1 00:00:00,956 --> 00:00:02,390 [SQUEAKING] 2 00:00:02,390 --> 00:00:03,346 [RUSTLING] 3 00:00:03,346 --> 00:00:10,038 [CLICKING] 4 00:00:10,038 --> 00:00:12,170 NANCY KANWISHER: So this is the line-up for today. 5 00:00:12,170 --> 00:00:14,210 We're going to be talking about language today 6 00:00:14,210 --> 00:00:15,613 and on Wednesday. 7 00:00:15,613 --> 00:00:17,030 But I want to start with something 8 00:00:17,030 --> 00:00:19,898 that I gave very short shrift at the end of lecture last time. 9 00:00:19,898 --> 00:00:21,440 And I'm going to give it short shrift 10 00:00:21,440 --> 00:00:23,720 again, but in a slightly different way. 11 00:00:23,720 --> 00:00:26,540 You'll need this for the reading, which hopefully you've 12 00:00:26,540 --> 00:00:28,820 already tried, started. 13 00:00:28,820 --> 00:00:30,680 Representational similarity analysis 14 00:00:30,680 --> 00:00:33,440 is subtle and rich and interesting. 15 00:00:33,440 --> 00:00:35,990 And it's taken me years of revisiting 16 00:00:35,990 --> 00:00:37,320 it to get its full force. 17 00:00:37,320 --> 00:00:40,100 So just keep going at it and hopefully every time you'll 18 00:00:40,100 --> 00:00:41,430 get it a little better. 19 00:00:41,430 --> 00:00:43,590 So let me try another brief version of this. 20 00:00:43,590 --> 00:00:47,120 So representational similarity analysis 21 00:00:47,120 --> 00:00:51,080 is like a generalized case of multiple voxel pattern 22 00:00:51,080 --> 00:00:53,840 analysis that applies to other methods 23 00:00:53,840 --> 00:00:58,200 and characterizes a bigger conceptual space. 24 00:00:58,200 --> 00:01:01,790 So to remind you, multiple voxel pattern analysis 25 00:01:01,790 --> 00:01:04,220 with functional MRI is this business where 26 00:01:04,220 --> 00:01:06,060 you split your data in half. 27 00:01:06,060 --> 00:01:07,937 So you have one set of scans where 28 00:01:07,937 --> 00:01:10,520 people are looking at, say, dogs and another set where they're 29 00:01:10,520 --> 00:01:13,310 looking at cats, and a whole other separate replication 30 00:01:13,310 --> 00:01:14,942 where they're looking at dogs and cats. 31 00:01:14,942 --> 00:01:16,400 You look at the pattern of response 32 00:01:16,400 --> 00:01:19,940 across voxels in each of those four conditions, dog 1, dog 33 00:01:19,940 --> 00:01:21,530 2, cat 1, cat 2. 34 00:01:21,530 --> 00:01:25,160 And you ask if the pattern is more similar for the two 35 00:01:25,160 --> 00:01:27,680 different splits of the data in the same condition, 36 00:01:27,680 --> 00:01:32,210 dog 1, dog 2, and cat 1, cat 2, the diagonal here, than 37 00:01:32,210 --> 00:01:35,420 in the two cases where they're different, dogs to cats. 38 00:01:35,420 --> 00:01:37,460 Everybody remember that? 39 00:01:37,460 --> 00:01:40,070 If you're having trouble with this, come see me or the TAs. 40 00:01:40,070 --> 00:01:42,500 That's not good. 41 00:01:42,500 --> 00:01:45,410 So now, that's MVPA. 42 00:01:45,410 --> 00:01:48,230 And you can use that to ask of a given region of interest 43 00:01:48,230 --> 00:01:52,370 in the brain or the whole brain, if the pattern of response 44 00:01:52,370 --> 00:01:54,770 in that region can distinguish between class A 45 00:01:54,770 --> 00:01:57,870 and Class B. That's what it's good for. 46 00:01:57,870 --> 00:02:02,220 So that's worth knowing, but it's impoverished. 47 00:02:02,220 --> 00:02:03,270 It's binary. 48 00:02:03,270 --> 00:02:04,770 I mean, cats versus dogs. 49 00:02:04,770 --> 00:02:06,210 It's a dopey example I choose. 50 00:02:06,210 --> 00:02:08,800 But whatever you choose, it's just going to be two things. 51 00:02:08,800 --> 00:02:11,310 It only takes you so far in characterizing what's 52 00:02:11,310 --> 00:02:13,480 represented in that region. 53 00:02:13,480 --> 00:02:15,990 You can make it richer if you force it to generalize. 54 00:02:15,990 --> 00:02:20,430 So if these two are a smaller size and a different viewpoint 55 00:02:20,430 --> 00:02:22,500 from those, and it still works, then we 56 00:02:22,500 --> 00:02:24,000 show that there's generality. 57 00:02:24,000 --> 00:02:26,610 Train on one kind of condition, test on a slightly different 58 00:02:26,610 --> 00:02:27,600 version of them. 59 00:02:27,600 --> 00:02:29,130 That tests the invariants. 60 00:02:29,130 --> 00:02:31,380 That's richer and more interesting. 61 00:02:31,380 --> 00:02:34,860 But even so, it's limited. 62 00:02:34,860 --> 00:02:37,500 So representational similarity analysis 63 00:02:37,500 --> 00:02:41,100 is a bigger, richer way of characterizing representations 64 00:02:41,100 --> 00:02:43,200 by looking at the pattern of response 65 00:02:43,200 --> 00:02:47,400 across multiple conditions, not just two and their variations. 66 00:02:47,400 --> 00:02:49,350 So instead of something like this, 67 00:02:49,350 --> 00:02:51,780 we'd have something like this with a whole bunch 68 00:02:51,780 --> 00:02:53,580 of different stimuli or conditions 69 00:02:53,580 --> 00:02:55,050 that we scan people on. 70 00:02:55,050 --> 00:02:58,080 And then we look at all the pairwise combinations-- 71 00:02:58,080 --> 00:03:01,110 how similar is dog to cat, how similar is it 72 00:03:01,110 --> 00:03:06,240 to pig or horse or table or chair or whatever. 73 00:03:06,240 --> 00:03:09,640 So then we have all of these pairwise similarities, 74 00:03:09,640 --> 00:03:15,030 which gives us a richer idea of what's going on there. 75 00:03:15,030 --> 00:03:18,390 And so now we don't have to choose a binary classification 76 00:03:18,390 --> 00:03:18,990 in there. 77 00:03:18,990 --> 00:03:21,390 We can look at that entire space. 78 00:03:21,390 --> 00:03:26,190 We can think of this whole space as our proxy 79 00:03:26,190 --> 00:03:30,950 for what is represented in that region of the brain. 80 00:03:30,950 --> 00:03:32,540 So now that's cool. 81 00:03:32,540 --> 00:03:35,720 So everybody get the gist of how this set 82 00:03:35,720 --> 00:03:39,350 of pairwise similarities in a region of the brain 83 00:03:39,350 --> 00:03:43,010 is a richer idea of what's going on in that region 84 00:03:43,010 --> 00:03:44,810 and what it cares about? 85 00:03:44,810 --> 00:03:47,120 Everybody got that? 86 00:03:47,120 --> 00:03:51,770 Now, chunk that matrix as one thing. 87 00:03:51,770 --> 00:03:55,640 That's a representation of what's represented 88 00:03:55,640 --> 00:03:57,680 in this part of the brain. 89 00:03:57,680 --> 00:04:01,130 But now we can take that unit and we can say, 90 00:04:01,130 --> 00:04:06,450 we can do the same thing on a totally different kind of data. 91 00:04:06,450 --> 00:04:08,670 So here's what we just did. 92 00:04:08,670 --> 00:04:12,360 Here's like some region of the brain, voxels. 93 00:04:12,360 --> 00:04:15,130 We can do the same thing in behavior. 94 00:04:15,130 --> 00:04:17,820 Now we can say, OK, you rate for me 95 00:04:17,820 --> 00:04:21,810 how similar is a dog to a cat on a scale from one to 10. 96 00:04:21,810 --> 00:04:24,480 I don't know, six or something. 97 00:04:24,480 --> 00:04:28,920 How similar is a cat to a pig? 98 00:04:28,920 --> 00:04:30,840 Four, I don't know. 99 00:04:30,840 --> 00:04:33,455 You can see, you imagine you get some similarity space. 100 00:04:33,455 --> 00:04:34,830 You could just get people to rate 101 00:04:34,830 --> 00:04:39,270 them and you could make a whole new matrix here. 102 00:04:39,270 --> 00:04:42,510 Now you're characterizing your conceptual space 103 00:04:42,510 --> 00:04:44,910 over those same items behaviorally 104 00:04:44,910 --> 00:04:47,430 by asking people how similar each thing are. 105 00:04:47,430 --> 00:04:49,890 Here, we're comparing similarity of patterns 106 00:04:49,890 --> 00:04:51,420 of responses across voxels. 107 00:04:51,420 --> 00:04:54,690 Here, we're doing it by asking how similar it 108 00:04:54,690 --> 00:04:56,490 seems to people behaviorally. 109 00:04:56,490 --> 00:05:01,210 Everybody get how that's a similar kind of enterprise? 110 00:05:01,210 --> 00:05:04,450 Or, we could record from neurons in monkey brains 111 00:05:04,450 --> 00:05:06,550 and show them the same pictures. 112 00:05:06,550 --> 00:05:08,530 And just look at the response across, say, 113 00:05:08,530 --> 00:05:12,070 100 neurons in the monkey brain to a dog and a cat and a pig 114 00:05:12,070 --> 00:05:13,670 and so forth. 115 00:05:13,670 --> 00:05:15,190 And then we could, ask how similar 116 00:05:15,190 --> 00:05:19,960 is a response across neurons in the monkey to each pair 117 00:05:19,960 --> 00:05:26,050 of stimuli, just as we did that across each pair of stimuli 118 00:05:26,050 --> 00:05:27,640 across voxels. 119 00:05:27,640 --> 00:05:29,030 Everybody got that? 120 00:05:29,030 --> 00:05:33,550 So in each case, we're getting a matrix like this. 121 00:05:33,550 --> 00:05:37,060 Now, we can do the totally cool-- oh, sorry, 122 00:05:37,060 --> 00:05:37,970 we're not quite yet. 123 00:05:37,970 --> 00:05:41,110 We can also do that not just on functional MRI 124 00:05:41,110 --> 00:05:43,720 voxels in the whole brain or in one region, 125 00:05:43,720 --> 00:05:45,603 but we can make separate matrices. 126 00:05:45,603 --> 00:05:47,020 These are obviously all fake data. 127 00:05:47,020 --> 00:05:49,187 I didn't take the trouble to make different matrices 128 00:05:49,187 --> 00:05:50,500 for each, right. 129 00:05:50,500 --> 00:05:52,120 But we can make different matrices 130 00:05:52,120 --> 00:05:53,770 for different regions of interest 131 00:05:53,770 --> 00:05:56,380 in the brain, one for each. 132 00:05:56,380 --> 00:05:59,230 Voxels here, what's their pairwise set of similarities 133 00:05:59,230 --> 00:06:00,160 across those stimuli? 134 00:06:00,160 --> 00:06:05,620 Voxels over here, what's their pairwise set of similarities? 135 00:06:05,620 --> 00:06:10,340 Now, we can correlate these matrices to each other. 136 00:06:10,340 --> 00:06:13,190 So we can say, for example, we had 137 00:06:13,190 --> 00:06:15,110 a bunch of people do ratings and give us 138 00:06:15,110 --> 00:06:18,203 their behavioral similarities based over these stimuli. 139 00:06:18,203 --> 00:06:20,120 And then we looked in some region of the brain 140 00:06:20,120 --> 00:06:24,380 and got the brain's similarity space and their responses 141 00:06:24,380 --> 00:06:25,640 across voxels. 142 00:06:25,640 --> 00:06:27,515 How similar are those to each other? 143 00:06:30,800 --> 00:06:33,080 So it's like we've moved up a level. 144 00:06:33,080 --> 00:06:35,660 Each matrix is a set of correlations 145 00:06:35,660 --> 00:06:37,280 between each pair of stimuli. 146 00:06:37,280 --> 00:06:39,560 But then once we have that set of correlations, 147 00:06:39,560 --> 00:06:43,970 we can take the whole matrix and correlate it to another matrix. 148 00:06:43,970 --> 00:06:47,180 This would be a way of asking in some region of the brain 149 00:06:47,180 --> 00:06:52,370 how well does the representation in this chunk of brain match 150 00:06:52,370 --> 00:06:55,580 people's subjective impression of that similarity space 151 00:06:55,580 --> 00:06:57,360 when you ask them about it. 152 00:06:57,360 --> 00:07:02,060 Everybody see how that's a way to ask that question? 153 00:07:02,060 --> 00:07:05,570 We can also relate functional MRI voxels 154 00:07:05,570 --> 00:07:09,410 to neurophysiology responses across neurons. 155 00:07:09,410 --> 00:07:14,330 We can ask how similar is your FFAs-- 156 00:07:14,330 --> 00:07:15,440 let's not take the FFA-- 157 00:07:15,440 --> 00:07:19,700 your LO that likes object shape, how similar is its shape 158 00:07:19,700 --> 00:07:23,210 space in your brain measured with functional MRI 159 00:07:23,210 --> 00:07:26,600 to shape space in this part of the monkey's brain 160 00:07:26,600 --> 00:07:29,760 registered with neurophysiology. 161 00:07:29,760 --> 00:07:32,323 It's pretty cosmic, right. 162 00:07:32,323 --> 00:07:34,740 We're asking if the monkey sees the world the same way you 163 00:07:34,740 --> 00:07:39,090 do, in a sense, for this method, by using these matrices 164 00:07:39,090 --> 00:07:42,360 and asking how similar they are across species and methods. 165 00:07:42,360 --> 00:07:42,915 Yeah? 166 00:07:42,915 --> 00:07:46,575 AUDIENCE: So are the function for [INAUDIBLE] similarity. 167 00:07:46,575 --> 00:07:48,750 All of them are the same or? 168 00:07:48,750 --> 00:07:51,380 NANCY KANWISHER: You could do whatever you like. 169 00:07:51,380 --> 00:07:55,850 So you can do garden variety functional MRI 170 00:07:55,850 --> 00:07:58,280 like we've been talking about in here just like the Haxby 171 00:07:58,280 --> 00:07:59,420 thing from 2001. 172 00:07:59,420 --> 00:08:01,370 That's when it all started, right. 173 00:08:01,370 --> 00:08:05,960 Just get a vector across voxels for one condition, 174 00:08:05,960 --> 00:08:08,240 a vector across voxels for the two condition, 175 00:08:08,240 --> 00:08:09,950 and correlate them. 176 00:08:09,950 --> 00:08:13,160 You can do that in responses across neurons. 177 00:08:13,160 --> 00:08:15,560 But you can also do more exotic things. 178 00:08:15,560 --> 00:08:18,590 You can train a linear classifier on a bunch of voxels 179 00:08:18,590 --> 00:08:22,550 and say, how well can it decode the response to pig 180 00:08:22,550 --> 00:08:24,380 to the response to dog. 181 00:08:24,380 --> 00:08:27,950 And you can put that number in that cell. 182 00:08:27,950 --> 00:08:30,650 So you can do it different ways, any measure of similarity. 183 00:08:30,650 --> 00:08:33,559 Or, very confusingly, there's an increasing trend 184 00:08:33,559 --> 00:08:36,590 to talk about dissimilarity, not similarity, 185 00:08:36,590 --> 00:08:38,750 by subtracting the r values from 1. 186 00:08:38,750 --> 00:08:41,090 I find that annoying, but it's all over the literature. 187 00:08:41,090 --> 00:08:44,059 And who cares whether it's similarity or dissimilarity. 188 00:08:44,059 --> 00:08:45,060 Doesn't really matter. 189 00:08:45,060 --> 00:08:48,600 They're both ways of collecting a representational space. 190 00:08:48,600 --> 00:08:49,346 Yeah? 191 00:08:49,346 --> 00:08:51,735 AUDIENCE: Are there any caveats into the [INAUDIBLE] 192 00:08:51,735 --> 00:08:53,360 that we should be available, since this 193 00:08:53,360 --> 00:08:55,188 is like a correlation of coordination. 194 00:08:55,188 --> 00:08:56,480 NANCY KANWISHER: Oh, a million. 195 00:08:56,480 --> 00:08:58,660 You're supposed to Fisher transform it 196 00:08:58,660 --> 00:08:59,660 and do all that garbage. 197 00:08:59,660 --> 00:09:01,070 And we're not discussing that in here. 198 00:09:01,070 --> 00:09:02,210 I'm just trying to give you the idea. 199 00:09:02,210 --> 00:09:03,710 I don't mean to be dismissive. 200 00:09:03,710 --> 00:09:05,750 I'm skipping over all of that stuff 201 00:09:05,750 --> 00:09:07,940 to just give you the gist of the idea. 202 00:09:07,940 --> 00:09:11,687 For purposes in this class, you could just eyeball 203 00:09:11,687 --> 00:09:12,270 that and that. 204 00:09:12,270 --> 00:09:13,970 And you'd say, oh, they're really ident-- 205 00:09:13,970 --> 00:09:15,053 no, they're not identical. 206 00:09:15,053 --> 00:09:17,340 I guess, I did switch it. 207 00:09:17,340 --> 00:09:18,660 I did switch a few of them, oh. 208 00:09:18,660 --> 00:09:20,400 OK, anyway, whatever. 209 00:09:20,400 --> 00:09:23,040 For purposes in this class, you could just eyeball them. 210 00:09:23,040 --> 00:09:25,290 Mathematically, an r-value-- we're 211 00:09:25,290 --> 00:09:26,670 leaving out all the details. 212 00:09:26,670 --> 00:09:29,040 Yeah, OK. 213 00:09:29,040 --> 00:09:31,800 And, of course, we can compare behavior 214 00:09:31,800 --> 00:09:35,250 in a person to physiology in a monkey, or behavior in a monkey 215 00:09:35,250 --> 00:09:36,990 to physiology in a monkey. 216 00:09:36,990 --> 00:09:39,162 And here's one thing you need for the reading. 217 00:09:39,162 --> 00:09:40,620 I hope it didn't already stump you. 218 00:09:40,620 --> 00:09:43,170 It's in a tiny part of one of the figures. 219 00:09:43,170 --> 00:09:46,410 We could make up a hypothesis of what's represented here. 220 00:09:46,410 --> 00:09:50,940 We might say, hey, consider this patch of brain. 221 00:09:50,940 --> 00:09:53,760 Maybe it represents the animate/inanimate distinction. 222 00:09:53,760 --> 00:09:56,430 In the ideal case, that would mean all 223 00:09:56,430 --> 00:10:00,510 it knows is animals versus non animals. 224 00:10:00,510 --> 00:10:02,070 And so that would mean this should 225 00:10:02,070 --> 00:10:04,502 be the representational similarity space. 226 00:10:04,502 --> 00:10:06,210 If these are all the animals, they're all 227 00:10:06,210 --> 00:10:07,830 exactly the same as each other. 228 00:10:07,830 --> 00:10:10,140 All the non-animals are the same as each other. 229 00:10:10,140 --> 00:10:12,940 But any animal and any non-animal are different. 230 00:10:12,940 --> 00:10:16,440 So this is a hypothesized similarity space 231 00:10:16,440 --> 00:10:18,210 of our guess of what's represented 232 00:10:18,210 --> 00:10:20,460 in a region, a model of what we think 233 00:10:20,460 --> 00:10:22,020 is represented in a region. 234 00:10:22,020 --> 00:10:25,530 And we can correlate that to any of these matrices 235 00:10:25,530 --> 00:10:29,040 to ask whether our hypothesis of what's in there is right. 236 00:10:29,040 --> 00:10:30,910 Does that make sense? 237 00:10:30,910 --> 00:10:32,780 OK. 238 00:10:32,780 --> 00:10:36,410 So why is that so-- oh, this whole thing so totally cool? 239 00:10:36,410 --> 00:10:39,860 It enables us to compare representational spaces 240 00:10:39,860 --> 00:10:43,210 across regions of interest in the brain-- 241 00:10:43,210 --> 00:10:45,940 the FFA to the PPA, do they have similar representational 242 00:10:45,940 --> 00:10:47,050 spaces-- 243 00:10:47,050 --> 00:10:49,480 across subject groups-- this batch 244 00:10:49,480 --> 00:10:51,310 of subjects and that batch of subjects-- 245 00:10:51,310 --> 00:10:54,100 without having to align voxels. 246 00:10:54,100 --> 00:10:55,270 We're not aligning voxels. 247 00:10:55,270 --> 00:10:56,860 We've left voxels behind. 248 00:10:56,860 --> 00:10:58,375 We're only using these matrices. 249 00:11:02,810 --> 00:11:05,930 We can do it across species, across methods, 250 00:11:05,930 --> 00:11:08,660 and across hypothesized models of what we 251 00:11:08,660 --> 00:11:12,760 think is going on, like that. 252 00:11:12,760 --> 00:11:16,360 So more generally, this probes representations 253 00:11:16,360 --> 00:11:17,410 in a richer way. 254 00:11:17,410 --> 00:11:20,080 We don't need to have just 10 or whatever I put there. 255 00:11:20,080 --> 00:11:22,390 We could have, if we keep subjects in the scanner long 256 00:11:22,390 --> 00:11:24,490 enough, or monkeys in the lab long enough, 257 00:11:24,490 --> 00:11:27,550 we can get hundreds of stimuli and really characterize 258 00:11:27,550 --> 00:11:30,400 a rich space. 259 00:11:30,400 --> 00:11:35,030 And we're looking at not just two discriminations, but lots. 260 00:11:35,030 --> 00:11:38,150 The key requirement for representational similarity 261 00:11:38,150 --> 00:11:41,420 analysis, to be able to do all this cool stuff, 262 00:11:41,420 --> 00:11:44,520 is the axes need to be the same. 263 00:11:44,520 --> 00:11:47,750 So the stimuli that you're getting the similarity of 264 00:11:47,750 --> 00:11:50,690 need to be the same in the person doing behavior, 265 00:11:50,690 --> 00:11:55,100 the person doing MRI, the monkey doing physiology, the model. 266 00:11:55,100 --> 00:11:56,870 If the axes are not the same, then there's 267 00:11:56,870 --> 00:11:59,550 no way to correlate the matrices. 268 00:11:59,550 --> 00:12:01,570 Make sense? 269 00:12:01,570 --> 00:12:03,380 We'll keep coming at this again and again. 270 00:12:03,380 --> 00:12:05,320 You'll see it in the paper for tomorrow night. 271 00:12:05,320 --> 00:12:10,140 And we'll come at it again in class on Wednesday. 272 00:12:10,140 --> 00:12:12,480 So that was all catch-up. 273 00:12:12,480 --> 00:12:16,240 So today, we are going to talk about language. 274 00:12:16,240 --> 00:12:20,080 And let's start by reflecting on what an amazing thing 275 00:12:20,080 --> 00:12:21,490 language is. 276 00:12:21,490 --> 00:12:25,610 So right now, there's a miraculous thing going on. 277 00:12:25,610 --> 00:12:31,090 I'm taking some weird, abstract, hard-to-grasp, even for me, 278 00:12:31,090 --> 00:12:33,550 ideas someplace in my head-- 279 00:12:33,550 --> 00:12:36,070 god knows where, somewhere in there-- 280 00:12:36,070 --> 00:12:38,440 and I'm trying to take those ideas 281 00:12:38,440 --> 00:12:42,010 and translate them into this bunch of noises coming out 282 00:12:42,010 --> 00:12:42,760 my mouth. 283 00:12:42,760 --> 00:12:44,230 That's already pretty astonishing. 284 00:12:44,230 --> 00:12:45,260 Like, what? 285 00:12:45,260 --> 00:12:46,780 What does that idea look like? 286 00:12:46,780 --> 00:12:47,710 Who the hell knows? 287 00:12:47,710 --> 00:12:49,720 How do you take an abstract idea and turn it 288 00:12:49,720 --> 00:12:51,250 into a string of sounds? 289 00:12:51,250 --> 00:12:52,600 That's wild. 290 00:12:52,600 --> 00:12:54,610 Nobody really knows pretty much a damn thing 291 00:12:54,610 --> 00:12:57,250 about how that works, fascinating mystery. 292 00:12:57,250 --> 00:12:59,800 But then that bunch of noises is going 293 00:12:59,800 --> 00:13:03,700 through the air and producing, let's hope, pretty similar 294 00:13:03,700 --> 00:13:05,830 ideas in your head. 295 00:13:05,830 --> 00:13:07,622 Wow. 296 00:13:07,622 --> 00:13:08,830 We do this all day every day. 297 00:13:08,830 --> 00:13:09,670 Big deal. 298 00:13:09,670 --> 00:13:11,320 But it's astonishing. 299 00:13:11,320 --> 00:13:15,190 It's just astonishing that that works at all. 300 00:13:15,190 --> 00:13:16,970 So that's the essence of language. 301 00:13:16,970 --> 00:13:19,060 That's why it's so cool. 302 00:13:19,060 --> 00:13:22,580 And let's think about how we're going to think about this. 303 00:13:22,580 --> 00:13:26,650 So the first thing to note is language is universally human. 304 00:13:26,650 --> 00:13:30,820 All neurologically intact humans have language. 305 00:13:30,820 --> 00:13:33,430 There are about 7,000 languages in the world. 306 00:13:33,430 --> 00:13:36,670 Sadly, this number is shrinking all the time. 307 00:13:36,670 --> 00:13:40,930 They are all richly expressive, including sign languages. 308 00:13:40,930 --> 00:13:43,060 There are no kind of impoverished languages 309 00:13:43,060 --> 00:13:45,400 that don't capture the full richness 310 00:13:45,400 --> 00:13:48,100 of expressible human experience. 311 00:13:48,100 --> 00:13:51,490 They're all equally rich. 312 00:13:51,490 --> 00:13:54,540 Language is uniquely human. 313 00:13:54,540 --> 00:13:57,360 Yes, chimps and parrots can accomplish 314 00:13:57,360 --> 00:14:00,180 all kinds of cool things, especially if you train them 315 00:14:00,180 --> 00:14:01,890 extensively. 316 00:14:01,890 --> 00:14:05,550 But what they have is not anything really like language. 317 00:14:05,550 --> 00:14:08,940 And to give you a vivid sense of this, 318 00:14:08,940 --> 00:14:12,240 let's look at Chaser, the Border Collie. 319 00:14:12,240 --> 00:14:14,490 And what I want you to think about as you look at this 320 00:14:14,490 --> 00:14:17,370 little video of Chaser the Border Collie is what is 321 00:14:17,370 --> 00:14:19,680 the difference between your language abilities 322 00:14:19,680 --> 00:14:20,550 and Chaser's. 323 00:14:20,550 --> 00:14:23,070 Chaser is pretty damned impressive, 324 00:14:23,070 --> 00:14:24,420 but you are more impressive. 325 00:14:24,420 --> 00:14:26,460 So watch it and enjoy and think about how it's 326 00:14:26,460 --> 00:14:28,841 different from what you do. 327 00:14:28,841 --> 00:14:29,508 [VIDEO PLAYBACK] 328 00:14:29,508 --> 00:14:32,880 - Some of us burst with pride if our dogs can respond 329 00:14:32,880 --> 00:14:34,930 to two or three commands. 330 00:14:34,930 --> 00:14:37,080 But what if we haven't begun to understand 331 00:14:37,080 --> 00:14:41,340 the possibilities of what the animal mind can really do? 332 00:14:41,340 --> 00:14:44,460 Our friend, astrophysicist Neil deGrasse Tyson, 333 00:14:44,460 --> 00:14:46,530 is host of Nova Science Now. 334 00:14:46,530 --> 00:14:49,833 And he brings us big news from the frontier. 335 00:14:49,833 --> 00:14:51,000 - Walk up, walk up, walk up. 336 00:14:51,000 --> 00:14:54,990 - Meet Chaser, beloved six-year-old Border Collie 337 00:14:54,990 --> 00:14:56,820 of Psychology Professor John Pilley. 338 00:14:56,820 --> 00:14:58,200 - Good girl. 339 00:14:58,200 --> 00:15:01,620 She was born to live in the Scottish mountains-- 340 00:15:01,620 --> 00:15:03,270 Chase, toe, toe, toe-- 341 00:15:03,270 --> 00:15:04,830 and herd sheep. 342 00:15:04,830 --> 00:15:05,370 Go, go. 343 00:15:05,370 --> 00:15:08,220 - John has taught chaser to tend an extremely large. 344 00:15:08,220 --> 00:15:12,280 if unconventional herd, of 1,000 toys. 345 00:15:12,280 --> 00:15:14,520 And she knows the name of every single one of these? 346 00:15:14,520 --> 00:15:15,420 - I hope. 347 00:15:15,420 --> 00:15:18,510 - I find this hard to believe, so I test Chaser's memory 348 00:15:18,510 --> 00:15:19,950 with a random sampling. 349 00:15:19,950 --> 00:15:21,780 Chaser, find Inky. 350 00:15:26,340 --> 00:15:28,800 Well, she got one right. 351 00:15:28,800 --> 00:15:31,560 Find Seal. 352 00:15:31,560 --> 00:15:34,600 Whoa, and that one too. 353 00:15:34,600 --> 00:15:36,930 In fact, she got all nine right. 354 00:15:36,930 --> 00:15:39,720 But what about a new toy she's never seen or heard 355 00:15:39,720 --> 00:15:40,440 the name of? 356 00:15:40,440 --> 00:15:43,950 Chaser's never seen Darwin, hasn't even ever heard the name 357 00:15:43,950 --> 00:15:45,480 Darwin. 358 00:15:45,480 --> 00:15:48,690 So we're going to see if she picks out Darwin by inference. 359 00:15:48,690 --> 00:15:51,220 Find Darwin. 360 00:15:56,390 --> 00:15:58,510 I have to ask her again. 361 00:15:58,510 --> 00:16:04,430 OK, Chaser, Chaser, Chaser, Chaser, find Darwin. 362 00:16:06,950 --> 00:16:08,370 (EXCITEDLY) Darwin! 363 00:16:08,370 --> 00:16:10,050 He's got Darwin! 364 00:16:10,050 --> 00:16:11,220 She did it. 365 00:16:11,220 --> 00:16:13,380 Chaser's never seen that doll before, 366 00:16:13,380 --> 00:16:17,940 yet she settled on the one toy she didn't know by deduction. 367 00:16:17,940 --> 00:16:20,460 It's similar to the way children learn language. 368 00:16:20,460 --> 00:16:23,940 But how does Chaser's ability compare with other species? 369 00:16:23,940 --> 00:16:27,090 Besides us, chimps and bonobos are the animal kingdom's 370 00:16:27,090 --> 00:16:29,460 top linguists, capable of learning sign language, 371 00:16:29,460 --> 00:16:31,350 but very slowly. 372 00:16:31,350 --> 00:16:33,480 They can solve some sophisticated problems, 373 00:16:33,480 --> 00:16:35,730 but they don't always pay close attention to humans. 374 00:16:35,730 --> 00:16:36,355 - Is he coming? 375 00:16:36,355 --> 00:16:39,163 - When I see my dog, my dog wants me to be around. 376 00:16:39,163 --> 00:16:41,080 Whereas a bonobo and chimpanzee, they need me. 377 00:16:41,080 --> 00:16:42,780 They're basically like, hey, you got any food. 378 00:16:42,780 --> 00:16:43,980 Can I get any food off of you? 379 00:16:43,980 --> 00:16:45,765 They're not interested in making me happy. 380 00:16:45,765 --> 00:16:48,090 - Since dogs do like to please us, 381 00:16:48,090 --> 00:16:51,450 that humans need to find a way to tap the potential in all 382 00:16:51,450 --> 00:16:52,710 of our dogs. 383 00:16:52,710 --> 00:16:54,420 OK, put it in the tub. 384 00:16:54,420 --> 00:16:56,490 And dogs like Chaser are just waiting for us 385 00:16:56,490 --> 00:16:58,980 to discover all that they can do. 386 00:16:58,980 --> 00:17:02,100 [GRUNTS] Smart dog. 387 00:17:02,100 --> 00:17:04,440 - And Neil deGrasse Tyson is here 388 00:17:04,440 --> 00:17:07,530 with the astonishing Chaser here. 389 00:17:07,530 --> 00:17:12,264 Tell me what you learned about animal behavior and child 390 00:17:12,264 --> 00:17:12,764 behavior. 391 00:17:12,764 --> 00:17:15,119 - Who would have thought that the animals 392 00:17:15,119 --> 00:17:17,940 are capable of this much display of intellect. 393 00:17:17,940 --> 00:17:21,280 I think we like thinking of humans as top of some ladder 394 00:17:21,280 --> 00:17:23,490 and don't even imagine that other animals could 395 00:17:23,490 --> 00:17:25,230 even approximate what we do. 396 00:17:25,230 --> 00:17:27,390 - All right, I think we all want to see. 397 00:17:27,390 --> 00:17:28,755 - You want the demo. 398 00:17:28,755 --> 00:17:29,380 - Can we do it? 399 00:17:29,380 --> 00:17:30,000 - A demo of this. 400 00:17:30,000 --> 00:17:30,690 - Do you think we can it. 401 00:17:30,690 --> 00:17:30,780 - Sure. 402 00:17:30,780 --> 00:17:31,210 We can try it. 403 00:17:31,297 --> 00:17:32,297 - This is so astounding. 404 00:17:32,297 --> 00:17:33,540 Can we take away the stool. 405 00:17:33,540 --> 00:17:33,795 - Sure. 406 00:17:33,795 --> 00:17:34,290 Let's try this. 407 00:17:34,290 --> 00:17:35,207 - We'll give it a try. 408 00:17:35,207 --> 00:17:35,810 [INAUDIBLE] 409 00:17:35,810 --> 00:17:37,080 Thanks. 410 00:17:37,080 --> 00:17:38,590 All right, so we get down? 411 00:17:38,590 --> 00:17:39,840 - Let's get down on dog level. 412 00:17:39,840 --> 00:17:40,650 That's always better. 413 00:17:40,650 --> 00:17:41,692 - All right, [INAUDIBLE]. 414 00:17:41,692 --> 00:17:45,532 - OK, Chaser, find Goose. 415 00:17:45,532 --> 00:17:48,969 [STUFFED TOY SQUEALING] 416 00:17:48,969 --> 00:17:49,680 - OK. 417 00:17:49,680 --> 00:17:50,580 - Can I do this one? 418 00:17:50,580 --> 00:17:51,497 - You can do this one. 419 00:17:51,497 --> 00:17:54,430 - Chaser, Chaser, find ABC. 420 00:18:01,480 --> 00:18:04,240 ABC-- you did it! 421 00:18:04,240 --> 00:18:04,960 We thank you. 422 00:18:04,960 --> 00:18:08,080 And we want everyone to know that it's a truly remarkable 423 00:18:08,080 --> 00:18:09,190 NOVA tonight. 424 00:18:09,190 --> 00:18:13,090 Four wheels reporting tonight on NOVA Science Now on PBS. 425 00:18:13,090 --> 00:18:18,318 And to you and your brilliant dogs at home, goodnight. 426 00:18:18,318 --> 00:18:18,901 [END PLAYBACK] 427 00:18:18,901 --> 00:18:20,380 NANCY KANWISHER: OK. 428 00:18:20,380 --> 00:18:22,540 She's a very good girl. 429 00:18:22,540 --> 00:18:29,290 And she knows a lot of nouns, right, 1,000 nouns, apparently. 430 00:18:29,290 --> 00:18:33,107 But what can't she do that you guys can do? 431 00:18:33,107 --> 00:18:33,815 Is this language? 432 00:18:37,890 --> 00:18:38,475 Yeah? 433 00:18:38,475 --> 00:18:39,990 AUDIENCE: It's word identification. 434 00:18:39,990 --> 00:18:41,125 It's not language. 435 00:18:41,125 --> 00:18:43,120 You modify actions [INAUDIBLE] language 436 00:18:43,120 --> 00:18:47,020 to be able to put verbs and nouns together. 437 00:18:47,020 --> 00:18:49,720 NANCY KANWISHER: That's good-- verbs and nouns together. 438 00:18:49,720 --> 00:18:51,250 What else? 439 00:18:51,250 --> 00:18:52,000 Yeah, [INAUDIBLE]? 440 00:18:52,000 --> 00:18:53,710 AUDIENCE: It's fortification of things. 441 00:18:53,710 --> 00:18:56,170 If they were like a bigger ABC and a smaller ABC 442 00:18:56,170 --> 00:18:59,886 type of thing, that distinction wouldn't be possible. 443 00:18:59,886 --> 00:19:01,990 NANCY KANWISHER: Alex the Parrot can do that one. 444 00:19:01,990 --> 00:19:03,160 I don't have the video of Alex, and I 445 00:19:03,160 --> 00:19:04,743 don't want to get too hung up on this, 446 00:19:04,743 --> 00:19:07,580 but some animals can do that kind of stuff. 447 00:19:07,580 --> 00:19:08,080 What else? 448 00:19:11,590 --> 00:19:12,090 Yeah? 449 00:19:12,090 --> 00:19:13,140 AUDIENCE: Yeah, it's probably closer 450 00:19:13,140 --> 00:19:14,850 to like sound identification. like, 451 00:19:14,850 --> 00:19:18,280 how I can identify the sound of a train or the sound of a car. 452 00:19:18,280 --> 00:19:20,280 NANCY KANWISHER: So just some rudimentary thing, 453 00:19:20,280 --> 00:19:22,950 like, visual form and sound. 454 00:19:22,950 --> 00:19:25,551 How about when she found Darwin? 455 00:19:25,551 --> 00:19:26,732 AUDIENCE: [INAUDIBLE]. 456 00:19:26,732 --> 00:19:27,690 NANCY KANWISHER: Sorry? 457 00:19:27,690 --> 00:19:30,107 AUDIENCE: Wasn't that case just, like you said, deduction? 458 00:19:30,107 --> 00:19:32,145 It was just like, it wasn't any of the words. 459 00:19:32,145 --> 00:19:35,010 NANCY KANWISHER: That's right, that's right. 460 00:19:35,010 --> 00:19:37,320 But that's pretty impressive, isn't it? 461 00:19:37,320 --> 00:19:39,690 Turns out, kids use that rule too in learning language. 462 00:19:39,690 --> 00:19:42,390 It's a whole set of studies of how 463 00:19:42,390 --> 00:19:44,940 kids use rules to try to figure out what people are referring 464 00:19:44,940 --> 00:19:46,232 to when they learn novel words. 465 00:19:46,232 --> 00:19:48,510 And that's one of the things that kids use. 466 00:19:48,510 --> 00:19:50,430 If there's a thing here that I don't know 467 00:19:50,430 --> 00:19:52,470 and somebody's saying a sound here I don't know, 468 00:19:52,470 --> 00:19:54,480 that thing probably goes with the sound. 469 00:19:54,480 --> 00:19:55,125 Yeah? 470 00:19:55,125 --> 00:19:57,570 AUDIENCE: I was about to say, I took 9.85 last semester. 471 00:19:57,570 --> 00:19:59,250 We talked about like an exact experiment 472 00:19:59,250 --> 00:20:02,730 where kids were able to learn the words of toys 473 00:20:02,730 --> 00:20:04,530 that were like not English words, 474 00:20:04,530 --> 00:20:05,710 but like "dax" and stuff. 475 00:20:05,710 --> 00:20:07,912 But then when they were given like a new object, 476 00:20:07,912 --> 00:20:09,870 they would be able to identify it as different. 477 00:20:09,870 --> 00:20:10,890 NANCY KANWISHER: Exactly. 478 00:20:10,890 --> 00:20:12,182 It's called mutual exclusivity. 479 00:20:12,182 --> 00:20:14,790 And that's exactly what Chaser is showing here. 480 00:20:14,790 --> 00:20:17,730 OK, so pretty impressive, but not fully 481 00:20:17,730 --> 00:20:21,180 language, more like memorizing a bunch of nouns 482 00:20:21,180 --> 00:20:26,400 plus mutual exclusivity plus some other stuff, maybe. 483 00:20:26,400 --> 00:20:31,770 She certainly can't understand who did what to who and why. 484 00:20:31,770 --> 00:20:33,510 This is not even in the ballpark. 485 00:20:33,510 --> 00:20:37,350 This is the essence of what we talk to each other about is 486 00:20:37,350 --> 00:20:40,710 this kind of stuff, all kinds of complicated relationships 487 00:20:40,710 --> 00:20:46,200 between different concepts that we communicate in language. 488 00:20:46,200 --> 00:20:49,980 So animals in-- not just taught English, 489 00:20:49,980 --> 00:20:51,900 but animals in their natural environments 490 00:20:51,900 --> 00:20:55,260 communicate in rich and detailed ways with each other. 491 00:20:55,260 --> 00:20:59,860 But usually in each case, about a very restricted domain. 492 00:20:59,860 --> 00:21:02,760 What kind of danger is around? 493 00:21:02,760 --> 00:21:04,530 What kind of food source is around? 494 00:21:04,530 --> 00:21:06,930 Those basic kinds of narrow things 495 00:21:06,930 --> 00:21:08,850 that are of survival value, those 496 00:21:08,850 --> 00:21:11,970 are the things that animal communication systems usually 497 00:21:11,970 --> 00:21:12,840 deal with. 498 00:21:12,840 --> 00:21:15,360 And in contrast, human languages are 499 00:21:15,360 --> 00:21:17,970 open-ended and compositional. 500 00:21:17,970 --> 00:21:20,250 Compositional means that we combine 501 00:21:20,250 --> 00:21:23,730 words to say new things, things no human being 502 00:21:23,730 --> 00:21:26,040 has ever said before. 503 00:21:26,040 --> 00:21:29,700 So that you don't see in animals. 504 00:21:29,700 --> 00:21:31,710 So what is language cognitively? 505 00:21:31,710 --> 00:21:36,070 That is, what do you have to know to know a language? 506 00:21:36,070 --> 00:21:37,810 Bunch of basic things. 507 00:21:37,810 --> 00:21:40,090 One is phonology, the sounds of language. 508 00:21:40,090 --> 00:21:42,900 We've talked about this a bit in the case of speech perception. 509 00:21:42,900 --> 00:21:46,380 Just hearing the difference between a ba and a pa, 510 00:21:46,380 --> 00:21:48,570 or seeing the equivalent gesture. 511 00:21:48,570 --> 00:21:51,540 American Sign Language is a fully expressive 512 00:21:51,540 --> 00:21:52,710 natural language. 513 00:21:52,710 --> 00:21:55,710 And there the phonemes are different pieces 514 00:21:55,710 --> 00:21:58,350 of hand movements rather than sounds, but function 515 00:21:58,350 --> 00:22:00,810 as phonemes all the same. 516 00:22:00,810 --> 00:22:03,510 And we talked about a region of the brain that 517 00:22:03,510 --> 00:22:07,830 responds very specifically to speech sounds in humans. 518 00:22:07,830 --> 00:22:09,810 Moving up into the language system, 519 00:22:09,810 --> 00:22:12,595 that's just an input system-- 520 00:22:12,595 --> 00:22:14,970 and by the way, we also talked about the visual word form 521 00:22:14,970 --> 00:22:18,812 area, a very recent addition to the input system in language. 522 00:22:18,812 --> 00:22:20,520 But that's only a few thousand years old. 523 00:22:20,520 --> 00:22:23,640 It's really phonology that's the native form of language that's 524 00:22:23,640 --> 00:22:27,180 been around for tens, if not hundreds, of thousands 525 00:22:27,180 --> 00:22:29,290 of years in human evolution. 526 00:22:29,290 --> 00:22:32,010 So semantics, we need to know what words mean. 527 00:22:32,010 --> 00:22:33,240 That's lexical semantics. 528 00:22:33,240 --> 00:22:34,740 But we also need to know how meaning 529 00:22:34,740 --> 00:22:36,690 arises when words go together. 530 00:22:39,740 --> 00:22:41,510 And related to how words go together, 531 00:22:41,510 --> 00:22:44,150 we need to know about the syntax of a language. 532 00:22:44,150 --> 00:22:48,170 That is, the structure or grammar of a language. 533 00:22:48,170 --> 00:22:50,600 And so each language has a set of rules 534 00:22:50,600 --> 00:22:54,170 about how you string words together in that language. 535 00:22:54,170 --> 00:22:57,050 And usually central to that-- it's not the only thing, 536 00:22:57,050 --> 00:22:59,360 but a central part of that-- is word order. 537 00:22:59,360 --> 00:23:02,270 And that whole set of rules for how you string together 538 00:23:02,270 --> 00:23:04,730 words, following word order rules, 539 00:23:04,730 --> 00:23:08,070 determines the meaning of the string of words. 540 00:23:08,070 --> 00:23:13,610 For example, shark bites man is different than man bites shark. 541 00:23:13,610 --> 00:23:15,620 And that just comes out of the syntax 542 00:23:15,620 --> 00:23:19,310 that we know that in English in this kind of construction 543 00:23:19,310 --> 00:23:22,790 the first word is going to be the agent, the one who's 544 00:23:22,790 --> 00:23:23,530 doing the thing. 545 00:23:23,530 --> 00:23:26,030 And the third word is going to be the patient, the one who's 546 00:23:26,030 --> 00:23:27,710 receiving the doing. 547 00:23:27,710 --> 00:23:30,200 And that's just built into your language system, 548 00:23:30,200 --> 00:23:33,990 that you know that implicitly. 549 00:23:33,990 --> 00:23:36,300 There's also the pragmatics of language. 550 00:23:36,300 --> 00:23:39,690 That is, how we understand what somebody actually 551 00:23:39,690 --> 00:23:41,640 means when they say something to us, 552 00:23:41,640 --> 00:23:44,040 which isn't always just a function 553 00:23:44,040 --> 00:23:47,340 of the actual string of words coming out of their mouth. 554 00:23:47,340 --> 00:23:51,870 So if somebody says it will be awesome if you pass the salt, 555 00:23:51,870 --> 00:23:54,120 it's not all that awesome to have the salt. 556 00:23:54,120 --> 00:23:56,760 It really means, please pass the salt. 557 00:23:56,760 --> 00:23:59,550 The pragmatics of the situation tells you the actual intent. 558 00:24:02,210 --> 00:24:04,730 And so to do pragmatics involves thinking 559 00:24:04,730 --> 00:24:07,577 about the other person's intent, what are they thinking, what 560 00:24:07,577 --> 00:24:09,410 do they want, what's going on in their head, 561 00:24:09,410 --> 00:24:11,300 and using all that background knowledge 562 00:24:11,300 --> 00:24:16,690 to constrain what do they mean by this particular utterance. 563 00:24:16,690 --> 00:24:19,720 So let's just survey of the main pieces 564 00:24:19,720 --> 00:24:21,490 of what we mean by language. 565 00:24:21,490 --> 00:24:23,350 But for the next two lectures, we're 566 00:24:23,350 --> 00:24:24,820 going to focus on the core, which 567 00:24:24,820 --> 00:24:27,820 is syntax and semantics, this stuff in here. 568 00:24:27,820 --> 00:24:29,950 And I will sloppily use the word "language" 569 00:24:29,950 --> 00:24:34,420 to refer to this stuff, not all the other stuff. 570 00:24:34,420 --> 00:24:37,787 And we'll focus really on the sentence understanding. 571 00:24:37,787 --> 00:24:40,120 So what do we want to know about sentence understanding? 572 00:24:40,120 --> 00:24:43,540 Well, the first thing we want to know is, is it even a thing. 573 00:24:43,540 --> 00:24:46,855 Is language a thing separate from the rest of thought? 574 00:24:49,700 --> 00:24:52,295 Second thing we want to know is, if it is at least something 575 00:24:52,295 --> 00:24:55,760 of kind of a thing, does language itself 576 00:24:55,760 --> 00:24:58,590 have component structure within it? 577 00:24:58,590 --> 00:25:00,590 Are there different parts of the language system 578 00:25:00,590 --> 00:25:04,570 that maybe do different things? 579 00:25:04,570 --> 00:25:06,700 And if so, what is represented and computed 580 00:25:06,700 --> 00:25:08,990 in each of those parts? 581 00:25:08,990 --> 00:25:13,798 And third, how do we represent meaning in the brain? 582 00:25:13,798 --> 00:25:16,090 So these are the things we'll address over the next two 583 00:25:16,090 --> 00:25:16,900 lectures. 584 00:25:16,900 --> 00:25:18,370 And let's start with this question 585 00:25:18,370 --> 00:25:21,190 that'll probably take up the bulk of this lecture. 586 00:25:21,190 --> 00:25:24,820 Is language distinct from the rest of thought? 587 00:25:24,820 --> 00:25:27,190 Another way of putting this, a more familiar way, 588 00:25:27,190 --> 00:25:30,100 is to ask, what is the relationship between language 589 00:25:30,100 --> 00:25:31,060 and thought? 590 00:25:31,060 --> 00:25:35,895 Or even more pointedly, could you think without language? 591 00:25:35,895 --> 00:25:37,520 Probably, every one of you has wondered 592 00:25:37,520 --> 00:25:38,940 about that at some point. 593 00:25:38,940 --> 00:25:40,910 So take like two or three minutes, talk 594 00:25:40,910 --> 00:25:43,485 to your neighbors about this, see if you can figure out 595 00:25:43,485 --> 00:25:45,110 whether you can think without language, 596 00:25:45,110 --> 00:25:47,690 and then let's pool your insights. 597 00:25:47,690 --> 00:25:48,830 Talk, think. 598 00:25:50,778 --> 00:25:51,752 [SIDE CONVERSATIONS] 599 00:25:51,752 --> 00:25:53,920 NANCY KANWISHER: OK, if you guys all nailed it, 600 00:25:53,920 --> 00:25:56,530 I'm sure you solve the whole thing. 601 00:25:56,530 --> 00:25:59,300 People have been talking about this for probably millennia. 602 00:25:59,300 --> 00:26:03,380 So, what do you guys think? 603 00:26:03,380 --> 00:26:05,725 What were some of your reflections on this question? 604 00:26:08,910 --> 00:26:09,660 Come on, you guys. 605 00:26:09,660 --> 00:26:11,430 Yes, Carly? 606 00:26:11,430 --> 00:26:13,530 AUDIENCE: I said I think that they 607 00:26:13,530 --> 00:26:17,010 could think without language because of like we talked 608 00:26:17,010 --> 00:26:19,560 previously about how [INAUDIBLE] babies 609 00:26:19,560 --> 00:26:21,640 are given very complex thought. 610 00:26:21,640 --> 00:26:26,880 But, like, he was arguing that the whale research, 611 00:26:26,880 --> 00:26:29,170 there's also the thing that babies kind of form 612 00:26:29,170 --> 00:26:31,060 their own language that we don't understand, 613 00:26:31,060 --> 00:26:32,703 but I don't think [INAUDIBLE]. 614 00:26:32,703 --> 00:26:33,870 NANCY KANWISHER: Not really. 615 00:26:33,870 --> 00:26:35,880 If you take three-month-old babies-- not really. 616 00:26:35,880 --> 00:26:39,120 So perfect, absolutely, you can hear this. 617 00:26:39,120 --> 00:26:41,010 Babies can think. 618 00:26:41,010 --> 00:26:42,592 You take 9.85, you'll learn more. 619 00:26:42,592 --> 00:26:44,550 They can really think about all kinds of stuff. 620 00:26:44,550 --> 00:26:47,640 It's really amazing how much they understand. 621 00:26:47,640 --> 00:26:51,160 And at three to six months, there's little or no language. 622 00:26:51,160 --> 00:26:53,580 So there's a beautiful case of thinking without language. 623 00:26:53,580 --> 00:26:54,300 Yeah, David? 624 00:26:54,300 --> 00:26:57,235 AUDIENCE: On the other side, if you 625 00:26:57,235 --> 00:26:59,610 don't give a name to something, if you don't give it word 626 00:26:59,610 --> 00:27:01,820 to something, then it's hard to really know it. 627 00:27:01,820 --> 00:27:04,950 Like, maybe there are 20 different types 628 00:27:04,950 --> 00:27:06,700 of the color green. 629 00:27:06,700 --> 00:27:09,110 And if you don't decide to call one of them 630 00:27:09,110 --> 00:27:11,630 olive and another one khaki green 631 00:27:11,630 --> 00:27:13,302 or something like that, then-- 632 00:27:13,302 --> 00:27:15,427 NANCY KANWISHER: Then you can't see the difference? 633 00:27:15,427 --> 00:27:17,580 AUDIENCE: Well, well, I don't know if you'd ever 634 00:27:17,580 --> 00:27:19,140 think of the difference. 635 00:27:19,140 --> 00:27:21,227 NANCY KANWISHER: OK, let's think about this. 636 00:27:21,227 --> 00:27:22,810 Do you think could see the difference? 637 00:27:22,810 --> 00:27:27,210 Suppose I held up an olive patch and a khaki patch to you. 638 00:27:27,210 --> 00:27:29,940 And for whatever reason, you had been raised with deprivation 639 00:27:29,940 --> 00:27:31,350 of the words olive and khaki. 640 00:27:31,350 --> 00:27:33,600 AUDIENCE: But somehow it's not about just a perception 641 00:27:33,600 --> 00:27:34,140 question. 642 00:27:34,140 --> 00:27:35,460 It's about remembering. 643 00:27:35,460 --> 00:27:37,437 NANCY KANWISHER: Yeah, bingo, bingo. 644 00:27:37,437 --> 00:27:39,270 So that's roughly what the literature shows. 645 00:27:39,270 --> 00:27:40,228 Anya, help me out here. 646 00:27:40,228 --> 00:27:41,400 I forgot to look this up. 647 00:27:41,400 --> 00:27:43,980 The literature still show that perceptually you can 648 00:27:43,980 --> 00:27:45,147 discriminate them just fine. 649 00:27:45,147 --> 00:27:46,813 It doesn't make a damn bit of difference 650 00:27:46,813 --> 00:27:48,190 if you have words for it. 651 00:27:48,190 --> 00:27:49,630 But if you have to remember it-- 652 00:27:49,630 --> 00:27:50,130 sorry. 653 00:27:50,130 --> 00:27:51,010 AUDIENCE: Faster. 654 00:27:51,010 --> 00:27:53,670 NANCY KANWISHER: Faster, faster. 655 00:27:53,670 --> 00:27:55,830 But accuracy in D prime, I don't think 656 00:27:55,830 --> 00:27:59,970 is different, maybe a little bit. 657 00:27:59,970 --> 00:28:00,580 Oops, caught. 658 00:28:00,580 --> 00:28:01,580 I meant to look this up. 659 00:28:01,580 --> 00:28:03,630 I knew this is going to come up. 660 00:28:03,630 --> 00:28:05,820 Write me an email to look this up and help 661 00:28:05,820 --> 00:28:07,013 me find the relevant stuff. 662 00:28:07,013 --> 00:28:09,180 Anyway, doesn't make a huge difference perceptually, 663 00:28:09,180 --> 00:28:11,710 but it does if you have to remember it for later. 664 00:28:11,710 --> 00:28:12,210 Yeah? 665 00:28:12,210 --> 00:28:13,890 AUDIENCE: That's actually what I say because I'm actually 666 00:28:13,890 --> 00:28:16,020 reproducing the experiment that found 667 00:28:16,020 --> 00:28:18,410 that there was a difference in color [INAUDIBLE].. 668 00:28:18,410 --> 00:28:19,680 NANCY KANWISHER: Aha, aha. 669 00:28:19,680 --> 00:28:20,180 What? 670 00:28:20,180 --> 00:28:21,495 In perception or memory? 671 00:28:21,495 --> 00:28:23,640 AUDIENCE: So they found that-- 672 00:28:23,640 --> 00:28:25,275 I believe it was-- 673 00:28:25,275 --> 00:28:27,900 NANCY KANWISHER: Because there's been a long history with this. 674 00:28:27,900 --> 00:28:29,970 They find one thing and they-- that's partly why I'm-- 675 00:28:29,970 --> 00:28:31,900 AUDIENCE: It's like a difference in the reaction time. 676 00:28:31,900 --> 00:28:34,230 Interesting enough, they found that if they introduce 677 00:28:34,230 --> 00:28:36,638 interference in their linguistic system, 678 00:28:36,638 --> 00:28:37,930 then that difference went away. 679 00:28:37,930 --> 00:28:39,480 So that's evidence that the language 680 00:28:39,480 --> 00:28:40,742 is causing the difference. 681 00:28:40,742 --> 00:28:43,200 NANCY KANWISHER: And that's in a perceptual discrimination. 682 00:28:43,200 --> 00:28:43,510 OK. 683 00:28:43,510 --> 00:28:44,677 AUDIENCE: It's pretty small. 684 00:28:44,677 --> 00:28:47,910 NANCY KANWISHER: Yeah, yeah, well, behavioral, well, yeah, 685 00:28:47,910 --> 00:28:49,350 effects often are. 686 00:28:49,350 --> 00:28:50,070 Yeah, Isabel? 687 00:28:50,070 --> 00:28:52,900 AUDIENCE: I remember one of the first neuroscience talks I 688 00:28:52,900 --> 00:28:57,680 went to in college was a woman who had been [INAUDIBLE].. 689 00:28:57,680 --> 00:28:59,590 She got in a terrible stroke and she's 690 00:28:59,590 --> 00:29:03,630 suffering from aphasia [INAUDIBLE] the speaking part 691 00:29:03,630 --> 00:29:05,320 and forgot all the language she learned. 692 00:29:05,320 --> 00:29:09,108 It took over her over a year to regain [INAUDIBLE].. 693 00:29:09,108 --> 00:29:12,412 And I remember the question that I asked was, 694 00:29:12,412 --> 00:29:16,832 you have this really terrible pain [INAUDIBLE].. 695 00:29:16,832 --> 00:29:18,540 But what did your inner voice sound like? 696 00:29:18,540 --> 00:29:21,840 And she said, well, I don't really have one, [INAUDIBLE].. 697 00:29:21,840 --> 00:29:23,340 And then she said, well, I must have 698 00:29:23,340 --> 00:29:25,350 thought in images and feelings. 699 00:29:25,350 --> 00:29:28,740 And the interesting thing that I experienced 700 00:29:28,740 --> 00:29:32,130 when I was relearning to talk was that, the more English 701 00:29:32,130 --> 00:29:34,890 I learned, the more my thoughts was with grammar. 702 00:29:34,890 --> 00:29:36,870 So I still could have these thoughts, 703 00:29:36,870 --> 00:29:40,340 but they were formulated in a different way 704 00:29:40,340 --> 00:29:43,230 than they were when I had [INAUDIBLE] 705 00:29:43,230 --> 00:29:45,023 the structured language department. 706 00:29:45,023 --> 00:29:46,440 NANCY KANWISHER: OK, that's great. 707 00:29:46,440 --> 00:29:50,550 So we're going to learn more about all of that, absolutely. 708 00:29:50,550 --> 00:29:52,590 OK, very good. 709 00:29:52,590 --> 00:29:54,750 So cool question, not obvious. 710 00:29:54,750 --> 00:29:56,680 Let's see what the data say. 711 00:29:56,680 --> 00:29:59,522 So first of all, you guys talked about babies 712 00:29:59,522 --> 00:30:00,480 and how they can think. 713 00:30:00,480 --> 00:30:04,600 But animals can think too, maybe not fully as richly as we can, 714 00:30:04,600 --> 00:30:07,350 but they can think in all kinds of subtle, rich ways. 715 00:30:07,350 --> 00:30:08,730 And animals don't have language. 716 00:30:08,730 --> 00:30:10,890 And so that's another case, animals and infants. 717 00:30:10,890 --> 00:30:12,892 And I'm mentioning numerosity because these 718 00:30:12,892 --> 00:30:14,850 are things we happen to have mentioned in here. 719 00:30:14,850 --> 00:30:17,130 Remember, the approximate number system. 720 00:30:17,130 --> 00:30:18,420 Animals are great at that. 721 00:30:18,420 --> 00:30:20,670 Very young infants are greater that when they 722 00:30:20,670 --> 00:30:22,860 don't have language at all. 723 00:30:22,860 --> 00:30:25,620 Also, by the way, people whose language 724 00:30:25,620 --> 00:30:29,370 do not have any number words whatsoever 725 00:30:29,370 --> 00:30:31,740 can do approximate numerosity. 726 00:30:31,740 --> 00:30:36,220 So here's a cool study from Ted Gibson's lab a few years ago. 727 00:30:36,220 --> 00:30:39,930 They went down into remote parts of the Amazon 728 00:30:39,930 --> 00:30:42,390 to study this group of people, the Piraha. 729 00:30:42,390 --> 00:30:44,760 Here they are in their canoe. 730 00:30:44,760 --> 00:30:49,790 They are a hunter-gatherer tribe of just a few hundred people. 731 00:30:49,790 --> 00:30:51,540 Their language is, as far as linguists can 732 00:30:51,540 --> 00:30:53,530 tell, unrelated to anyone else. 733 00:30:53,530 --> 00:30:55,170 And it has no number words. 734 00:30:55,170 --> 00:30:57,120 There's a whole dispute about that, 735 00:30:57,120 --> 00:30:58,890 but the current view is there are really 736 00:30:58,890 --> 00:31:02,880 no number words at all, not even for zero or one. 737 00:31:02,880 --> 00:31:06,600 So how do they do at approximate magnitude? 738 00:31:06,600 --> 00:31:07,980 Well, let's see. 739 00:31:07,980 --> 00:31:12,450 So here is the testing session down in the Amazon. 740 00:31:12,450 --> 00:31:16,500 And this is the experimenter lining up a bunch of, I think, 741 00:31:16,500 --> 00:31:17,650 they're batteries. 742 00:31:17,650 --> 00:31:19,920 And this guy is asked to match the number of balloons 743 00:31:19,920 --> 00:31:21,045 to the number of batteries. 744 00:31:21,045 --> 00:31:23,670 And he has to do it aligned this way so he can't just 745 00:31:23,670 --> 00:31:25,290 put them one next to the other. 746 00:31:25,290 --> 00:31:28,290 If you let him, he'll put them one next to the other. 747 00:31:28,290 --> 00:31:32,220 But this is designed to test it better. 748 00:31:32,220 --> 00:31:33,690 And he puts down four balloons. 749 00:31:33,690 --> 00:31:34,065 [VIDEO PLAYBACK] 750 00:31:34,065 --> 00:31:34,898 [SIDE CONVERSATIONS] 751 00:31:34,898 --> 00:31:36,690 Bingo, very good. 752 00:31:36,690 --> 00:31:41,730 OK, no number words in his language. 753 00:31:41,730 --> 00:31:47,666 What about this case? 754 00:31:47,666 --> 00:31:48,907 - Hi, people. 755 00:31:48,907 --> 00:31:50,740 NANCY KANWISHER: Oh, the plot is thickening. 756 00:31:50,740 --> 00:31:55,852 - Six or five, [INAUDIBLE] 757 00:31:55,852 --> 00:31:58,310 - [INAUDIBLE] lot of thread. 758 00:31:58,310 --> 00:31:59,350 [INAUDIBLE] of thread. 759 00:31:59,350 --> 00:32:02,390 NANCY KANWISHER: He laughs, he thinks that's pretty funny. 760 00:32:02,390 --> 00:32:03,452 But watch. 761 00:32:03,452 --> 00:32:05,745 - Five, five. 762 00:32:05,745 --> 00:32:07,550 NANCY KANWISHER: [? Valiant ?] goes ahead. 763 00:32:07,550 --> 00:32:11,302 - [INAUDIBLE] I think it is [INAUDIBLE] five. 764 00:32:11,302 --> 00:32:14,040 Lots, lots. 765 00:32:14,040 --> 00:32:19,810 [INAUDIBLE] and intensifier, like lots and lots. 766 00:32:19,810 --> 00:32:21,200 - You're doing well. 767 00:32:33,325 --> 00:32:35,105 NANCY KANWISHER: Right, right, right. 768 00:32:35,105 --> 00:32:35,327 - There you go. 769 00:32:35,327 --> 00:32:35,820 - Good. 770 00:32:35,820 --> 00:32:36,820 - You can see which one. 771 00:32:36,820 --> 00:32:37,820 - Nine-- nine, 10. 772 00:32:37,820 --> 00:32:38,320 - 10? 773 00:32:38,320 --> 00:32:38,820 That was 10? 774 00:32:38,820 --> 00:32:39,745 [END PLAYBACK] 775 00:32:39,745 --> 00:32:42,210 NANCY KANWISHER: So I think he gave nine for 10, 776 00:32:42,210 --> 00:32:45,460 or something like that. 777 00:32:45,460 --> 00:32:47,615 Anyway, if I had any of you guys do this task 778 00:32:47,615 --> 00:32:48,990 and I prevented you from counting 779 00:32:48,990 --> 00:32:51,210 by having you do verbal shadowing or something else 780 00:32:51,210 --> 00:32:53,010 to tie up your language system, you 781 00:32:53,010 --> 00:32:55,980 would do exactly the same as this guy does. 782 00:32:55,980 --> 00:32:59,580 So the approximate number system doesn't require language, 783 00:32:59,580 --> 00:33:01,830 doesn't require number words in your language 784 00:33:01,830 --> 00:33:03,120 to get the concept. 785 00:33:03,120 --> 00:33:06,756 And it doesn't require use of language to do the task. 786 00:33:06,756 --> 00:33:08,842 AUDIENCE: He saw him put [INAUDIBLE]?? 787 00:33:08,842 --> 00:33:09,800 NANCY KANWISHER: Sorry? 788 00:33:09,800 --> 00:33:11,560 AUDIENCE: He actually saw him put all of them? 789 00:33:11,560 --> 00:33:11,960 He saw? 790 00:33:11,960 --> 00:33:13,543 NANCY KANWISHER: Yeah, just like you-- 791 00:33:13,543 --> 00:33:14,680 AUDIENCE: [INAUDIBLE]. 792 00:33:14,680 --> 00:33:16,270 NANCY KANWISHER: I mean, that's the actual experiment being 793 00:33:16,270 --> 00:33:17,187 conducted right there. 794 00:33:19,970 --> 00:33:23,330 OK, so we've just argued that at least the approximate number 795 00:33:23,330 --> 00:33:27,200 system is present in animals who don't have number words, 796 00:33:27,200 --> 00:33:32,340 infants who don't, and adults who don't have number words. 797 00:33:32,340 --> 00:33:34,730 What about other aspects of thought? 798 00:33:34,730 --> 00:33:36,965 And what can we learn from studying brain disorders, 799 00:33:36,965 --> 00:33:41,332 as Isabel mentioned a moment ago, a very rich source. 800 00:33:41,332 --> 00:33:43,040 So here's the question we're considering. 801 00:33:43,040 --> 00:33:45,890 We're taking language and thought, or cognition, 802 00:33:45,890 --> 00:33:48,710 and we're asking whether they're totally separate in the mind 803 00:33:48,710 --> 00:33:52,910 and brain or whether they're totally the same thing 804 00:33:52,910 --> 00:33:55,790 or whether there's some relationship that they're 805 00:33:55,790 --> 00:33:57,320 somewhat different. 806 00:33:57,320 --> 00:33:58,890 So that's the question. 807 00:33:58,890 --> 00:34:00,500 What do we learn from brain disorders? 808 00:34:00,500 --> 00:34:03,720 Well, let's start with developmental disorders. 809 00:34:03,720 --> 00:34:06,800 And there are unfortunately a large number of these. 810 00:34:06,800 --> 00:34:09,350 For example, there are language savants, 811 00:34:09,350 --> 00:34:12,199 people with Down syndrome, Williams syndrome, 812 00:34:12,199 --> 00:34:13,440 Turner syndrome. 813 00:34:13,440 --> 00:34:15,949 These are all developmental disorders 814 00:34:15,949 --> 00:34:20,330 in which people have very low IQs, but, 815 00:34:20,330 --> 00:34:23,780 notably, in each of these cases, very good language. 816 00:34:23,780 --> 00:34:26,690 Perhaps the most striking is Williams syndrome. 817 00:34:26,690 --> 00:34:28,620 These kids are remarkable. 818 00:34:28,620 --> 00:34:30,230 They have very low IQs. 819 00:34:30,230 --> 00:34:33,739 They can't do the most basic spatial reasoning tasks. 820 00:34:33,739 --> 00:34:35,570 They can't cross the street safely. 821 00:34:35,570 --> 00:34:39,230 They can't live independently at all. 822 00:34:39,230 --> 00:34:41,060 And yet they're highly social. 823 00:34:41,060 --> 00:34:45,350 And their language is almost indistinguishable from any 824 00:34:45,350 --> 00:34:46,730 of yours. 825 00:34:46,730 --> 00:34:48,290 Not quite-- if you test them subtly, 826 00:34:48,290 --> 00:34:51,530 can find some differences, but it is rich and complex. 827 00:34:51,530 --> 00:34:52,550 And it's bizarre. 828 00:34:52,550 --> 00:34:54,650 Because you'd think if your thoughts are so 829 00:34:54,650 --> 00:34:56,690 impoverished because your IQ is low, 830 00:34:56,690 --> 00:34:58,255 how could you have rich language. 831 00:34:58,255 --> 00:35:00,380 But that's the weird thing about Williams syndrome. 832 00:35:00,380 --> 00:35:02,810 Their language is extremely rich and, in fact, 833 00:35:02,810 --> 00:35:05,580 poetic and quite beautiful and expressive. 834 00:35:05,580 --> 00:35:08,600 So that's really surprising and suggests 835 00:35:08,600 --> 00:35:13,310 that you can have quite severely impaired cognition and very 836 00:35:13,310 --> 00:35:14,610 good language. 837 00:35:14,610 --> 00:35:16,878 So that's the first crack that these things are 838 00:35:16,878 --> 00:35:18,170 more separate than you'd guess. 839 00:35:18,170 --> 00:35:21,530 Actually, I find this one more surprising than all the others. 840 00:35:21,530 --> 00:35:25,280 But in cases of brain damage, which 841 00:35:25,280 --> 00:35:30,330 was the first mental function localized in the brain. 842 00:35:30,330 --> 00:35:31,970 So this is historically important. 843 00:35:31,970 --> 00:35:35,510 Way back in 1861, Paul Broca stood up 844 00:35:35,510 --> 00:35:37,970 in front of the Anthropology Society of Paris 845 00:35:37,970 --> 00:35:40,610 and he announced that the left frontal lobe 846 00:35:40,610 --> 00:35:42,680 was the seat of speech. 847 00:35:42,680 --> 00:35:46,490 And this is on the basis of his patient Tan, who 848 00:35:46,490 --> 00:35:48,380 had a big nasty lesion right there 849 00:35:48,380 --> 00:35:51,650 in what became known as Broca's area. 850 00:35:51,650 --> 00:35:55,190 Tan was his name because, after that lesion, that was all 851 00:35:55,190 --> 00:35:57,410 he could say. 852 00:35:57,410 --> 00:36:01,100 So this is back when the mainstream view 853 00:36:01,100 --> 00:36:03,470 was very much against localization of function 854 00:36:03,470 --> 00:36:04,490 in the brain. 855 00:36:04,490 --> 00:36:06,570 There were people like Franz Josef Gall who 856 00:36:06,570 --> 00:36:08,987 were going around saying that different parts of the brain 857 00:36:08,987 --> 00:36:11,480 did very different things, but Gall was kind of a nut 858 00:36:11,480 --> 00:36:14,660 and he was not taken seriously by the academic elite, 859 00:36:14,660 --> 00:36:19,460 whereas Broca was a fancy member of the French academic 860 00:36:19,460 --> 00:36:21,530 societies and a muckety muck. 861 00:36:21,530 --> 00:36:23,960 And when he announced that the left frontal lobe is 862 00:36:23,960 --> 00:36:26,510 the seat of speech, everybody had to pay attention. 863 00:36:26,510 --> 00:36:27,990 So it was big stuff. 864 00:36:27,990 --> 00:36:32,960 Importantly, Broca noted that Tan wasn't globally impaired 865 00:36:32,960 --> 00:36:36,110 at thinking, that Tan could do all kinds of things, 866 00:36:36,110 --> 00:36:37,800 even though he could not speak. 867 00:36:37,800 --> 00:36:41,960 So he was already onto this critical idea way back in 1861. 868 00:36:41,960 --> 00:36:44,600 And he's just the most famous in that group. 869 00:36:44,600 --> 00:36:46,850 There were a bunch of people before him in the decades 870 00:36:46,850 --> 00:36:51,380 before who were reporting similar kinds of associations. 871 00:36:51,380 --> 00:36:56,030 So what would it be like to have intact thought despite impaired 872 00:36:56,030 --> 00:36:57,320 language? 873 00:36:57,320 --> 00:36:59,915 So Isabel mentioned, asking somebody who had a stroke. 874 00:37:04,080 --> 00:37:04,770 OK, Great. 875 00:37:04,770 --> 00:37:06,910 So here's another case. 876 00:37:06,910 --> 00:37:09,570 This is a case of this guy here, Tom Lubbock, 877 00:37:09,570 --> 00:37:13,620 who died a few years ago from a brain tumor 878 00:37:13,620 --> 00:37:17,460 in his temporal lobe that destroyed most of his language, 879 00:37:17,460 --> 00:37:19,450 but it destroyed it gradually. 880 00:37:19,450 --> 00:37:20,970 And this guy was a writer. 881 00:37:20,970 --> 00:37:25,380 He was an art critic for a major English paper. 882 00:37:25,380 --> 00:37:28,830 And as he started to lose language, he wrote about it, 883 00:37:28,830 --> 00:37:31,400 and he wrote about it very beautifully. 884 00:37:31,400 --> 00:37:35,210 And he said, "my language to describe things in the world 885 00:37:35,210 --> 00:37:38,120 is very small, limited. 886 00:37:38,120 --> 00:37:39,770 My thoughts when I look at the world 887 00:37:39,770 --> 00:37:42,590 are vast, limitless, and normal. 888 00:37:42,590 --> 00:37:44,810 Same as they ever were. 889 00:37:44,810 --> 00:37:46,580 My experience of the world is not 890 00:37:46,580 --> 00:37:52,230 made less by lack of language but is essentially unchanged." 891 00:37:52,230 --> 00:37:55,490 So that's a very powerful and surprising piece of writing. 892 00:37:55,490 --> 00:37:57,568 It's a little bit mysterious, because here's 893 00:37:57,568 --> 00:37:59,360 this guy writing beautifully and telling us 894 00:37:59,360 --> 00:38:00,840 his language is impaired. 895 00:38:00,840 --> 00:38:03,960 So his idea of language impairment may not be mine. 896 00:38:03,960 --> 00:38:06,020 I wish I could write that well. 897 00:38:06,020 --> 00:38:08,690 Nonetheless, he's clearly reflecting 898 00:38:08,690 --> 00:38:13,955 on what is a very big loss of his previous language ability. 899 00:38:13,955 --> 00:38:15,830 And I'm sure it was very painstaking to write 900 00:38:15,830 --> 00:38:16,980 these sentences. 901 00:38:16,980 --> 00:38:18,980 And he's still telling us that, even though he's 902 00:38:18,980 --> 00:38:22,770 lost a lot of language, it has not changed his experience. 903 00:38:22,770 --> 00:38:25,190 So that's just one subjective impression. 904 00:38:25,190 --> 00:38:27,530 So that argues against this extreme view 905 00:38:27,530 --> 00:38:32,010 that they're the same thing, but it leaves a lot of slop. 906 00:38:32,010 --> 00:38:32,510 Yes? 907 00:38:32,510 --> 00:38:35,720 AUDIENCE: [INAUDIBLE] because he had a [INAUDIBLE] 908 00:38:35,720 --> 00:38:39,560 of speaking and learning about the word before. 909 00:38:39,560 --> 00:38:40,660 NANCY KANWISHER: Yes. 910 00:38:40,660 --> 00:38:41,940 A very important point. 911 00:38:41,940 --> 00:38:43,040 Absolutely. 912 00:38:43,040 --> 00:38:44,570 So this is a case of somebody who 913 00:38:44,570 --> 00:38:48,950 had a lesion in mid-life 40, 50, something like that. 914 00:38:48,950 --> 00:38:53,420 He had a whole lifetime of using language to learn and bootstrap 915 00:38:53,420 --> 00:38:54,420 all of cognition. 916 00:38:54,420 --> 00:38:55,850 So absolutely we have to separate 917 00:38:55,850 --> 00:38:57,020 two different questions. 918 00:38:57,020 --> 00:39:02,090 Do you need language to become a normal, intelligent, functional 919 00:39:02,090 --> 00:39:04,467 human being? 920 00:39:04,467 --> 00:39:06,050 Do you need it throughout development? 921 00:39:06,050 --> 00:39:10,280 Or, once you've developed, do you still need it to think? 922 00:39:10,280 --> 00:39:12,260 And those are two very different questions. 923 00:39:12,260 --> 00:39:15,470 And, in fact, absolutely you need language to develop. 924 00:39:15,470 --> 00:39:19,430 If you reflect for a moment on all the things you know, 925 00:39:19,430 --> 00:39:23,360 take a quick mental inventory, survey all the things you know, 926 00:39:23,360 --> 00:39:27,830 it's a lot of things, almost all of those you learned 927 00:39:27,830 --> 00:39:29,270 because somebody told you. 928 00:39:31,900 --> 00:39:34,930 Most of what we know we learn from language. 929 00:39:34,930 --> 00:39:36,340 Maybe you read about it. 930 00:39:36,340 --> 00:39:40,620 But that's somebody telling you in a different way. 931 00:39:40,620 --> 00:39:43,980 So language is crucial for development of cognition 932 00:39:43,980 --> 00:39:44,760 and for learning. 933 00:39:44,760 --> 00:39:45,990 Absolutely. 934 00:39:45,990 --> 00:39:48,120 But now we're asking a different question 935 00:39:48,120 --> 00:39:50,160 of whether you need it, whether it's 936 00:39:50,160 --> 00:39:53,920 the same thing in adulthood. 937 00:39:53,920 --> 00:39:57,040 So this guy is a little bit complicated, 938 00:39:57,040 --> 00:39:59,530 because he obviously still has a lot of language left. 939 00:39:59,530 --> 00:40:01,840 Let's consider cases of people who 940 00:40:01,840 --> 00:40:05,480 have essentially no language due to brain damage. 941 00:40:05,480 --> 00:40:07,970 So this is known as global aphasia. 942 00:40:07,970 --> 00:40:10,810 And Rosemary Varley in England has been studying 943 00:40:10,810 --> 00:40:11,867 a group of three people-- 944 00:40:11,867 --> 00:40:13,450 I think she's got a few more, but here 945 00:40:13,450 --> 00:40:15,340 are her three main ones-- 946 00:40:15,340 --> 00:40:16,520 who have global aphasia. 947 00:40:16,520 --> 00:40:18,550 And she's been studying them for a few years. 948 00:40:18,550 --> 00:40:21,460 And, sorry, it doesn't show here at all. 949 00:40:21,460 --> 00:40:23,200 Sorry about this lousy projector. 950 00:40:23,200 --> 00:40:24,010 Shows on my screen. 951 00:40:24,010 --> 00:40:26,530 They're big, nasty, lesions taking up 952 00:40:26,530 --> 00:40:29,680 a lot of the left hemisphere and basically 953 00:40:29,680 --> 00:40:33,640 knocking out essentially all the language regions 954 00:40:33,640 --> 00:40:35,320 in these three individuals. 955 00:40:35,320 --> 00:40:38,020 And here's their performance on a bunch of different language 956 00:40:38,020 --> 00:40:39,610 tasks. 957 00:40:39,610 --> 00:40:43,090 They have to look at a picture and name it. 958 00:40:43,090 --> 00:40:46,150 They have to understand reversible sentences. 959 00:40:46,150 --> 00:40:49,588 That's like boy kiss girl versus girl kiss boy. 960 00:40:49,588 --> 00:40:51,130 They need to know who did the kissing 961 00:40:51,130 --> 00:40:54,990 and who got kissed, right, and a whole bunch of questions 962 00:40:54,990 --> 00:40:55,490 like that. 963 00:40:55,490 --> 00:40:58,270 And they are at chance at every one of these. 964 00:40:58,270 --> 00:41:00,970 So these are people-- not just people who can't speak. 965 00:41:00,970 --> 00:41:03,490 They're people who can't speak or understand 966 00:41:03,490 --> 00:41:06,280 language pretty much at all. 967 00:41:06,280 --> 00:41:09,100 So it's as close as we can get to a case of a person who 968 00:41:09,100 --> 00:41:11,530 has no language ability. 969 00:41:11,530 --> 00:41:15,370 So can these people think? 970 00:41:15,370 --> 00:41:18,870 So Rosemary Varley has done paper 971 00:41:18,870 --> 00:41:22,200 after paper in which she finds clever ways to communicate 972 00:41:22,200 --> 00:41:24,510 tasks to these people to find out what kind of thinking 973 00:41:24,510 --> 00:41:25,680 they're capable of. 974 00:41:25,680 --> 00:41:27,390 Here's one. 975 00:41:27,390 --> 00:41:30,563 You have to order this series of pictures. 976 00:41:30,563 --> 00:41:31,980 So look at it for a second and you 977 00:41:31,980 --> 00:41:38,760 can figure out that it goes basically from to left. 978 00:41:38,760 --> 00:41:42,000 So can people with global aphasia do this task? 979 00:41:42,000 --> 00:41:47,140 Yes, they're perfect at it, no problem whatsoever. 980 00:41:47,140 --> 00:41:49,140 Now you might dispute, is that cause and effect. 981 00:41:49,140 --> 00:41:50,408 Is it knowledge of sequences? 982 00:41:50,408 --> 00:41:51,200 Are they different? 983 00:41:51,200 --> 00:41:51,742 I don't know. 984 00:41:51,742 --> 00:41:54,920 But anyway, it's a pretty rich task here. 985 00:41:54,920 --> 00:41:56,960 Here's another task. 986 00:41:56,960 --> 00:41:58,940 Look at these pictures and tell which 987 00:41:58,940 --> 00:42:00,710 of them are things you know and which 988 00:42:00,710 --> 00:42:04,280 of them are things you have never seen before that I drew. 989 00:42:07,710 --> 00:42:10,325 Takes a moment, but you can figure it out. 990 00:42:10,325 --> 00:42:11,700 Top three things are real things, 991 00:42:11,700 --> 00:42:15,210 and those three things are things I drew. 992 00:42:15,210 --> 00:42:17,347 So we could ask, does a person with globalization 993 00:42:17,347 --> 00:42:18,180 know the difference. 994 00:42:18,180 --> 00:42:20,333 Basically, do you have to be able to name 995 00:42:20,333 --> 00:42:22,500 things to know the difference of what's a real thing 996 00:42:22,500 --> 00:42:24,300 and what's not? 997 00:42:24,300 --> 00:42:27,180 Here's another task. 998 00:42:27,180 --> 00:42:30,820 Which of these is the plausible event? 999 00:42:30,820 --> 00:42:34,630 That's more complicated, because here we just need to know, 1000 00:42:34,630 --> 00:42:36,280 is that a real thing that I know. 1001 00:42:36,280 --> 00:42:38,380 Here we need to know, who's doing what to who, 1002 00:42:38,380 --> 00:42:39,760 and does it make sense? 1003 00:42:39,760 --> 00:42:41,768 So it taps world knowledge, figuring out 1004 00:42:41,768 --> 00:42:43,810 who's doing what to whom, which many people think 1005 00:42:43,810 --> 00:42:46,040 is at the core of language. 1006 00:42:46,040 --> 00:42:49,900 So how do people with global aphasia do? 1007 00:42:49,900 --> 00:42:52,210 Perfectly at both of these things. 1008 00:42:52,210 --> 00:42:55,420 Well, not perfectly, but the same as control subjects. 1009 00:42:55,420 --> 00:42:56,620 Yeah, Carly? 1010 00:42:56,620 --> 00:42:57,820 AUDIENCE: I'm just confused. 1011 00:42:57,820 --> 00:43:00,250 Like, how do you get the question 1012 00:43:00,250 --> 00:43:02,140 across what they need to do? 1013 00:43:02,140 --> 00:43:03,723 NANCY KANWISHER: I don't know exactly, 1014 00:43:03,723 --> 00:43:11,110 but you do something like, for example. 1015 00:43:11,110 --> 00:43:12,430 Do you ever play charades? 1016 00:43:12,430 --> 00:43:13,275 Like that. 1017 00:43:13,275 --> 00:43:17,740 AUDIENCE: So, like, it's not exactly-- 1018 00:43:17,740 --> 00:43:19,690 couldn't someone argue that there's 1019 00:43:19,690 --> 00:43:24,446 actions that you're doing or some kind of form of language? 1020 00:43:24,446 --> 00:43:26,300 NANCY KANWISHER: They're communication. 1021 00:43:26,300 --> 00:43:28,030 They're not language. 1022 00:43:28,030 --> 00:43:30,280 So when we say language, we really mean language. 1023 00:43:30,280 --> 00:43:32,110 Not necessarily noises coming out of 1024 00:43:32,110 --> 00:43:35,345 the mouth, because American Sign Language counts. 1025 00:43:35,345 --> 00:43:37,720 And I didn't have time to put that in this lecture, which 1026 00:43:37,720 --> 00:43:40,090 is a damn shame, because it really does count in every way 1027 00:43:40,090 --> 00:43:42,520 and is very interesting and uses similar neural structures 1028 00:43:42,520 --> 00:43:44,710 and all that stuff. 1029 00:43:44,710 --> 00:43:47,140 But language is different than communication. 1030 00:43:47,140 --> 00:43:50,270 There's all kinds of ways of communicating. 1031 00:43:50,270 --> 00:43:50,770 Yeah? 1032 00:43:50,770 --> 00:43:53,267 AUDIENCE: And how old are these patients again? 1033 00:43:53,267 --> 00:43:54,850 NANCY KANWISHER: I don't know exactly, 1034 00:43:54,850 --> 00:43:57,010 but it's almost always strokes. 1035 00:43:57,010 --> 00:43:58,337 They're probably 40 to 60. 1036 00:43:58,337 --> 00:43:59,920 AUDIENCE: So it's definitely an adult. 1037 00:43:59,920 --> 00:44:03,313 I mean, it's not a infant thing. 1038 00:44:03,313 --> 00:44:04,480 NANCY KANWISHER: No, no, no. 1039 00:44:04,480 --> 00:44:07,510 These are all people who had brain damage in midlife 1040 00:44:07,510 --> 00:44:09,910 or later in life. 1041 00:44:09,910 --> 00:44:13,660 So that's pretty impressive, OK. 1042 00:44:13,660 --> 00:44:17,590 So, basically, these people with global aphasia 1043 00:44:17,590 --> 00:44:22,150 are able to do every single task that Rosemary Varley has 1044 00:44:22,150 --> 00:44:23,290 tested them on. 1045 00:44:23,290 --> 00:44:27,320 So I just showed you causality, nonverbal meaning. 1046 00:44:27,320 --> 00:44:28,150 Here's a cool one. 1047 00:44:28,150 --> 00:44:30,790 Remember reorientation-- you should. 1048 00:44:30,790 --> 00:44:32,350 May well be on the final exam. 1049 00:44:32,350 --> 00:44:35,680 To remind you, I did a whole most of a lecture 1050 00:44:35,680 --> 00:44:37,630 on this thing about reorientation. 1051 00:44:37,630 --> 00:44:41,860 Remember, rats and infants, if you hide food there and put 1052 00:44:41,860 --> 00:44:45,700 them in this box, they later go 50-50 to the two corners, 1053 00:44:45,700 --> 00:44:48,400 even though that wall should disambiguate which 1054 00:44:48,400 --> 00:44:50,530 is the exactly correct corner. 1055 00:44:50,530 --> 00:44:52,240 They should always go here. 1056 00:44:52,240 --> 00:44:56,170 They have the knowledge that it's there, but they go 50-50. 1057 00:44:56,170 --> 00:44:58,090 And, remember, I said that Liz Spelke 1058 00:44:58,090 --> 00:45:00,820 has this interesting argument that the key thing you 1059 00:45:00,820 --> 00:45:04,210 need to be able to solve that task is language. 1060 00:45:04,210 --> 00:45:06,118 Because, in fact, if you test adults 1061 00:45:06,118 --> 00:45:07,660 and you tie up their language system, 1062 00:45:07,660 --> 00:45:10,030 they behave like infants and rats. 1063 00:45:10,030 --> 00:45:12,190 But if you don't tie up their language system, 1064 00:45:12,190 --> 00:45:15,430 they can do the task, which is pretty suggestive that language 1065 00:45:15,430 --> 00:45:16,990 is a crux of the matter. 1066 00:45:16,990 --> 00:45:20,560 However, the global aphasic do this task just fine. 1067 00:45:23,240 --> 00:45:26,810 So now, we have to go to [? Min ?] Young's hypothesis, 1068 00:45:26,810 --> 00:45:30,610 which is that maybe the role of language in reorientation 1069 00:45:30,610 --> 00:45:33,730 is learning about that whole spatial system 1070 00:45:33,730 --> 00:45:36,580 during childhood, which the global aphasics could do, 1071 00:45:36,580 --> 00:45:41,200 not maintaining the ability once you've gained it. 1072 00:45:41,200 --> 00:45:42,280 All right, they can do-- 1073 00:45:42,280 --> 00:45:43,905 I won't give you all the data on this-- 1074 00:45:43,905 --> 00:45:45,490 but they can do arithmetic tasks, 1075 00:45:45,490 --> 00:45:48,340 logic tasks, algebra tasks. 1076 00:45:48,340 --> 00:45:50,200 They appreciate music. 1077 00:45:50,200 --> 00:45:54,130 They can think about what other people are thinking. 1078 00:45:54,130 --> 00:45:57,130 So everything that all that kind of high level 1079 00:45:57,130 --> 00:46:00,010 abstract quintessentially human abilities 1080 00:46:00,010 --> 00:46:03,100 that we are impressed with ourselves for being able to do, 1081 00:46:03,100 --> 00:46:05,380 these people can do without language. 1082 00:46:05,380 --> 00:46:10,170 So language and thought are not the same thing. 1083 00:46:10,170 --> 00:46:12,430 You can still think in lots of different ways, 1084 00:46:12,430 --> 00:46:14,610 even after you lose language. 1085 00:46:14,610 --> 00:46:17,670 On the other hand, as has already been brought up, 1086 00:46:17,670 --> 00:46:20,795 global aphasics had language during development. 1087 00:46:20,795 --> 00:46:22,920 So saying that you don't need it as an adult is not 1088 00:46:22,920 --> 00:46:25,510 the same as saying you don't need it during development. 1089 00:46:25,510 --> 00:46:28,800 You absolutely do need it during development, 1090 00:46:28,800 --> 00:46:32,170 because it's a key way we learn about the world. 1091 00:46:32,170 --> 00:46:36,030 And for example, there are studies from Rebecca Sachs' lab 1092 00:46:36,030 --> 00:46:39,420 showing that deaf kids who learn language later-- 1093 00:46:39,420 --> 00:46:42,250 for example, if they're born not to deaf parents 1094 00:46:42,250 --> 00:46:45,210 but to hearing parents who don't cotton on to the fact 1095 00:46:45,210 --> 00:46:48,060 that it's important for them to learn ASL early, 1096 00:46:48,060 --> 00:46:51,030 and hence they don't get language until later, 1097 00:46:51,030 --> 00:46:53,820 those kids are not as good at understanding 1098 00:46:53,820 --> 00:46:56,520 what other people are thinking, something that we usually 1099 00:46:56,520 --> 00:46:57,765 learn about through language. 1100 00:47:02,370 --> 00:47:05,610 Further, even though I'm making a big deal 1101 00:47:05,610 --> 00:47:07,650 about how you can think without language, 1102 00:47:07,650 --> 00:47:12,600 I'm not saying that language is irrelevant to thinking. 1103 00:47:12,600 --> 00:47:14,220 Every time I write a grant proposal, 1104 00:47:14,220 --> 00:47:16,320 I think, oh god, I have all these ideas in my head 1105 00:47:16,320 --> 00:47:17,700 and now I have to waste weeks and weeks 1106 00:47:17,700 --> 00:47:19,408 and weeks, blah, blah, blah, putting them 1107 00:47:19,408 --> 00:47:23,100 all down on paper to try to get money to fund my habit. 1108 00:47:23,100 --> 00:47:24,990 And then I get into like sentence 3 1109 00:47:24,990 --> 00:47:26,910 and I suddenly realize, oh, oh no, I 1110 00:47:26,910 --> 00:47:29,160 haven't been thinking about this clearly at all. 1111 00:47:29,160 --> 00:47:31,440 So this is my very informal introspection 1112 00:47:31,440 --> 00:47:33,690 on the role of language in my own thinking. 1113 00:47:33,690 --> 00:47:36,060 Like, even when I think there's a clear thought, 1114 00:47:36,060 --> 00:47:37,710 the same thing happens when I go to prepare a lecture. 1115 00:47:37,710 --> 00:47:39,335 It's, like, oh yeah, I know this stuff. 1116 00:47:39,335 --> 00:47:41,860 But I put together some slides, I'm like slide 2, 1117 00:47:41,860 --> 00:47:45,000 no, I don't really know this stuff. 1118 00:47:45,000 --> 00:47:48,810 So there is some role for language and thinking. 1119 00:47:48,810 --> 00:47:52,440 And I'll give you one example here. 1120 00:47:52,440 --> 00:47:54,660 One of the many things that language can do 1121 00:47:54,660 --> 00:47:58,350 is to make information more salient. 1122 00:47:58,350 --> 00:48:02,070 So right now, close your eyes, everyone close your eyes. 1123 00:48:02,070 --> 00:48:04,860 I mean it, I see if they're open. 1124 00:48:04,860 --> 00:48:09,220 While keeping your eyes closed, point south. 1125 00:48:09,220 --> 00:48:11,430 You may not exactly know where south 1126 00:48:11,430 --> 00:48:12,870 is, but make a good guess. 1127 00:48:15,590 --> 00:48:17,600 Use your whole arm so everyone can 1128 00:48:17,600 --> 00:48:19,170 see when they open their eyes. 1129 00:48:22,220 --> 00:48:24,290 Keep pointing, but now you can open your eyes 1130 00:48:24,290 --> 00:48:25,430 and you can look around and see where 1131 00:48:25,430 --> 00:48:26,513 everyone else is pointing. 1132 00:48:26,513 --> 00:48:29,900 You guys are not bad, not bad, not bad. 1133 00:48:29,900 --> 00:48:33,530 But we got some over here, we're a little turned around here. 1134 00:48:33,530 --> 00:48:35,135 Anyway, it's roughly over there. 1135 00:48:37,790 --> 00:48:40,490 So, yeah, hang on, wait a second. 1136 00:48:40,490 --> 00:48:41,260 Yes, right. 1137 00:48:43,890 --> 00:48:44,592 Well, hang on. 1138 00:48:44,592 --> 00:48:45,800 Yeah, right, it's over there. 1139 00:48:48,890 --> 00:48:51,290 So your vector average was closer 1140 00:48:51,290 --> 00:48:56,100 to the true thing than a random vector, but not so hot. 1141 00:48:56,100 --> 00:48:58,830 If your language forced you to keep track of this, 1142 00:48:58,830 --> 00:49:00,450 you'd be better at it. 1143 00:49:00,450 --> 00:49:04,410 And we know that from the case of the Pormpuraaw, 1144 00:49:04,410 --> 00:49:07,050 these guys here, who live in Australia. 1145 00:49:07,050 --> 00:49:08,820 It's Aboriginal people. 1146 00:49:08,820 --> 00:49:13,500 And they spend a lot of time going around 1147 00:49:13,500 --> 00:49:15,940 in the remote outback of Australia, 1148 00:49:15,940 --> 00:49:17,700 where they need to know where they are. 1149 00:49:17,700 --> 00:49:21,090 And who is going where and when is really 1150 00:49:21,090 --> 00:49:22,950 of the essence in their lives and 1151 00:49:22,950 --> 00:49:24,760 in their social interactions. 1152 00:49:24,760 --> 00:49:26,640 So when they run into each other, 1153 00:49:26,640 --> 00:49:29,460 they don't say, hi, how are you. 1154 00:49:29,460 --> 00:49:32,670 Instead, they say, which way are you going. 1155 00:49:32,670 --> 00:49:35,550 And a typical answer might be, "North northwest 1156 00:49:35,550 --> 00:49:37,500 in the middle distance, how about you?" 1157 00:49:40,820 --> 00:49:43,430 They don't talk about things being left or right 1158 00:49:43,430 --> 00:49:45,590 or behind them, reference frames that 1159 00:49:45,590 --> 00:49:48,650 have to do with the person's own body, which are frankly 1160 00:49:48,650 --> 00:49:50,330 really stupid reference frames. 1161 00:49:50,330 --> 00:49:52,610 Because I can say this thing is to the left. 1162 00:49:52,610 --> 00:49:54,920 And then I turn, and now it's not to the left anymore. 1163 00:49:54,920 --> 00:49:57,260 Like, how stupid is that, right? 1164 00:49:57,260 --> 00:49:59,210 These guys have a much better system. 1165 00:49:59,210 --> 00:50:00,710 They would rather say, oh, "You have 1166 00:50:00,710 --> 00:50:04,150 a bug on your southeast leg." 1167 00:50:04,150 --> 00:50:05,840 Right, OK. 1168 00:50:05,840 --> 00:50:08,780 So these guys, people who speak this language, 1169 00:50:08,780 --> 00:50:11,630 they have to be aware of absolute compass directions 1170 00:50:11,630 --> 00:50:14,870 all the time just to speak. 1171 00:50:14,870 --> 00:50:18,230 And so they're oriented all the time, unlike us. 1172 00:50:18,230 --> 00:50:20,060 And in that sense, their language 1173 00:50:20,060 --> 00:50:22,920 makes salient certain kinds of information. 1174 00:50:22,920 --> 00:50:24,852 It's not that we can't think about direction. 1175 00:50:24,852 --> 00:50:26,810 It's just that most of the time we're not aware 1176 00:50:26,810 --> 00:50:29,143 because our language doesn't force us to think about it. 1177 00:50:31,760 --> 00:50:34,770 So interim summary. 1178 00:50:34,770 --> 00:50:37,250 We've been asking this question of whether thought 1179 00:50:37,250 --> 00:50:39,897 is separate from and possible without language. 1180 00:50:39,897 --> 00:50:41,480 Before you guys take off, you wrote it 1181 00:50:41,480 --> 00:50:43,010 on the board, this board right here? 1182 00:50:43,010 --> 00:50:44,690 Awesome. 1183 00:50:44,690 --> 00:50:47,190 You guys need to tell me when there's time to take the quiz. 1184 00:50:47,190 --> 00:50:49,398 So you're going to have seven minutes because there's 1185 00:50:49,398 --> 00:50:50,240 seven questions. 1186 00:50:50,240 --> 00:50:59,180 And so at 12:18 let me know and I will turn the board around. 1187 00:50:59,180 --> 00:51:01,670 OK, 12:17 because it'll take me a minute to turn around. 1188 00:51:01,670 --> 00:51:02,840 All right, thank you. 1189 00:51:02,840 --> 00:51:05,390 Take notes, tell me about that time. 1190 00:51:05,390 --> 00:51:07,490 So here's a question we've been engaging in, 1191 00:51:07,490 --> 00:51:10,370 is thought separate from and possible without language? 1192 00:51:10,370 --> 00:51:13,490 And the literature from neuroscience psych patients 1193 00:51:13,490 --> 00:51:16,550 says, yes, absolutely, they're totally separate. 1194 00:51:16,550 --> 00:51:20,460 Global aphasics have many forms of thought without language. 1195 00:51:20,460 --> 00:51:24,910 So given that, what would you predict from functional MRI? 1196 00:51:24,910 --> 00:51:27,720 So if I told you, which is true, that these are the brain 1197 00:51:27,720 --> 00:51:30,930 regions that are active during language tasks, 1198 00:51:30,930 --> 00:51:34,690 for example, when you understand the meaning of a sentence, 1199 00:51:34,690 --> 00:51:35,800 what would you predict? 1200 00:51:35,800 --> 00:51:39,550 Should they be activated only by language, not 1201 00:51:39,550 --> 00:51:41,440 by non-linguistic tasks? 1202 00:51:41,440 --> 00:51:42,320 What do you think? 1203 00:51:42,320 --> 00:51:43,653 Take a moment to think about it. 1204 00:51:43,653 --> 00:51:46,070 These are the regions that are engaged when you understand 1205 00:51:46,070 --> 00:51:47,200 the meaning of a sentence. 1206 00:51:47,200 --> 00:51:49,840 Would you expect them to be engaged based on what I've just 1207 00:51:49,840 --> 00:51:53,260 told you when you do mental arithmetic, when 1208 00:51:53,260 --> 00:51:57,040 you think about spatial orientations, 1209 00:51:57,040 --> 00:52:00,290 when you appreciate music? 1210 00:52:00,290 --> 00:52:01,437 No, right. 1211 00:52:01,437 --> 00:52:03,020 If they're separate, they're separate. 1212 00:52:03,020 --> 00:52:04,895 They should go on in different brain regions. 1213 00:52:04,895 --> 00:52:06,440 Everybody have that intuition? 1214 00:52:06,440 --> 00:52:07,985 No, you don't have that intuition? 1215 00:52:07,985 --> 00:52:10,200 AUDIENCE: No. 1216 00:52:10,200 --> 00:52:14,360 I mean, do you think about things in terms of words, 1217 00:52:14,360 --> 00:52:16,913 even as a mental crutch, even if you didn't have to? 1218 00:52:16,913 --> 00:52:18,830 NANCY KANWISHER: OK, fair enough, fair enough. 1219 00:52:18,830 --> 00:52:20,970 So it doesn't nail this case. 1220 00:52:20,970 --> 00:52:24,500 It could well be that you have separate systems for all 1221 00:52:24,500 --> 00:52:27,900 those other things, but you still lean on the system. 1222 00:52:27,900 --> 00:52:30,350 Not necessarily, but you use it sometimes. 1223 00:52:30,350 --> 00:52:31,790 In fact, there's evidence for that 1224 00:52:31,790 --> 00:52:34,550 that we won't get to today. 1225 00:52:34,550 --> 00:52:39,320 But the initial thought is, you don't need to activate it. 1226 00:52:39,320 --> 00:52:41,090 Well, here's a surprise. 1227 00:52:41,090 --> 00:52:44,750 Up until recently, pretty much the whole brain imaging 1228 00:52:44,750 --> 00:52:47,330 literature says that language overlaps 1229 00:52:47,330 --> 00:52:49,690 with all of these things in the brain, 1230 00:52:49,690 --> 00:52:51,440 that the activations overlap in the brain. 1231 00:52:51,440 --> 00:52:53,360 They're all the same thing. 1232 00:52:53,360 --> 00:52:57,110 That's been the received story for 20 years or so 1233 00:52:57,110 --> 00:52:58,580 of brain imaging. 1234 00:52:58,580 --> 00:53:01,200 And that just does not fit with the patient literature. 1235 00:53:01,200 --> 00:53:03,020 So we have a conundrum. 1236 00:53:03,020 --> 00:53:05,640 Here are just a few examples. 1237 00:53:05,640 --> 00:53:09,650 Stan Dehaene says, "arithmetic recruits networks involved 1238 00:53:09,650 --> 00:53:12,680 in word association processes." 1239 00:53:12,680 --> 00:53:14,330 People who study music say regions 1240 00:53:14,330 --> 00:53:16,800 such as Broca's area and Wernicke's area, 1241 00:53:16,800 --> 00:53:19,280 which have been considered specific to language, 1242 00:53:19,280 --> 00:53:22,460 are also activated by certain aspects of music. 1243 00:53:22,460 --> 00:53:24,530 Thus, the idea of language specificity 1244 00:53:24,530 --> 00:53:26,675 has been called into question, and on and on. 1245 00:53:26,675 --> 00:53:27,800 There's a million of these. 1246 00:53:27,800 --> 00:53:30,700 I just put a few of them up there. 1247 00:53:30,700 --> 00:53:33,010 So what's going on. 1248 00:53:33,010 --> 00:53:35,060 How are we going to resolve this contradiction? 1249 00:53:35,060 --> 00:53:37,490 On the one hand, the patient literature 1250 00:53:37,490 --> 00:53:39,490 suggests that language is separate from the rest 1251 00:53:39,490 --> 00:53:40,150 of thought. 1252 00:53:40,150 --> 00:53:42,950 And on the other hand, most of the neuroimaging literature 1253 00:53:42,950 --> 00:53:44,950 says that if you look at those language regions, 1254 00:53:44,950 --> 00:53:48,200 you find them activated in all these other kinds of things. 1255 00:53:48,200 --> 00:53:51,820 One hypothesis is David's, that they're activated but not 1256 00:53:51,820 --> 00:53:53,380 essentially so. 1257 00:53:53,380 --> 00:53:55,957 But there's another hypothesis. 1258 00:53:55,957 --> 00:53:58,290 And that is that there's a methodological flaw with most 1259 00:53:58,290 --> 00:54:01,340 of the prior research. 1260 00:54:01,340 --> 00:54:03,410 What is that methodological flaw? 1261 00:54:03,410 --> 00:54:05,420 It's an inappropriate use of something 1262 00:54:05,420 --> 00:54:06,560 called a group analysis. 1263 00:54:06,560 --> 00:54:08,600 I've alluded to this a few times briefly, 1264 00:54:08,600 --> 00:54:10,430 but let me do it for real now. 1265 00:54:10,430 --> 00:54:13,880 Let me first say, it's not that a group analysis 1266 00:54:13,880 --> 00:54:15,680 with functional MRI is an evil thing that 1267 00:54:15,680 --> 00:54:16,790 should never be done. 1268 00:54:16,790 --> 00:54:17,960 They have uses. 1269 00:54:17,960 --> 00:54:19,850 But particularly for the question 1270 00:54:19,850 --> 00:54:23,030 of asking whether common regions of the brain 1271 00:54:23,030 --> 00:54:24,860 are engaged in two different tasks, 1272 00:54:24,860 --> 00:54:27,380 it is not a good method, for the following reason. 1273 00:54:27,380 --> 00:54:29,930 So, first, let's say, what is a group analysis. 1274 00:54:29,930 --> 00:54:31,550 With functional MRI, it just means-- 1275 00:54:31,550 --> 00:54:33,633 and, again, I'm going to be very sketchy with this 1276 00:54:33,633 --> 00:54:35,913 because this is not an actual hands-on methods class. 1277 00:54:35,913 --> 00:54:37,580 I'm just trying to get you to understand 1278 00:54:37,580 --> 00:54:39,200 the gist of the methods. 1279 00:54:39,200 --> 00:54:41,120 You take a bunch of scanned brains 1280 00:54:41,120 --> 00:54:44,510 and you align them in a common space as best you can. 1281 00:54:44,510 --> 00:54:46,640 You can't do it perfectly because brains 1282 00:54:46,640 --> 00:54:50,240 are anatomically different from one person to the next. 1283 00:54:50,240 --> 00:54:53,480 But you do your best to align them as best you can. 1284 00:54:53,480 --> 00:54:57,110 Then you do an analysis across those aligned brains. 1285 00:54:57,110 --> 00:55:00,470 And you ask, what is consistent across this group of subjects. 1286 00:55:00,470 --> 00:55:02,300 That's a very useful question to ask. 1287 00:55:02,300 --> 00:55:05,180 If we want to know overall what are the brain regions that 1288 00:55:05,180 --> 00:55:07,880 are consistently activated when you understand language 1289 00:55:07,880 --> 00:55:09,410 across this whole group of subjects, 1290 00:55:09,410 --> 00:55:11,620 that's a good use of a group analysis. 1291 00:55:11,620 --> 00:55:13,370 You'll find that picture I just showed you 1292 00:55:13,370 --> 00:55:15,537 before with stuff going down the left temporal lobe, 1293 00:55:15,537 --> 00:55:17,270 a bunch of left frontal lobe stuff. 1294 00:55:17,270 --> 00:55:19,250 And that will be a very blurry picture 1295 00:55:19,250 --> 00:55:21,902 of the regions that are most consistent across subjects. 1296 00:55:21,902 --> 00:55:22,610 Yes, [INAUDIBLE]. 1297 00:55:22,610 --> 00:55:24,980 AUDIENCE: Do you line them anatomically 1298 00:55:24,980 --> 00:55:26,990 as you new [INAUDIBLE] side to each other? 1299 00:55:26,990 --> 00:55:29,070 Or do you line them functionally, 1300 00:55:29,070 --> 00:55:32,090 so you can look at the scans in the functional [? region? ?] 1301 00:55:32,090 --> 00:55:34,940 NANCY KANWISHER: So therein lies a universe of options. 1302 00:55:34,940 --> 00:55:37,040 What I'm talking about now is a group analysis 1303 00:55:37,040 --> 00:55:39,050 is aligning them anatomically. 1304 00:55:39,050 --> 00:55:40,850 And that's where the problem comes in. 1305 00:55:40,850 --> 00:55:42,683 And where we're going to go from that is you 1306 00:55:42,683 --> 00:55:44,570 need to align them functionally. 1307 00:55:44,570 --> 00:55:47,900 If you just align them anatomically, 1308 00:55:47,900 --> 00:55:50,000 then the following can happen. 1309 00:55:50,000 --> 00:55:52,430 So you do a standard group analysis 1310 00:55:52,430 --> 00:55:55,310 and you say, for example, let's do a language task 1311 00:55:55,310 --> 00:55:57,860 and arithmetic task and a music task. 1312 00:55:57,860 --> 00:55:59,720 And let's suppose you find this-- 1313 00:55:59,720 --> 00:56:03,860 basically, Broca's area vicinity is activated in an overlapping 1314 00:56:03,860 --> 00:56:06,050 fashion in all three. 1315 00:56:06,050 --> 00:56:10,310 Each of those is based on an analysis of 12 or 20 subjects 1316 00:56:10,310 --> 00:56:12,830 aligned as best we can. 1317 00:56:12,830 --> 00:56:14,810 So that's basically what the literature shows 1318 00:56:14,810 --> 00:56:17,390 is lots of stuff like that. 1319 00:56:17,390 --> 00:56:18,770 But here's the problem. 1320 00:56:18,770 --> 00:56:21,860 You can get that result in a group analysis, 1321 00:56:21,860 --> 00:56:25,370 even if the actual data looks like this 1322 00:56:25,370 --> 00:56:27,740 in each individual subject. 1323 00:56:27,740 --> 00:56:31,430 No overlap at all in any subject, 1324 00:56:31,430 --> 00:56:34,470 but those regions are in slightly different locations. 1325 00:56:34,470 --> 00:56:39,340 And so if you average across this, you get that. 1326 00:56:39,340 --> 00:56:41,300 Everybody see the problem? 1327 00:56:41,300 --> 00:56:43,870 So it's not that it's a bad idea to do a group analysis. 1328 00:56:43,870 --> 00:56:45,940 It's a nice, initial, blurry picture 1329 00:56:45,940 --> 00:56:48,970 of the approximate consistent locations 1330 00:56:48,970 --> 00:56:50,650 in the brain for a given task. 1331 00:56:50,650 --> 00:56:53,290 The problem is when you say, oh, there's overlap, 1332 00:56:53,290 --> 00:56:55,330 therefore they're the same thing, 1333 00:56:55,330 --> 00:56:57,040 because you can get this result even 1334 00:56:57,040 --> 00:57:00,940 if there's no overlap in any subject at all. 1335 00:57:00,940 --> 00:57:02,890 So the whole literature did this for 20 years 1336 00:57:02,890 --> 00:57:04,660 and made all this talk about how language 1337 00:57:04,660 --> 00:57:06,670 is on top of everything else in the brain. 1338 00:57:06,670 --> 00:57:09,160 And for a long time I was sitting by the sidelines 1339 00:57:09,160 --> 00:57:11,740 going, oh my god. 1340 00:57:11,740 --> 00:57:15,763 And then, eventually, Ev Fedorenko came along 1341 00:57:15,763 --> 00:57:16,930 and she knew about language. 1342 00:57:16,930 --> 00:57:19,390 And I said, let's figure out, maybe they're right, 1343 00:57:19,390 --> 00:57:22,610 maybe that's true, or maybe it's like this. 1344 00:57:22,610 --> 00:57:24,850 Let's find out. 1345 00:57:24,850 --> 00:57:27,250 So how do we do that? 1346 00:57:27,250 --> 00:57:29,410 What you do is exactly what [INAUDIBLE] 1347 00:57:29,410 --> 00:57:30,520 mentioned a moment ago. 1348 00:57:30,520 --> 00:57:33,970 You align them not anatomically but functionally. 1349 00:57:33,970 --> 00:57:36,580 That's a whole reason to use functional reasons of interest. 1350 00:57:36,580 --> 00:57:38,860 We've encountered this before when I was carrying on 1351 00:57:38,860 --> 00:57:41,800 about why we do functional localizers with the fusiform 1352 00:57:41,800 --> 00:57:42,310 face area. 1353 00:57:42,310 --> 00:57:43,600 This is the same deal. 1354 00:57:43,600 --> 00:57:46,675 It's just that that insight started in the back of the head 1355 00:57:46,675 --> 00:57:48,550 and hasn't reached the front of the head yet, 1356 00:57:48,550 --> 00:57:49,820 or it's about here. 1357 00:57:49,820 --> 00:57:51,028 So some people get it here. 1358 00:57:51,028 --> 00:57:53,320 And the farther forward you go, the less people realize 1359 00:57:53,320 --> 00:57:55,390 this is an issue, which is really ridiculous, 1360 00:57:55,390 --> 00:57:57,900 because it gets more and more important as you go this way. 1361 00:57:57,900 --> 00:57:59,650 Some stuff is actually aligned in the back 1362 00:57:59,650 --> 00:58:01,330 and nothing is aligned in the front. 1363 00:58:01,330 --> 00:58:03,280 Anyway, so what do you do? 1364 00:58:03,280 --> 00:58:06,910 You do just what we did with the FFA and all the other regions. 1365 00:58:06,910 --> 00:58:09,550 One, in each subject individually, 1366 00:58:09,550 --> 00:58:12,370 you identify those language regions. 1367 00:58:12,370 --> 00:58:13,687 You run some localizer. 1368 00:58:13,687 --> 00:58:15,520 It's like, OK, I got this and that and that. 1369 00:58:15,520 --> 00:58:17,740 And then once you identified them, 1370 00:58:17,740 --> 00:58:20,620 you can ask, OK, does that region in that subject 1371 00:58:20,620 --> 00:58:22,060 show activation for arithmetic. 1372 00:58:22,060 --> 00:58:25,090 No, that's next door, right, et cetera. 1373 00:58:25,090 --> 00:58:25,990 Everybody got this? 1374 00:58:25,990 --> 00:58:27,520 This is really important. 1375 00:58:27,520 --> 00:58:30,380 I guess just because I'm obsessed with it. 1376 00:58:30,380 --> 00:58:32,380 I honestly don't know if it's globally important 1377 00:58:32,380 --> 00:58:33,963 or if it's just my personal obsession, 1378 00:58:33,963 --> 00:58:35,770 but you need to know it for this course. 1379 00:58:35,770 --> 00:58:38,560 We'll leave it at that. 1380 00:58:38,560 --> 00:58:41,020 So this is standard in people who study vision 1381 00:58:41,020 --> 00:58:43,660 and it's less standard in people who work in other domains. 1382 00:58:43,660 --> 00:58:46,720 But they're slowly cottoning on. 1383 00:58:46,720 --> 00:58:49,780 So how do we identify language regions in each subject 1384 00:58:49,780 --> 00:58:51,310 individually? 1385 00:58:51,310 --> 00:58:53,500 There are lots of possible ways to do this. 1386 00:58:53,500 --> 00:58:56,320 But here's the way I'm going to show you 1387 00:58:56,320 --> 00:58:59,180 that's been used a bunch by Fedorenko and others. 1388 00:58:59,180 --> 00:59:02,410 So we start by saying, OK, let's find candidate brain 1389 00:59:02,410 --> 00:59:06,070 regions that respond to language, which I told you, 1390 00:59:06,070 --> 00:59:07,840 by language, I mean sentence understanding 1391 00:59:07,840 --> 00:59:09,590 for present purposes. 1392 00:59:09,590 --> 00:59:11,680 So if we want to look at sentence understanding, 1393 00:59:11,680 --> 00:59:13,700 we've got to start with sentence understanding. 1394 00:59:13,700 --> 00:59:15,340 So if you look at the screen, you'll 1395 00:59:15,340 --> 00:59:16,930 see some of the stimuli we use. 1396 00:59:20,870 --> 00:59:24,200 So subject is lying in the scanner and they see that. 1397 00:59:24,200 --> 00:59:26,220 And then we can either give them a task or not. 1398 00:59:26,220 --> 00:59:28,800 And we'll talk about that in a second. 1399 00:59:28,800 --> 00:59:30,380 What are we going to compare it to? 1400 00:59:30,380 --> 00:59:32,422 Well, there are lots and lots of different things 1401 00:59:32,422 --> 00:59:34,920 we could compare it to that control for different things. 1402 00:59:34,920 --> 00:59:38,825 But we started off with this, if you read this here. 1403 00:59:44,410 --> 00:59:47,080 So the idea is, it's visually similar. 1404 00:59:47,080 --> 00:59:48,910 You can hear the sounds in your head. 1405 00:59:48,910 --> 00:59:51,070 You can pronounce those things to yourself. 1406 00:59:51,070 --> 00:59:55,720 But there's really no syntax and no meaning-- 1407 00:59:55,720 --> 00:59:59,770 not perfect, but a first pass. 1408 00:59:59,770 --> 01:00:02,290 So when you do that, you get activations 1409 01:00:02,290 --> 01:00:03,160 that look like this. 1410 01:00:03,160 --> 01:00:04,660 Here are four different subjects. 1411 01:00:04,660 --> 01:00:07,120 And you can see they're very systematic things. 1412 01:00:07,120 --> 01:00:08,290 See these three blobs-- 1413 01:00:08,290 --> 01:00:10,270 boom, boom, boom, boom, boom, boom-- 1414 01:00:10,270 --> 01:00:13,180 in each subject, and a bunch of stuff in the temporal lobe 1415 01:00:13,180 --> 01:00:14,770 like that in each subject. 1416 01:00:14,770 --> 01:00:19,880 They're quite systematic but absolutely not identical. 1417 01:00:19,880 --> 01:00:22,820 All right, so, yeah, it's just what I did. 1418 01:00:22,820 --> 01:00:24,120 So now what do you do next? 1419 01:00:24,120 --> 01:00:25,670 Well, we just made this up, sentences 1420 01:00:25,670 --> 01:00:26,765 versus non-word strings. 1421 01:00:26,765 --> 01:00:29,270 Well, who says that's a good thing to do? 1422 01:00:29,270 --> 01:00:31,040 So the next thing you do is you've 1423 01:00:31,040 --> 01:00:34,250 got to validate your localizer task to make sure it isn't just 1424 01:00:34,250 --> 01:00:35,820 like trivial in some sense. 1425 01:00:35,820 --> 01:00:39,200 So the first question is, is it reliable? 1426 01:00:39,200 --> 01:00:41,930 So, here's session 1, three different subjects' 1427 01:00:41,930 --> 01:00:43,370 activations. 1428 01:00:43,370 --> 01:00:45,050 Well, just scan them again. 1429 01:00:45,050 --> 01:00:47,925 There's a lot of talk about fancy statistics, blah, blah 1430 01:00:47,925 --> 01:00:48,590 blah. 1431 01:00:48,590 --> 01:00:51,060 Just scan them again. 1432 01:00:51,060 --> 01:00:54,390 Wow, look how similar these two little hot spots, 1433 01:00:54,390 --> 01:00:55,860 this elongated one. 1434 01:00:55,860 --> 01:00:58,380 I mean, it's remarkable, extremely 1435 01:00:58,380 --> 01:01:01,230 reliable within a subject, and yet somewhat 1436 01:01:01,230 --> 01:01:06,170 different across subjects, so check one, reliable. 1437 01:01:06,170 --> 01:01:09,850 More interestingly, does it generalize across task 1438 01:01:09,850 --> 01:01:11,650 and presentation modality? 1439 01:01:11,650 --> 01:01:13,635 So before we just had people reading sentences. 1440 01:01:13,635 --> 01:01:15,010 And I keep saying, reading is not 1441 01:01:15,010 --> 01:01:17,540 the native form of language. 1442 01:01:17,540 --> 01:01:20,050 So let's replicate that reading. 1443 01:01:20,050 --> 01:01:22,068 And now we're adding a memory task. 1444 01:01:22,068 --> 01:01:24,610 So at the end of each string, a little probe comes up and you 1445 01:01:24,610 --> 01:01:28,120 have to say, was this word or a non-word in the previous 1446 01:01:28,120 --> 01:01:29,200 thing-- 1447 01:01:29,200 --> 01:01:31,360 sequence. 1448 01:01:31,360 --> 01:01:34,870 And let's compare that to just listening to the sentences. 1449 01:01:34,870 --> 01:01:37,570 Wow, look how similar. 1450 01:01:37,570 --> 01:01:42,070 So that tells us that we're not studying reading or speech. 1451 01:01:42,070 --> 01:01:45,010 We're studying language after those things converge. 1452 01:01:45,010 --> 01:01:48,220 Those regions don't care if you saw a word or heard the word. 1453 01:01:48,220 --> 01:01:50,137 They just care if you're representing 1454 01:01:50,137 --> 01:01:51,220 the meaning of a sentence. 1455 01:01:51,220 --> 01:01:54,610 Everybody with me why that's important? 1456 01:01:54,610 --> 01:01:56,950 All right, check, check. 1457 01:01:56,950 --> 01:01:58,630 Does it generalize across languages? 1458 01:01:58,630 --> 01:02:02,110 Suppose you're bilingual and speak two different languages. 1459 01:02:02,110 --> 01:02:05,590 Here's two subjects who speak both English and Spanish. 1460 01:02:05,590 --> 01:02:08,350 Wow, look how similar. 1461 01:02:08,350 --> 01:02:10,390 So it's really language in general, 1462 01:02:10,390 --> 01:02:15,310 not English or Spanish or a particular language. 1463 01:02:15,310 --> 01:02:18,070 Does it generalize across materials? 1464 01:02:18,070 --> 01:02:20,598 So we could have reading sentences versus non-words 1465 01:02:20,598 --> 01:02:22,390 that we've been talking about here with two 1466 01:02:22,390 --> 01:02:24,610 different runs in one subject. 1467 01:02:24,610 --> 01:02:26,380 Are we going have subjects listening 1468 01:02:26,380 --> 01:02:28,602 to speech versus degraded speech, like this? 1469 01:02:28,602 --> 01:02:29,560 Here's the speech case. 1470 01:02:29,560 --> 01:02:30,227 [VIDEO PLAYBACK] 1471 01:02:30,227 --> 01:02:32,470 - During my days of house arrest, 1472 01:02:32,470 --> 01:02:36,250 it felt as though I were no longer part of the real world. 1473 01:02:36,250 --> 01:02:37,855 NANCY KANWISHER: OK, versus this. 1474 01:02:37,855 --> 01:02:43,727 - [INAUDIBLE] 1475 01:02:43,727 --> 01:02:44,310 [END PLAYBACK] 1476 01:02:44,310 --> 01:02:45,770 NANCY KANWISHER: OK, so very degraded. 1477 01:02:45,770 --> 01:02:47,430 You can't understand what's being said, 1478 01:02:47,430 --> 01:02:50,720 but it has similar prosody and some similar structure. 1479 01:02:50,720 --> 01:02:53,900 And the point is, you get very similar activations 1480 01:02:53,900 --> 01:02:58,190 with those very different kinds of contrasts. 1481 01:02:58,190 --> 01:03:01,480 So now we have really validated this thing. 1482 01:03:01,480 --> 01:03:03,650 It checks out in all the ways it should. 1483 01:03:03,650 --> 01:03:05,780 It doesn't care about modality. 1484 01:03:05,780 --> 01:03:09,650 It does care about meaning. 1485 01:03:09,650 --> 01:03:11,610 And it's highly reliable. 1486 01:03:11,610 --> 01:03:13,520 So now we can put it to use. 1487 01:03:13,520 --> 01:03:17,120 Now we can ask, what does each of those regions do? 1488 01:03:17,120 --> 01:03:20,660 All right, so to do that, in each participant 1489 01:03:20,660 --> 01:03:24,170 then we find those regions with this localizer. 1490 01:03:24,170 --> 01:03:25,790 Now let me just step back a second. 1491 01:03:25,790 --> 01:03:28,610 There's nothing magic about this localizer per se. 1492 01:03:28,610 --> 01:03:31,160 When you want to study something, 1493 01:03:31,160 --> 01:03:32,510 you use common sense. 1494 01:03:32,510 --> 01:03:34,850 You try something, you validate it. 1495 01:03:34,850 --> 01:03:37,248 It may turn out later that of the thing that we thought 1496 01:03:37,248 --> 01:03:39,290 we were identifying language with this localizer, 1497 01:03:39,290 --> 01:03:40,500 it's got this other stuff. 1498 01:03:40,500 --> 01:03:42,368 And then maybe you refine your localizer 1499 01:03:42,368 --> 01:03:43,410 into something different. 1500 01:03:43,410 --> 01:03:45,740 So it's not that this is the only possible way. 1501 01:03:45,740 --> 01:03:49,250 It was just a sensible approach. 1502 01:03:49,250 --> 01:03:51,320 So you use this to find those regions. 1503 01:03:51,320 --> 01:03:53,540 Here they are in these four subjects. 1504 01:03:53,540 --> 01:03:57,313 And now, you can say, let's find. 1505 01:03:57,313 --> 01:03:58,730 So you have to figure out some way 1506 01:03:58,730 --> 01:04:01,850 to say that thing corresponds to that to that to that. 1507 01:04:01,850 --> 01:04:03,440 And there's a whole bunch of math 1508 01:04:03,440 --> 01:04:05,420 that was invented to do that. 1509 01:04:05,420 --> 01:04:08,480 You can basically see it with your eyeballs 1510 01:04:08,480 --> 01:04:11,510 that those guys roughly correspond and those guys 1511 01:04:11,510 --> 01:04:13,220 roughly correspond. 1512 01:04:13,220 --> 01:04:15,480 The math is just a way to do that. 1513 01:04:15,480 --> 01:04:17,225 And then once you've found that region, 1514 01:04:17,225 --> 01:04:19,850 you can measure its response in a whole bunch of new conditions 1515 01:04:19,850 --> 01:04:22,820 and ask what it does. 1516 01:04:22,820 --> 01:04:27,680 And in particular, so this is different from a group analysis 1517 01:04:27,680 --> 01:04:29,810 where you don't identify those regions. 1518 01:04:29,810 --> 01:04:34,070 You just choose regions anatomically 1519 01:04:34,070 --> 01:04:37,760 So if we just align them and said, OK, that's a region, 1520 01:04:37,760 --> 01:04:39,980 well, we don't have much of the language stuff there, 1521 01:04:39,980 --> 01:04:42,560 not much there, a lot there, not much there. 1522 01:04:42,560 --> 01:04:44,360 OK, that's not great. 1523 01:04:44,360 --> 01:04:46,640 Then we take another one and we define this. 1524 01:04:46,640 --> 01:04:47,420 This is a problem. 1525 01:04:47,420 --> 01:04:49,160 No language stuff here, lots of language 1526 01:04:49,160 --> 01:04:50,750 stuff there, none and lots. 1527 01:04:50,750 --> 01:04:52,880 Not good. 1528 01:04:52,880 --> 01:04:54,560 Everybody see how that's a problem? 1529 01:04:54,560 --> 01:04:56,960 OK, I guess I'm flogging this. 1530 01:04:56,960 --> 01:04:58,820 We can move on now. 1531 01:04:58,820 --> 01:05:02,030 But the main problems with the group analysis 1532 01:05:02,030 --> 01:05:05,360 are you might fail to detect neural activity that's actually 1533 01:05:05,360 --> 01:05:07,880 there, because it doesn't align well enough across subjects 1534 01:05:07,880 --> 01:05:09,560 and so it doesn't reach threshold. 1535 01:05:09,560 --> 01:05:11,450 It's not consistent. 1536 01:05:11,450 --> 01:05:13,910 But for present purposes, the more relevant problem 1537 01:05:13,910 --> 01:05:17,930 is, you might fail to distinguish between two 1538 01:05:17,930 --> 01:05:19,790 different functions, because they 1539 01:05:19,790 --> 01:05:24,830 invariably coexist within that region or not. 1540 01:05:24,830 --> 01:05:27,290 So we're not doing that for present purposes. 1541 01:05:27,290 --> 01:05:29,150 Instead, we're going to now go back 1542 01:05:29,150 --> 01:05:32,720 to the conundrum of why do the patient studies suggest 1543 01:05:32,720 --> 01:05:35,180 that language is distinct from the rest of thought, 1544 01:05:35,180 --> 01:05:37,130 but the past functional MRI studies 1545 01:05:37,130 --> 01:05:39,320 suggest that language overlaps with other functions 1546 01:05:39,320 --> 01:05:40,640 in the brain. 1547 01:05:40,640 --> 01:05:43,040 And we're going to consider the hypothesis 1548 01:05:43,040 --> 01:05:44,960 that if you study individual brains 1549 01:05:44,960 --> 01:05:48,680 and localize those regions individually in each subject, 1550 01:05:48,680 --> 01:05:51,950 then the story might be different. 1551 01:05:51,950 --> 01:05:53,030 And it is. 1552 01:05:53,030 --> 01:05:55,310 So here's the task that Fedorenko and I 1553 01:05:55,310 --> 01:05:56,610 did a few years ago. 1554 01:05:56,610 --> 01:05:58,557 We came up with seven different tasks. 1555 01:05:58,557 --> 01:06:00,140 I won't bore you with all the details. 1556 01:06:00,140 --> 01:06:01,500 It doesn't really matter. 1557 01:06:01,500 --> 01:06:04,610 We just had lots of stuff, arithmetic, spatial working 1558 01:06:04,610 --> 01:06:07,010 memory, various cognitive control tasks, 1559 01:06:07,010 --> 01:06:09,560 working memory tests, all kinds of stuff, 1560 01:06:09,560 --> 01:06:12,560 focusing on things that-- music, focusing on stuff 1561 01:06:12,560 --> 01:06:14,330 that other people had said overlaps 1562 01:06:14,330 --> 01:06:17,193 with language in the brain. 1563 01:06:17,193 --> 01:06:18,860 And so first thing is you've got to make 1564 01:06:18,860 --> 01:06:21,520 sure those other tasks actually produce activations, 1565 01:06:21,520 --> 01:06:24,020 because it's easy to make up a task and have it not do much, 1566 01:06:24,020 --> 01:06:26,010 and then that's not very interesting. 1567 01:06:26,010 --> 01:06:29,450 So, yes, each one of those tasks produce lots of activation. 1568 01:06:29,450 --> 01:06:31,040 Look at all that red stuff. 1569 01:06:31,040 --> 01:06:34,580 Looks like a bunch of pizzas. 1570 01:06:34,580 --> 01:06:36,170 So they produce good activations. 1571 01:06:36,170 --> 01:06:38,342 Now the question is, do those activations overlap 1572 01:06:38,342 --> 01:06:39,425 with the language regions. 1573 01:06:42,720 --> 01:06:45,130 So let's consider two of them. 1574 01:06:45,130 --> 01:06:46,710 This is basically Wernicke's area 1575 01:06:46,710 --> 01:06:49,800 and Broca's area, two well-known language regions, 1576 01:06:49,800 --> 01:06:52,420 identified individually in each subject 1577 01:06:52,420 --> 01:06:55,800 and now averaging the response over all the conditions. 1578 01:06:55,800 --> 01:06:57,300 Here's a response when subjects read 1579 01:06:57,300 --> 01:07:01,680 sentences and non-word strings, sentences and non-word strings. 1580 01:07:01,680 --> 01:07:03,550 That's how we define those regions, 1581 01:07:03,550 --> 01:07:05,302 but this is in data that wasn't actually 1582 01:07:05,302 --> 01:07:06,510 used to define those regions. 1583 01:07:06,510 --> 01:07:10,198 We held out some data and just cross-validated it. 1584 01:07:10,198 --> 01:07:12,240 Now the question is, how do those regions respond 1585 01:07:12,240 --> 01:07:15,180 to all of these other things? 1586 01:07:15,180 --> 01:07:18,480 They don't, pretty much at all. 1587 01:07:21,600 --> 01:07:23,190 So notice what's happened here. 1588 01:07:23,190 --> 01:07:26,220 The prior literature shows massive overlap 1589 01:07:26,220 --> 01:07:28,680 between language and all these other things. 1590 01:07:28,680 --> 01:07:32,100 In our data, when you identify those language regions 1591 01:07:32,100 --> 01:07:34,450 in each subject individually and measure 1592 01:07:34,450 --> 01:07:36,450 the magnitude of response in those other things, 1593 01:07:36,450 --> 01:07:38,470 they don't respond. 1594 01:07:38,470 --> 01:07:41,280 So this shows stunning specificity of the language 1595 01:07:41,280 --> 01:07:43,530 regions consistent with the picture that 1596 01:07:43,530 --> 01:07:45,150 comes from the patient literature, 1597 01:07:45,150 --> 01:07:46,950 from studies of brain damage. 1598 01:07:46,950 --> 01:07:48,750 Language really is separate in the brain 1599 01:07:48,750 --> 01:07:52,310 from all of these things Everybody get that picture? 1600 01:07:52,310 --> 01:07:54,060 And the reason the literature had it wrong 1601 01:07:54,060 --> 01:07:56,760 is they were mushing brains together and blurring 1602 01:07:56,760 --> 01:08:01,140 the hell out of their data and drawing wrong conclusions. 1603 01:08:01,140 --> 01:08:04,470 I'm speeding up because I don't want to run out of time. 1604 01:08:04,470 --> 01:08:06,930 So we started with these questions here. 1605 01:08:06,930 --> 01:08:08,880 Is language distinct from the rest of thought? 1606 01:08:08,880 --> 01:08:12,240 I'm saying, yes, language may be necessary to learn to think. 1607 01:08:12,240 --> 01:08:13,960 And it is indeed. 1608 01:08:13,960 --> 01:08:17,550 But the evidence from the neurological patients 1609 01:08:17,550 --> 01:08:18,810 is pretty powerful. 1610 01:08:18,810 --> 01:08:21,000 Global aphasics with pretty much no language 1611 01:08:21,000 --> 01:08:24,250 can think in myriad, sophisticated ways. 1612 01:08:24,250 --> 01:08:27,060 And when you do your functional MRI studies right, 1613 01:08:27,060 --> 01:08:29,100 you find that the language regions in the brain, 1614 01:08:29,100 --> 01:08:34,210 in fact, are not active during non-linguistic thinking. 1615 01:08:34,210 --> 01:08:36,180 Make sense? 1616 01:08:36,180 --> 01:08:38,270 Questions? 1617 01:08:38,270 --> 01:08:40,840 Wow, I finished on time.