 00:44:03,000 Here are the pronouns. 710 00:44:03,000 --> 00:44:05,940 One of them wanted to buy a kite. 711 00:44:05,940 --> 00:44:07,240 He has one, he said. 712 00:44:07,240 --> 00:44:10,230 He will make you take it back. 713 00:44:10,230 --> 00:44:11,030 So that's pretty hard. 714 00:44:11,030 --> 00:44:14,790 Actually, the principal here that I'm driving at is, when 715 00:44:14,790 --> 00:44:17,750 you have a new story or any story, if you have an old 716 00:44:17,750 --> 00:44:19,080 story but you want to read it-- when you want to 717 00:44:19,080 --> 00:44:23,350 understand it quickly, the instantiation of it quickly, 718 00:44:23,350 --> 00:44:26,980 you need to be sure that, if you're the storyteller, you 719 00:44:26,980 --> 00:44:29,650 don't add to the burden of the understanding any syntactic 720 00:44:29,650 --> 00:44:30,840 difficulty. 721 00:44:30,840 --> 00:44:32,790 So that's an example of telling a story with 722 00:44:32,790 --> 00:44:35,470 additional syntactic difficulty. 723 00:44:35,470 --> 00:44:39,360 No newspaper journalist would ever write 724 00:44:39,360 --> 00:44:41,000 the story like that. 725 00:44:41,000 --> 00:44:42,790 Here's how they would write it. 726 00:44:47,760 --> 00:44:50,130 They would even give you a clue that there's certain 727 00:44:50,130 --> 00:44:52,720 information you're never going to get, like 728 00:44:52,720 --> 00:44:54,600 who told the story. 729 00:44:54,600 --> 00:44:58,070 It's that reliable sources business. 730 00:44:58,070 --> 00:45:02,550 So this brings us to the final bit that I want to deal with 731 00:45:02,550 --> 00:45:05,150 today, and what I'm going to do is, if you came into this 732 00:45:05,150 --> 00:45:09,660 class today as a 97 pound writing weakling, you're about 733 00:45:09,660 --> 00:45:14,730 to emerge as a 250 pound mountain of a writer. 734 00:45:14,730 --> 00:45:16,630 Because I want to tell you a few tricks that will make you 735 00:45:16,630 --> 00:45:20,540 usually better as a writer, especially if 736 00:45:20,540 --> 00:45:25,200 you're Russian or German. 737 00:45:25,200 --> 00:45:27,120 And here's how it works. 738 00:45:27,120 --> 00:45:28,370 Rule one. 739 00:45:44,920 --> 00:45:47,640 Because they place additional syntactic burden on the 740 00:45:47,640 --> 00:45:50,520 understanding of the story by the reader. 741 00:45:50,520 --> 00:45:53,450 So if you're telling somebody about some difficult new 742 00:45:53,450 --> 00:45:56,740 technical idea, the last thing you want them to do is to 743 00:45:56,740 --> 00:45:58,990 burden their syntactic processor with figuring out 744 00:45:58,990 --> 00:46:01,040 pronoun antecedents. 745 00:46:01,040 --> 00:46:03,240 So don't use pronouns in you're writing, technical 746 00:46:03,240 --> 00:46:05,800 writing, at least, will be much clearer. 747 00:46:05,800 --> 00:46:08,490 By the way, why does this especially apply to Germans 748 00:46:08,490 --> 00:46:10,230 and Russians? 749 00:46:10,230 --> 00:46:11,980 Is it because-- 750 00:46:11,980 --> 00:46:14,900 is this an ethnic origin slur or is this a 751 00:46:14,900 --> 00:46:16,310 fact about their language? 752 00:46:16,310 --> 00:46:17,370 It's a fact about their language. 753 00:46:17,370 --> 00:46:20,380 Where is the fact? 754 00:46:20,380 --> 00:46:23,570 Why can they get away with pronoun usages that we cannot 755 00:46:23,570 --> 00:46:25,910 get away with in English? 756 00:46:25,910 --> 00:46:26,470 Yes, Andrew? 757 00:46:26,470 --> 00:46:29,802 STUDENT: Gender and also-- 758 00:46:29,802 --> 00:46:35,240 PATRICK WINSTON: Gender, because if you have all of 759 00:46:35,240 --> 00:46:42,530 your nouns decorated with gender, that reduces by three, 760 00:46:42,530 --> 00:46:45,960 the potential for ambiguity in the pronoun antecedent. 761 00:46:45,960 --> 00:46:48,540 So you'll frequently find that German and Russian writers 762 00:46:48,540 --> 00:46:50,970 will have pronouns all over the place that are perfectly 763 00:46:50,970 --> 00:46:52,930 clear to them because of gender. 764 00:46:52,930 --> 00:46:55,170 They are interpretable by English speakers when 765 00:46:55,170 --> 00:46:56,890 translated because we don't have the gender to 766 00:46:56,890 --> 00:46:59,720 help us zero in. 767 00:46:59,720 --> 00:47:03,890 So these things all have to do with minimizing extra, 768 00:47:03,890 --> 00:47:06,170 superfluous, gratuitous, unnecessary 769 00:47:06,170 --> 00:47:07,330 burden on the reader. 770 00:47:07,330 --> 00:47:13,775 Number two is don't use former or latter. 771 00:47:16,950 --> 00:47:19,660 You see those words used frequently in technical 772 00:47:19,660 --> 00:47:21,310 writing, and guess what? 773 00:47:21,310 --> 00:47:24,400 No human being ever encounters those words without having to 774 00:47:24,400 --> 00:47:28,370 stop and go back to figure out what they refer to it. 775 00:47:28,370 --> 00:47:30,970 So that's another example of not placing any unnecessary 776 00:47:30,970 --> 00:47:33,370 syntactic burden on the reader. 777 00:47:33,370 --> 00:47:41,500 And finally, don't call a shovel a spade. 778 00:47:45,000 --> 00:47:47,680 There's a habit, probably instilled by well-meaning but 779 00:47:47,680 --> 00:47:49,500 misadvised high school teachers, that you shouldn't 780 00:47:49,500 --> 00:47:53,530 repeat words, and so people go to great lengths to use some 781 00:47:53,530 --> 00:47:54,830 different word. 782 00:47:54,830 --> 00:47:57,640 The problem is that the reader doesn't know if the shift in 783 00:47:57,640 --> 00:48:01,640 the word is deliberate and attached to some subtle 784 00:48:01,640 --> 00:48:05,320 meaning shift, or if it's just adhering to some high school 785 00:48:05,320 --> 00:48:08,720 teacher's admonition against using the same word again. 786 00:48:08,720 --> 00:48:12,540 So you don't want to say, oh, the right way to dig this 787 00:48:12,540 --> 00:48:15,230 particular hole is with a spade and then switch to a 788 00:48:15,230 --> 00:48:18,350 shovel because the reader can't tell if it's deliberate, 789 00:48:18,350 --> 00:48:21,840 accidental, or a consequence of just the desire not to use 790 00:48:21,840 --> 00:48:23,450 the same word. 791 00:48:23,450 --> 00:48:28,510 So this is how you, with some very simple mechanisms 792 00:48:28,510 --> 00:48:30,660 grounded in AI, you can actually make yourself into a 793 00:48:30,660 --> 00:48:33,280 better writer by avoiding those kinds of things that put 794 00:48:33,280 --> 00:48:35,700 an unnecessary syntactic burden on the people who are 795 00:48:35,700 --> 00:48:36,950 reading your stuff.