1 00:00:00,500 --> 00:00:04,720 Human language can be complex and bewildering. 2 00:00:04,720 --> 00:00:06,160 PHONE RINGS 3 00:00:06,160 --> 00:00:09,200 Oh, dear. Sorry, I've got to take this. 4 00:00:10,560 --> 00:00:13,360 Hello? I can't talk now. 5 00:00:13,360 --> 00:00:17,080 I'm doing the Christmas lectures! 6 00:00:17,080 --> 00:00:18,400 What? 7 00:00:19,720 --> 00:00:21,920 She said DAVID didn't take his money? 8 00:00:23,960 --> 00:00:26,480 What? She said David didn't take HIS money? 9 00:00:27,640 --> 00:00:31,880 Oh, she SAID David didn't take his money. 10 00:00:31,880 --> 00:00:33,600 Why didn't you just say that, then? 11 00:00:33,600 --> 00:00:37,560 Sorry. Now, what you have there is three very different meanings from 12 00:00:37,560 --> 00:00:40,600 exactly the same sentence. 13 00:00:40,600 --> 00:00:44,960 Will anything other than another human being ever be able to cope 14 00:00:44,960 --> 00:00:46,360 with that level of complexity? 15 00:00:47,800 --> 00:00:51,280 In this lecture, I'm going to find out what makes language the ultimate 16 00:00:51,280 --> 00:00:55,960 communication tool and why humans are absolute masters of it. 17 00:01:20,160 --> 00:01:23,360 Welcome to the third Royal Institution Christmas lecture 18 00:01:23,360 --> 00:01:24,760 of 2017. 19 00:01:24,760 --> 00:01:26,800 I'm Professor Sophie Scott. 20 00:01:26,800 --> 00:01:31,480 Now humans have got an incredibly powerful ability - language. 21 00:01:31,480 --> 00:01:36,040 I can convey very precise meanings to anyone within earshot if they 22 00:01:36,040 --> 00:01:41,000 speak my language. To give you a taste, please let me introduce scientist, 23 00:01:41,000 --> 00:01:43,800 comedian and rapper, Alex Lethbridge. 24 00:01:55,000 --> 00:01:57,280 So I've listened to Doc Brown, 25 00:01:57,280 --> 00:02:00,880 Akala and Syntax, they showed me how to flow off my grammar and syntax. 26 00:02:00,880 --> 00:02:03,120 The RI told me, Alex, what's your language? 27 00:02:03,120 --> 00:02:05,720 I checked my head, do you mean English, Fante or Spanish? 28 00:02:05,720 --> 00:02:08,080 Now, my PhD's crazy, I'll do wordplay till it pays me. 29 00:02:08,080 --> 00:02:09,880 And when I get bored, a language or two. 30 00:02:09,880 --> 00:02:11,680 So while you're getting PTSD from your GCSEs 31 00:02:11,680 --> 00:02:13,600 and wondering should I RSVP to GCHQ? 32 00:02:13,600 --> 00:02:16,560 Now I'm not sure, Sophie says language is complex. 33 00:02:16,560 --> 00:02:18,960 You've got the rules like subjects, verbs and objects. 34 00:02:18,960 --> 00:02:21,400 It's more than words, you've got intonation and context. 35 00:02:21,400 --> 00:02:24,040 Final lecture, we're learning all of these concepts. 36 00:02:24,040 --> 00:02:25,360 APPLAUSE 37 00:02:32,520 --> 00:02:36,920 I don't know about you, when I'm listening to rap music, I like to count all the words. 38 00:02:36,920 --> 00:02:39,280 And I reckon in about 25 seconds there, Alex, 39 00:02:39,280 --> 00:02:41,320 you said about 110 words. 40 00:02:41,320 --> 00:02:45,440 Yeah, exactly. And got over about 15 ideas, does that sound about right? 41 00:02:45,440 --> 00:02:47,880 Amazing. Thank you very much, Alex. 42 00:02:47,880 --> 00:02:49,280 No worries, thanks. 43 00:02:54,600 --> 00:02:59,040 We're used to thinking of telepathy as a science-fiction concept, but 44 00:02:59,040 --> 00:03:01,760 Alex just achieved the exact same result. 45 00:03:01,760 --> 00:03:04,720 We share the content of our minds, our brains, 46 00:03:04,720 --> 00:03:08,080 whenever we want to speak, or rap, to anyone. 47 00:03:08,080 --> 00:03:10,600 Don't worry, I'm not going to rap! 48 00:03:10,600 --> 00:03:15,320 Are we unique in having these skills or will we one day have a full 49 00:03:15,320 --> 00:03:17,360 conversation with another species? 50 00:03:17,360 --> 00:03:20,920 When I was a little girl, I so wanted to be able to talk to animals. 51 00:03:20,920 --> 00:03:22,800 Will that ever happen? 52 00:03:22,800 --> 00:03:27,040 And will computers ever be able to fully get their processors around our 53 00:03:27,040 --> 00:03:30,080 language well enough to understand a joke? 54 00:03:30,080 --> 00:03:34,120 Tonight, I'm going to explore what makes language so amazing 55 00:03:34,120 --> 00:03:36,400 and so very difficult. 56 00:03:37,520 --> 00:03:40,520 But what do we mean when we talk about language? 57 00:03:40,520 --> 00:03:42,160 Languages can come in many forms. 58 00:03:42,160 --> 00:03:45,320 We can talk, we can write, we can sign. 59 00:03:45,320 --> 00:03:48,160 And I've got a very basic form of language here. 60 00:03:48,160 --> 00:03:50,560 MORSE BEEPING 61 00:03:56,840 --> 00:03:58,720 Anybody speak Morse? 62 00:04:00,000 --> 00:04:01,800 That was a cry for help! 63 00:04:01,800 --> 00:04:05,440 Now, I don't speak Morse beyond being able to do that. 64 00:04:05,440 --> 00:04:09,440 But basically you can think of language like Morse code as being 65 00:04:09,440 --> 00:04:12,040 a message which we're sending with a code. 66 00:04:12,040 --> 00:04:13,560 And to make a code, 67 00:04:13,560 --> 00:04:18,280 the first thing you need to do is to produce a signal that's got some 68 00:04:18,280 --> 00:04:21,200 kind of structure. Now, that means a signal 69 00:04:21,200 --> 00:04:25,280 that's not just a random stream of noises without any order, 70 00:04:25,280 --> 00:04:29,600 nor can it be a very simple pattern just repeated again and again. 71 00:04:29,600 --> 00:04:33,640 You need to have a capacity to send information like the short and long 72 00:04:33,640 --> 00:04:35,320 patterns of the Morse. 73 00:04:35,320 --> 00:04:37,560 Now, humans do this when we speak aloud. 74 00:04:37,560 --> 00:04:40,960 We're using the sounds of our voices that we use in our language to 75 00:04:40,960 --> 00:04:43,000 express a code. 76 00:04:43,000 --> 00:04:47,360 Can we find any signs of a similar kind of structure in other animals' 77 00:04:47,360 --> 00:04:48,640 voices? And if so, 78 00:04:48,640 --> 00:04:51,480 could we crack their code and have a proper conversation with them? 79 00:04:52,680 --> 00:04:57,120 There are some animals who are very good candidates for being able to 80 00:04:57,120 --> 00:05:00,040 produce these sorts of sounds with structured elements, 81 00:05:00,040 --> 00:05:01,040 and those are birds. 82 00:05:03,680 --> 00:05:08,920 Now, these guys, who are a couple of zebra finches, 83 00:05:08,920 --> 00:05:12,000 and a couple of canaries, they're songbirds. 84 00:05:13,680 --> 00:05:15,720 Songbirds, when they're babies, 85 00:05:15,720 --> 00:05:19,640 they learn all the songs they're going to sing when they're adults. 86 00:05:19,640 --> 00:05:24,040 The most impressive can learn over 1,000 different songs. 87 00:05:24,040 --> 00:05:27,480 Could these songs contain coded information? 88 00:05:27,480 --> 00:05:30,400 Now, I can hear a couple of cheeps coming out of here, 89 00:05:30,400 --> 00:05:32,400 but I think we have a recording of one of the canaries. 90 00:05:32,400 --> 00:05:33,720 Can we listen to that? 91 00:05:33,720 --> 00:05:36,560 CANARY CHIRPS 92 00:05:44,600 --> 00:05:46,240 It's a beautiful sound. 93 00:05:46,240 --> 00:05:50,200 But does it contain enough structure in that signal that it could be used 94 00:05:50,200 --> 00:05:51,280 to transmit a code? 95 00:05:52,600 --> 00:05:55,880 Well, I've got an example of the canary's song here. 96 00:05:55,880 --> 00:06:00,720 And what I'm showing it to you as, is what's called a spectrogram. 97 00:06:00,720 --> 00:06:05,280 Now, a spectrogram is a way of looking at the structure in a sound. 98 00:06:05,280 --> 00:06:09,120 So, what you have along this direction is time. 99 00:06:09,120 --> 00:06:12,200 So, this is the sound unfurling over time, 100 00:06:12,200 --> 00:06:16,360 how it's changing over time. This direction, we've got frequency. 101 00:06:16,360 --> 00:06:18,240 And that's roughly telling you, 102 00:06:18,240 --> 00:06:20,560 low-pitched sounds up to high-pitched sounds. 103 00:06:20,560 --> 00:06:23,160 And where the colours are warmer and brighter, 104 00:06:23,160 --> 00:06:24,960 that's where there's more energy. 105 00:06:24,960 --> 00:06:28,480 And we can see in these individual elements, these little notes, 106 00:06:28,480 --> 00:06:31,520 and we're seeing some quite structured elements to this. 107 00:06:31,520 --> 00:06:34,480 We've got a similar sequence here and repeating there. 108 00:06:34,480 --> 00:06:39,200 And then these sequences of lower and higher alternating notes. 109 00:06:39,200 --> 00:06:41,920 Now, I need to compare this with another kind of voice. 110 00:06:41,920 --> 00:06:45,000 So, I would like a human volunteer, please. 111 00:06:45,000 --> 00:06:47,760 Can I have you in the middle there, with the penguin? 112 00:06:47,760 --> 00:06:49,280 Thank you very much. 113 00:06:52,760 --> 00:06:54,440 Now, what's your name? 114 00:06:54,440 --> 00:06:58,000 Ruth. Ruth. I'm going to ask you to come over here and say the first two 115 00:06:58,000 --> 00:07:00,840 lines of Humpty Dumpty into my computer. 116 00:07:00,840 --> 00:07:02,080 OK, I'll tell you when to go. 117 00:07:02,080 --> 00:07:03,800 If you could just stand about there. 118 00:07:03,800 --> 00:07:06,400 Brilliant. And go now. 119 00:07:06,400 --> 00:07:08,840 Humpty Dumpty sat on the wall. 120 00:07:08,840 --> 00:07:11,520 Humpty Dumpty had a great fall. 121 00:07:11,520 --> 00:07:13,280 Brilliant. Thank you very much, Ruth. 122 00:07:13,280 --> 00:07:14,760 Exemplary, I think you'd agree. 123 00:07:21,600 --> 00:07:24,920 So, here's Ruth's version of Humpty Dumpty shown in the same way 124 00:07:24,920 --> 00:07:28,400 on a spectrogram. You can see immediately there are some differences. 125 00:07:28,400 --> 00:07:30,520 There's the canary, there's Ruth. 126 00:07:30,520 --> 00:07:32,600 You can also see some similarities. 127 00:07:32,600 --> 00:07:34,600 And I'm talking in the most general sense here, 128 00:07:34,600 --> 00:07:39,320 but Ruth is producing individual rhythm in the syllables of what she's saying, 129 00:07:39,320 --> 00:07:42,480 and you're seeing a pattern of that over the sentence. 130 00:07:42,480 --> 00:07:46,640 And you're seeing something broadly comparable in the canary. 131 00:07:46,640 --> 00:07:49,320 We're seeing structure in those sounds. 132 00:07:49,320 --> 00:07:53,320 The canary and the speech sounds have both got rhythm, pitch, 133 00:07:53,320 --> 00:07:54,960 rate information in there. 134 00:07:54,960 --> 00:07:57,760 So it's at least possible that the songbird 135 00:07:57,760 --> 00:08:00,240 is producing something which has got 136 00:08:00,240 --> 00:08:04,360 similarities to the way we code information in our speech. 137 00:08:04,360 --> 00:08:06,640 Could it actually be a code, though? 138 00:08:07,720 --> 00:08:09,440 Well, probably not. 139 00:08:09,440 --> 00:08:12,000 It doesn't seem to be quite enough. 140 00:08:12,000 --> 00:08:14,880 So if you look at how songbirds use song, 141 00:08:14,880 --> 00:08:17,520 what you find is they don't generally change their songs 142 00:08:17,520 --> 00:08:19,200 once they've learned them. 143 00:08:19,200 --> 00:08:21,480 They will always sing the same whole song. 144 00:08:21,480 --> 00:08:25,120 And the other thing they don't do is chop songs up and rearrange them 145 00:08:25,120 --> 00:08:27,920 to make new songs. We do that all the time. 146 00:08:27,920 --> 00:08:30,520 We can use words in lots of different, very novel orders. 147 00:08:30,520 --> 00:08:32,400 The birds don't do this. 148 00:08:32,400 --> 00:08:36,080 So it's entirely possible that, complex though the songbirds are, 149 00:08:36,080 --> 00:08:39,280 they are not producing something that is conveying a complex, 150 00:08:39,280 --> 00:08:40,800 coded meaning. 151 00:08:40,800 --> 00:08:42,480 But there's another group of birds 152 00:08:42,480 --> 00:08:46,360 that can learn to say human-coded signals, use words. 153 00:08:46,360 --> 00:08:48,120 And those are parrots. 154 00:08:48,120 --> 00:08:50,400 Please meet Mike and his parrot. 155 00:08:53,640 --> 00:08:56,600 Hello. Hello. 156 00:08:59,960 --> 00:09:02,720 Hi. Hi. Hi, Mike, who have you brought with you? 157 00:09:02,720 --> 00:09:04,080 So this is Helly... You all right? 158 00:09:04,080 --> 00:09:06,040 You all right? ..and Helly is an Amazon parrot. 159 00:09:06,040 --> 00:09:09,080 So a South American bird from the rainforests. 160 00:09:09,080 --> 00:09:10,960 And she's already said hello, hasn't she? 161 00:09:10,960 --> 00:09:12,960 She has, yes. She knows that's an introduction, 162 00:09:12,960 --> 00:09:15,400 so it's the first point of call for a conversation 163 00:09:15,400 --> 00:09:16,760 or attention gathering. 164 00:09:16,760 --> 00:09:19,280 How many other words does she use? You all right? 165 00:09:19,280 --> 00:09:22,280 She uses about 80 different sounds and words. 166 00:09:22,280 --> 00:09:24,000 Yeah, human words, though? 167 00:09:24,000 --> 00:09:26,320 Oh, human words, I would say she knows about five or ten. 168 00:09:26,320 --> 00:09:27,600 Yeah. Can you say bye? 169 00:09:27,600 --> 00:09:28,560 Bye! 170 00:09:31,040 --> 00:09:33,280 So she's doing a set of human words. 171 00:09:33,280 --> 00:09:37,160 Did you teach her those or did she just pick them up from seeing how they were used? 172 00:09:37,160 --> 00:09:40,160 Yeah, she understands that "Hello" is a greeting because it's a 173 00:09:40,160 --> 00:09:42,000 word that we would use on arrival. 174 00:09:42,000 --> 00:09:44,960 And she would then understand that the "Bye" is a departure word. 175 00:09:44,960 --> 00:09:47,960 So she places the words with timings, as well. 176 00:09:47,960 --> 00:09:50,720 Yeah. But throughout her life she's had lots of trainers saying, 177 00:09:50,720 --> 00:09:52,280 "Are you all right, are you all right?" 178 00:09:52,280 --> 00:09:55,240 So if she actually gets worried in something like a studio environment, 179 00:09:55,240 --> 00:09:57,200 she will say "Are you all right?" You all right? 180 00:09:57,200 --> 00:09:59,440 And she knows that goes with almost an emotion, as well. 181 00:09:59,440 --> 00:10:03,760 Yes. So she links these words with timings and emotions. 182 00:10:03,760 --> 00:10:07,440 And she can do other sounds that aren't... she can do more than words, can't she? 183 00:10:07,440 --> 00:10:10,840 Yes, she uses lots of different sounds as well as words. 184 00:10:10,840 --> 00:10:11,840 Can we have a bomb? 185 00:10:11,840 --> 00:10:14,120 WHISTLE, BOOM 186 00:10:14,120 --> 00:10:15,560 LAUGHTER 187 00:10:18,800 --> 00:10:23,240 They can also be copying other birds as well, so local birds in the wild. 188 00:10:23,240 --> 00:10:25,560 And also other birds that we house. 189 00:10:25,560 --> 00:10:27,920 Thank you very much, Mike. And thank you very much, Helly. 190 00:10:27,920 --> 00:10:30,320 Lovely meeting you. Bye-bye. 191 00:10:39,760 --> 00:10:44,080 What gives birds their amazing ability to learn to produce these sounds? 192 00:10:44,080 --> 00:10:48,800 Well, the answer may lie within their brains. 193 00:10:48,800 --> 00:10:52,920 Now, birds are more closely related to dinosaurs than they are to us. 194 00:10:52,920 --> 00:10:56,600 But we share similarities in the ways that our brains control both 195 00:10:56,600 --> 00:11:00,400 the learning and the production of sounds we make with our voices 196 00:11:00,400 --> 00:11:03,400 and the genes that build those parts of the brain. 197 00:11:03,400 --> 00:11:04,480 If we look... 198 00:11:05,880 --> 00:11:07,480 ..at a bird brain, 199 00:11:07,480 --> 00:11:10,560 we can identify specific areas which are important 200 00:11:10,560 --> 00:11:14,920 in the learning of song and in the control of singing. 201 00:11:14,920 --> 00:11:18,360 And if we look at a human brain... 202 00:11:19,960 --> 00:11:24,240 ..we can see a similarity, in that there are specific networks 203 00:11:24,240 --> 00:11:26,200 recruited when we are speaking. 204 00:11:26,200 --> 00:11:28,640 When we're talking, these are human brains here, 205 00:11:28,640 --> 00:11:32,600 this is the right side of the brain and the left side of the brain. 206 00:11:32,600 --> 00:11:36,440 We see some areas which are strongly associated with the control of all 207 00:11:36,440 --> 00:11:39,720 the work we have to do to make the sounds of speech. 208 00:11:39,720 --> 00:11:44,480 We also find the very specific area just on the left side of the brain 209 00:11:44,480 --> 00:11:47,480 which seems to be very important in planning speech. 210 00:11:47,480 --> 00:11:51,040 And it's also important in learning new things to do with our voices. 211 00:11:52,280 --> 00:11:54,840 How can I find out more about this system? 212 00:11:54,840 --> 00:11:58,640 We can take these snapshots of the brain in action and we can work with 213 00:11:58,640 --> 00:12:02,360 people who have had strokes and have damaged these brain areas. 214 00:12:02,360 --> 00:12:05,680 But there's another technique that we can use where we can investigate 215 00:12:05,680 --> 00:12:07,800 what would happen if we could turn off 216 00:12:07,800 --> 00:12:10,480 that part of the brain in someone, and just that part of the brain. 217 00:12:10,480 --> 00:12:14,160 What I'd like to do is introduce you to my colleague from UCL, 218 00:12:14,160 --> 00:12:17,360 Dr Ricci Hannah, and comedian Robin Ince. 219 00:12:20,240 --> 00:12:23,880 Hello Robin, hello. I've been waiting for this day. 220 00:12:23,880 --> 00:12:25,120 Hello, Ricci. 221 00:12:25,120 --> 00:12:26,200 Hi. Hi. 222 00:12:26,200 --> 00:12:28,360 Now, Robin. 223 00:12:28,360 --> 00:12:30,840 Robin, what we're going to do is sit you down here. 224 00:12:30,840 --> 00:12:35,280 Right. And then I'm going to let Ricci explain what we're going to do next. OK? 225 00:12:35,280 --> 00:12:38,920 I have to emphasise, this is a temporary state of affairs, OK? 226 00:12:38,920 --> 00:12:42,640 We're not about to do some sort of terrible live brain surgery on you! 227 00:12:42,640 --> 00:12:45,680 I was quite worried, because I was watching what else was going on and I thought, 228 00:12:45,680 --> 00:12:48,240 "They're going to prove that I'm less intelligent than a parrot!" 229 00:12:48,240 --> 00:12:50,000 So let's find out what happens. 230 00:12:50,000 --> 00:12:52,040 Ricci, can you tell us what you've got here? 231 00:12:52,040 --> 00:12:55,800 So this is a transcranial magnetic stimulator. 232 00:12:55,800 --> 00:12:57,760 And what we can use it for is to probe 233 00:12:57,760 --> 00:13:00,240 how different parts of the brain play a role 234 00:13:00,240 --> 00:13:02,280 in different aspects of behaviour. 235 00:13:02,280 --> 00:13:03,560 In this case, speech. 236 00:13:03,560 --> 00:13:08,960 OK. So can we start just by pointing out, that the way that we get this, 237 00:13:08,960 --> 00:13:14,240 we change Robin's brain activity by passing electrical current through 238 00:13:14,240 --> 00:13:16,120 that coil, is that right? Mm-hmm. 239 00:13:16,120 --> 00:13:19,440 And actually, according to the principles of electromagnetic forces, 240 00:13:19,440 --> 00:13:22,400 which were first described here by Michael Faraday, 241 00:13:22,400 --> 00:13:25,920 what that lets us do is induce currents inside Robin's brain 242 00:13:25,920 --> 00:13:28,760 without actually having to get inside his brain. 243 00:13:28,760 --> 00:13:30,400 It's absolutely amazing. 244 00:13:30,400 --> 00:13:37,080 So can we start by getting Robin talking and then seeing if we can stop him talking? OK! 245 00:13:38,240 --> 00:13:40,200 People have been trying this for years! 246 00:13:40,200 --> 00:13:42,880 It's going to be popular! Can I get you to shuffle back, Robin? 247 00:13:42,880 --> 00:13:44,160 Yes. There we go. 248 00:13:44,160 --> 00:13:49,120 Are you OK? OK, so I'll just position the coil. 249 00:13:50,440 --> 00:13:57,200 What I want you to do is to say the months of the year really loudly and clearly. OK. 250 00:13:57,200 --> 00:13:58,840 When you're ready. 251 00:13:58,840 --> 00:14:03,520 January, February, March, April, M... 252 00:14:03,520 --> 00:14:05,000 Please start talking again! 253 00:14:06,200 --> 00:14:08,080 That is weird! 254 00:14:08,080 --> 00:14:09,680 That is very... 255 00:14:09,680 --> 00:14:12,920 I think Professor Brian Cox who I do a radio show with is going to want a 256 00:14:12,920 --> 00:14:16,440 beret that I wear with that in there, so he can stop me talking! 257 00:14:16,440 --> 00:14:19,080 I don't know what it looked like to you, but it was like just... 258 00:14:20,960 --> 00:14:23,880 It was like kind of Homer with a doughnut, sort of... 259 00:14:25,400 --> 00:14:27,200 If you could bear it, can we try it again? 260 00:14:27,200 --> 00:14:29,240 Yeah! It's really... I find it amazing. 261 00:14:29,240 --> 00:14:31,960 It is extraordinary, isn't it? And then it just comes back. 262 00:14:31,960 --> 00:14:34,080 I think I prefer quiet me. Let's do it again! 263 00:14:35,280 --> 00:14:39,000 Shall I try, and see how far I can get in Jabberwocky, shall I try that? 264 00:14:39,000 --> 00:14:40,960 Yes, excellent, yes. 265 00:14:40,960 --> 00:14:43,000 Tell me when you want me to. When you're ready. 266 00:14:43,000 --> 00:14:45,200 'Twas brillig, and the slithy toves. Did... 267 00:14:49,240 --> 00:14:50,160 It's fantastic! 268 00:14:54,600 --> 00:14:56,200 That's... 269 00:15:01,560 --> 00:15:04,120 So what Ricci's very specifically 270 00:15:04,120 --> 00:15:06,280 focusing on here is this part of the brain, 271 00:15:06,280 --> 00:15:08,680 it's called the inferior frontal gyrus, on the left. 272 00:15:08,680 --> 00:15:10,640 And in humans it's incredibly important for 273 00:15:10,640 --> 00:15:12,840 planning and controlling speech. 274 00:15:12,840 --> 00:15:15,520 If we move to a slightly different area, 275 00:15:15,520 --> 00:15:17,360 even within the same network, 276 00:15:17,360 --> 00:15:20,600 what we find is that Robin will be able to talk absolutely fine. 277 00:15:20,600 --> 00:15:22,120 Would it be OK to try that? 278 00:15:22,120 --> 00:15:23,200 Yeah, sure. OK. 279 00:15:25,760 --> 00:15:29,320 OK. I didn't even look at the health and safety form for this! 280 00:15:29,320 --> 00:15:32,200 I sent this to you! I know, I didn't look, just in case! 281 00:15:32,200 --> 00:15:33,840 You're fine, you're safe. 282 00:15:33,840 --> 00:15:34,880 OK, when you're ready. 283 00:15:34,880 --> 00:15:36,520 'Twas brillig, and the slithy toves 284 00:15:36,520 --> 00:15:37,880 Did gyre and gimble in the wabe 285 00:15:37,880 --> 00:15:40,360 All mimsy were the borogoves And the mome... Yeah, that's... 286 00:15:40,360 --> 00:15:41,560 There you go, there you go. 287 00:15:41,560 --> 00:15:43,760 I think I preferred the one at the side! 288 00:15:43,760 --> 00:15:47,320 I can't thank you enough for being prepared to come out in front of 289 00:15:47,320 --> 00:15:49,280 everybody and have us try and zap your brain! 290 00:15:49,280 --> 00:15:51,400 Thank you very much, Robin. Thank you very much, Ricci. 291 00:15:51,400 --> 00:15:52,960 That was amazing, thank you. 292 00:15:54,040 --> 00:15:55,800 It was brilliant, thank you. 293 00:15:59,000 --> 00:16:00,800 Thank you. 294 00:16:02,680 --> 00:16:07,880 So you could see how precise the effect of the transcranial magnetic 295 00:16:07,880 --> 00:16:12,640 stimulation was. We were only seeing Robin stopping talking when we were 296 00:16:12,640 --> 00:16:16,360 applying the TMS over his left inferior frontal gyrus. 297 00:16:16,360 --> 00:16:19,400 When we went elsewhere in the brain, other things happened, 298 00:16:19,400 --> 00:16:22,200 but it's not stopping him from talking. 299 00:16:22,200 --> 00:16:26,080 So what we're seeing here in the humans and in the birds 300 00:16:26,080 --> 00:16:28,000 are very dedicated brain regions 301 00:16:28,000 --> 00:16:31,080 that are important in vocal control and vocal learning. 302 00:16:31,080 --> 00:16:35,040 It's a strong hint that we really are seeing some commonality in 303 00:16:35,040 --> 00:16:37,240 the brain areas that are to do with the learning 304 00:16:37,240 --> 00:16:39,520 and the producing of vocal sounds. 305 00:16:39,520 --> 00:16:44,320 But can any of these birds really understand the words they're saying? 306 00:16:44,320 --> 00:16:46,360 Well, parrots don't just say human words. 307 00:16:46,360 --> 00:16:48,960 They'll mimic pretty much anything they hear a lot of. 308 00:16:48,960 --> 00:16:52,240 Car alarms, creaking doors, alarm clocks. 309 00:16:52,240 --> 00:16:54,000 Why do they do this at all? 310 00:16:54,000 --> 00:16:56,680 Well, birds are mimicking to show off to potential mates, 311 00:16:56,680 --> 00:16:57,960 to get attention. 312 00:16:57,960 --> 00:17:00,160 To impress other birds and other humans. 313 00:17:00,160 --> 00:17:02,000 To defend their nesting sites. 314 00:17:02,000 --> 00:17:05,080 Perhaps the more impressive a sound they can make, 315 00:17:05,080 --> 00:17:06,840 the more likely they'll find a companion, 316 00:17:06,840 --> 00:17:08,680 or scare off a rival. 317 00:17:08,680 --> 00:17:13,280 So birds aren't showing a great ability to decode our language. 318 00:17:13,280 --> 00:17:16,040 But what do I mean by decoding? 319 00:17:16,040 --> 00:17:23,320 We humans are exceptionally good at working out what words are and what words mean. 320 00:17:23,320 --> 00:17:26,560 I'd like you to watch and listen to a clip that I'm about to play you, 321 00:17:26,560 --> 00:17:29,000 and there's going to be a short test afterwards. 322 00:17:29,000 --> 00:17:31,560 And this will go down on your permanent record. 323 00:17:33,040 --> 00:17:34,560 Back. Deaf. Gidge. 324 00:17:34,560 --> 00:17:36,040 Hock. Fip. Nop. 325 00:17:36,040 --> 00:17:37,560 Fib. Wreck. Sit. 326 00:17:37,560 --> 00:17:39,000 They. Fip. Fip. 327 00:17:39,000 --> 00:17:40,560 Hock. Fib. Gack. 328 00:17:40,560 --> 00:17:42,040 Gin. Hock. Hock. 329 00:17:42,040 --> 00:17:43,600 Lun. Fip. Nat. 330 00:17:43,600 --> 00:17:45,040 Fip. Ros. Hock. 331 00:17:45,040 --> 00:17:46,040 Yas. Beth. 332 00:17:47,720 --> 00:17:50,320 OK. Short test. 333 00:17:50,320 --> 00:17:51,320 What's this called? 334 00:17:52,760 --> 00:17:53,760 CROWD: Fip. 335 00:17:54,880 --> 00:17:58,000 It's a fip. What's this? 336 00:17:58,000 --> 00:18:00,040 CROWD: Hock. Hock, excellent. 337 00:18:00,040 --> 00:18:04,280 Well done. Now, I didn't tell you to work out what was going on there. 338 00:18:04,280 --> 00:18:07,920 What you were doing was decoding the information we gave you. 339 00:18:07,920 --> 00:18:11,560 There were some novel sounds in there and they were being associated 340 00:18:11,560 --> 00:18:14,280 in quite a regular way with visual information. 341 00:18:14,280 --> 00:18:17,040 Even if you don't know that your brain is trying to do it, 342 00:18:17,040 --> 00:18:20,840 you're always trying to spot words and work out what those words mean. 343 00:18:20,840 --> 00:18:25,120 We are not the only animal who has an ability to do this. 344 00:18:25,120 --> 00:18:28,240 There's an animal you're probably all very familiar with who's actually 345 00:18:28,240 --> 00:18:30,840 really very, very good at sharing this ability with us. 346 00:18:30,840 --> 00:18:31,760 And that's dogs. 347 00:18:33,360 --> 00:18:37,320 Please give us a very nice doggy-friendly round of applause 348 00:18:37,320 --> 00:18:39,400 for Gable and his owner, Sally. 349 00:18:42,600 --> 00:18:45,040 Hello. Hello. Hello. 350 00:18:45,040 --> 00:18:48,800 Hello. Hey, Gable. 351 00:18:49,920 --> 00:18:52,040 So Sally, what's different about Gable? 352 00:18:53,480 --> 00:19:01,240 Gable's got the ability to identify a large number of objects and toys by name. 353 00:19:01,240 --> 00:19:05,240 He currently knows about 150 different names for toys and objects and articles. 354 00:19:05,240 --> 00:19:06,840 Goodness. 355 00:19:06,840 --> 00:19:09,560 Now, can we see a demonstration of that? 356 00:19:09,560 --> 00:19:12,800 So we've got some of Gable's toys here and what were going to do is 357 00:19:12,800 --> 00:19:15,240 get a random selection of those out and then see. 358 00:19:15,240 --> 00:19:17,640 OK. I think there is somebody... 359 00:19:17,640 --> 00:19:20,680 Hi, Sean. Now, we've spent a little bit of time with Sean, 360 00:19:20,680 --> 00:19:23,640 getting Sean used to being with Gable and Gable used to being with Sean. 361 00:19:23,640 --> 00:19:25,760 Can I bring you down? Thank you. 362 00:19:28,240 --> 00:19:31,440 Now Sean, if you could come over here. 363 00:19:31,440 --> 00:19:36,080 Can you pop these gloves on and can you just randomly select 15 toys 364 00:19:36,080 --> 00:19:37,880 from these buckets and spread them out? 365 00:19:37,880 --> 00:19:41,200 When did you first realise Gable could do this? 366 00:19:41,200 --> 00:19:42,720 When he was a young puppy, actually. 367 00:19:42,720 --> 00:19:44,800 He sort of invented the game. 368 00:19:44,800 --> 00:19:47,280 I was trying to watch telly one evening and he kept pestering me. 369 00:19:47,280 --> 00:19:48,840 He wanted to do something. 370 00:19:48,840 --> 00:19:52,360 And I just remembered oh, his red toy was upstairs. 371 00:19:52,360 --> 00:19:54,240 And I just said, "Oh, go and get your red toy." 372 00:19:54,240 --> 00:19:55,880 Sort of idly just dismissing him, 373 00:19:55,880 --> 00:19:58,360 hoping he'd be gone for ages because he couldn't find it. 374 00:19:58,360 --> 00:20:01,440 And he came back with it and he put it in front of me and looked at me. 375 00:20:01,440 --> 00:20:04,920 And I just thought, "I've actually never told you that's called Red Toy." 376 00:20:04,920 --> 00:20:07,000 And so I just then thought, I wonder what happens 377 00:20:07,000 --> 00:20:09,400 if I do teach him a name, and it's gone from there, really. 378 00:20:10,800 --> 00:20:13,400 Now... Wow, look at all your toys. ..Sean, don't go anywhere, 379 00:20:13,400 --> 00:20:15,360 I'm going to need you again. 380 00:20:15,360 --> 00:20:19,680 Can I ask you, Sally, to tell Gable to pick one of these toys, please? 381 00:20:19,680 --> 00:20:21,240 OK. Gable... 382 00:20:23,440 --> 00:20:24,760 triceratops. 383 00:20:26,800 --> 00:20:28,560 Get triceratops. 384 00:20:32,800 --> 00:20:36,640 Yes! Yes, good boy! Good boy! 385 00:20:36,640 --> 00:20:38,000 Good boy! Leave it. 386 00:20:39,880 --> 00:20:41,720 Come round here. Fantastic. 387 00:20:41,720 --> 00:20:45,800 So we've seen that Gable is really very good at working out what you're 388 00:20:45,800 --> 00:20:49,240 saying and trying to work out from looking at you which one you mean. 389 00:20:49,240 --> 00:20:51,480 But then Gable is your dog and he might be really, 390 00:20:51,480 --> 00:20:53,200 really familiar with your voice. 391 00:20:53,200 --> 00:20:55,560 It would be very interesting to know if we could see this happen with 392 00:20:55,560 --> 00:20:57,360 someone who's got a very different voice. 393 00:20:57,360 --> 00:21:00,080 So Sean, is it OK if Sean has a go? 394 00:21:00,080 --> 00:21:04,280 Yes. And what you need to do is whisper into Sean's ear 395 00:21:04,280 --> 00:21:06,360 which toy you'd like him to pick up. 396 00:21:06,360 --> 00:21:08,360 Do you remember Sean? 397 00:21:08,360 --> 00:21:12,000 OK. Gable, octopus. 398 00:21:12,000 --> 00:21:13,160 Get octopus. 399 00:21:13,160 --> 00:21:14,240 Go. 400 00:21:15,240 --> 00:21:16,160 Yes! 401 00:21:17,520 --> 00:21:19,280 Good boy, good boy! 402 00:21:19,280 --> 00:21:21,160 And well done, Sean. Leave it. 403 00:21:22,600 --> 00:21:24,160 Can Sean do one more? Of course, yes. 404 00:21:24,160 --> 00:21:26,000 Is that OK, Sean? 405 00:21:26,000 --> 00:21:28,000 Gable, get hammer. 406 00:21:28,000 --> 00:21:29,040 Get hammer. 407 00:21:31,760 --> 00:21:34,760 Yes! Amazing. 408 00:21:34,760 --> 00:21:37,000 Good boy! 409 00:21:38,160 --> 00:21:41,640 So this really does indicate that Gable must have some understanding of what 410 00:21:41,640 --> 00:21:45,120 words mean above and beyond just associating your voice with things. 411 00:21:45,120 --> 00:21:49,400 That's fantastic. Thank you so much, Sean, thank you so much, Sally, 412 00:21:49,400 --> 00:21:51,360 and particularly thank you so much, Gable. 413 00:21:51,360 --> 00:21:52,320 Come on, then! 414 00:21:56,360 --> 00:21:58,360 That's amazing. 415 00:21:58,360 --> 00:22:04,440 So, what's happening in Gable's brain so that he can do this? 416 00:22:04,440 --> 00:22:07,280 Well, there's a group of scientists in Hungary who have been doing some 417 00:22:07,280 --> 00:22:09,520 quite extraordinary experiments. 418 00:22:09,520 --> 00:22:11,840 They've been training dogs to lie very still 419 00:22:11,840 --> 00:22:14,680 and then they've been putting them into brain scanners. 420 00:22:16,720 --> 00:22:21,680 The brain scanners don't work if you move so the dog has to stay very still. 421 00:22:22,960 --> 00:22:26,120 And then what they are doing is taking pictures of the activity inside the 422 00:22:26,120 --> 00:22:30,840 dog's brains while they're listening to different sounds and words. 423 00:22:30,840 --> 00:22:35,960 When I scan people, they're not normally that happy! 424 00:22:35,960 --> 00:22:39,480 So this is called functional magnetic resonance imaging and it lets us 425 00:22:39,480 --> 00:22:41,440 take photographs of the brain in action. 426 00:22:43,440 --> 00:22:45,000 And this is showing the results. 427 00:22:45,000 --> 00:22:47,280 So a dog's brain, as you can see, 428 00:22:47,280 --> 00:22:50,360 it's different in shape to a human brain, but it has some of the same 429 00:22:50,360 --> 00:22:54,640 structures in there. And quite strikingly, 430 00:22:54,640 --> 00:22:59,080 one of the things they are finding in the results is when dogs hear words they understand, 431 00:22:59,080 --> 00:23:02,360 we see greater activation in the left side of the brain, 432 00:23:02,360 --> 00:23:05,360 particularly in brain areas to do with processing sound. 433 00:23:05,360 --> 00:23:10,600 And that is very similar to what you find in our brains. 434 00:23:10,600 --> 00:23:15,080 So it looks like it's at least possible that more than any other animal 435 00:23:15,080 --> 00:23:19,240 we've looked at, dogs might be sharing some of our ability 436 00:23:19,240 --> 00:23:24,240 to decode spoken words and they may even be doing it in similar ways to us. 437 00:23:25,320 --> 00:23:30,760 But of course, amazing as they are, dog's brains do have their limits. 438 00:23:30,760 --> 00:23:34,840 The largest number of words that any dog has been found to understand, 439 00:23:34,840 --> 00:23:38,280 and this was an exceptional dog, is about 1,000 words. 440 00:23:38,280 --> 00:23:39,680 Which sounds fantastic. 441 00:23:39,680 --> 00:23:42,240 Gable knows about 150 words, 442 00:23:42,240 --> 00:23:47,000 but every one of you could understand 1,000 words when you were three years old. 443 00:23:47,000 --> 00:23:50,400 And of course human language is more than just single words. 444 00:23:50,400 --> 00:23:55,400 We don't walk around going, "Steps, camera, man". 445 00:23:55,400 --> 00:23:57,800 We are putting words together into sequences. 446 00:23:57,800 --> 00:24:00,640 And when we put words into sequences, 447 00:24:00,640 --> 00:24:05,080 we actually add in an extra level of coded meaning. 448 00:24:06,720 --> 00:24:11,080 So perhaps, if we want to look for some more humanlike ability to put 449 00:24:11,080 --> 00:24:14,480 symbols into sequences like we do with sentences, 450 00:24:14,480 --> 00:24:17,480 we should be looking closer in the evolutionary tree. 451 00:24:17,480 --> 00:24:21,880 Perhaps we should be looking to our closest cousins, other apes. 452 00:24:21,880 --> 00:24:24,600 Now, because I'm not David Attenborough, 453 00:24:24,600 --> 00:24:26,200 I don't get to play with the chimpanzee. 454 00:24:27,760 --> 00:24:30,280 But we have the next best thing. 455 00:24:30,280 --> 00:24:32,760 Please release the apes. 456 00:24:32,760 --> 00:24:36,400 THEY IMITATE MONKEYS 457 00:24:50,800 --> 00:24:52,920 This didn't happen to David Attenborough! 458 00:24:55,040 --> 00:24:56,240 That's just brilliant. 459 00:24:57,480 --> 00:24:59,320 These are human apes. 460 00:25:01,520 --> 00:25:03,800 Please welcome Neil and Ace. 461 00:25:03,800 --> 00:25:07,000 So you guys have been working a lot with our ape cousins? 462 00:25:07,000 --> 00:25:10,080 We have, yeah. We were actually very fortunate to work with Andy Serkis 463 00:25:10,080 --> 00:25:12,800 on Planet Of The Apes Last Frontier, which is an interactive movie. 464 00:25:12,800 --> 00:25:14,600 Hold on a second, sorry. 465 00:25:14,600 --> 00:25:18,360 OK, just reassurance. 466 00:25:18,360 --> 00:25:21,240 We were very lucky, we got to study apes for quite a while before we 467 00:25:21,240 --> 00:25:24,400 started creating our own characters from ape work. 468 00:25:24,400 --> 00:25:27,800 We watched their movements, their behaviour patterns, hierarchies - 469 00:25:27,800 --> 00:25:30,760 here's some of the work we did - and also the speech they use, 470 00:25:30,760 --> 00:25:33,760 which is five different parts of language. 471 00:25:33,760 --> 00:25:37,760 Grunts, barks, hoots, whimpers and screams, which they do use as chimps. 472 00:25:37,760 --> 00:25:40,120 And this is my ape actually, Bryn. 473 00:25:40,120 --> 00:25:42,120 That's you? Yes, that's me. 474 00:25:42,120 --> 00:25:43,840 That's amazing. Yes. 475 00:25:43,840 --> 00:25:47,280 So what this is letting you do is have ape actors which are based on humans, 476 00:25:47,280 --> 00:25:50,080 so you can get them to do things which you could never actually ask 477 00:25:50,080 --> 00:25:51,800 another ape to do? Absolutely, 478 00:25:51,800 --> 00:25:53,880 and we get to create a bit of drama in the family. 479 00:25:53,880 --> 00:25:56,560 It's actually really about family, this interactive film, 480 00:25:56,560 --> 00:25:59,400 so it is centred on this particular family of apes that have splintered 481 00:25:59,400 --> 00:26:00,640 away from Caesar's group. 482 00:26:00,640 --> 00:26:03,400 So you've had to work really closely with the apes to actually learn 483 00:26:03,400 --> 00:26:05,480 about their body language and movements. 484 00:26:05,480 --> 00:26:08,880 Did you pick up on anything that felt like communication that you were looking at? 485 00:26:08,880 --> 00:26:12,360 Yeah, we studied a lot of their gesticulation and body language, 486 00:26:12,360 --> 00:26:15,880 especially between dominant and subservient male apes. 487 00:26:15,880 --> 00:26:18,760 Because what you get is a lot of different behaviour patterns that 488 00:26:18,760 --> 00:26:21,600 have been formed by a simple sign, for instance, the touch of hands. 489 00:26:21,600 --> 00:26:23,280 If somebody is in the dominant position, 490 00:26:23,280 --> 00:26:24,920 it's quite important to be able to... 491 00:26:24,920 --> 00:26:26,160 So if you were offering, 492 00:26:26,160 --> 00:26:30,000 if you're trying to get my attention and forgiveness, for instance. Go on. 493 00:26:30,000 --> 00:26:33,800 He gives me respect and I give it back to him. 494 00:26:33,800 --> 00:26:36,200 They are communicating with gestures. They are, yes. 495 00:26:36,200 --> 00:26:37,960 Amazing. Thank you so much, thank you. 496 00:26:49,120 --> 00:26:52,280 So Neil and Ace aren't just imitating the apes very well, 497 00:26:52,280 --> 00:26:55,120 they are clearly picking up on some aspects of communication, 498 00:26:55,120 --> 00:26:57,800 but is there really some kind of 499 00:26:57,800 --> 00:27:00,400 ape conversation going on? 500 00:27:00,400 --> 00:27:03,160 And would it look at all like the way we use language? 501 00:27:04,360 --> 00:27:08,640 To find out, please welcome chimpanzee researcher from the University of St Andrews, 502 00:27:08,640 --> 00:27:11,160 Dr Cat Hobaiter. 503 00:27:13,120 --> 00:27:14,960 Hello. Lovely to meet you. 504 00:27:16,760 --> 00:27:21,040 So you've been asking some really interesting questions about ape language 505 00:27:21,040 --> 00:27:23,040 and can you tell us more about your work? 506 00:27:23,040 --> 00:27:24,240 Yes, absolutely, 507 00:27:24,240 --> 00:27:29,920 so what people have done in the past was try to teach apes our language, 508 00:27:29,920 --> 00:27:32,240 so human language or sign language, 509 00:27:32,240 --> 00:27:34,080 or they've looked at their vocalisations. 510 00:27:34,080 --> 00:27:36,720 But what we've been looking at is their gestures. 511 00:27:36,720 --> 00:27:40,440 OK. And it turns out, they've got a lot of different gestures. 512 00:27:40,440 --> 00:27:44,640 So their own natural communication contains 60 or 70 513 00:27:44,640 --> 00:27:51,160 different hand and body movements that they're using every day to ask come here, go away, 514 00:27:51,160 --> 00:27:53,240 I want that, all the little meanings. 515 00:27:53,240 --> 00:27:55,920 And how do you work out what those gestures mean? 516 00:27:55,920 --> 00:27:58,920 What we do is we look for a particular gesture 517 00:27:58,920 --> 00:28:02,400 and then we are looking for what happens next. 518 00:28:02,400 --> 00:28:07,440 And if I were to do this and you responded back to me, 519 00:28:07,440 --> 00:28:10,840 one or two cases it could be anything or you could misunderstand me or I 520 00:28:10,840 --> 00:28:15,560 could keep going. But what I'm looking for is what stops me from 521 00:28:15,560 --> 00:28:18,320 signalling. So if I'm asking you for something, 522 00:28:18,320 --> 00:28:21,640 then the thing you do that makes me happy as a signaller 523 00:28:21,640 --> 00:28:26,040 is the thing that I wanted. Yes, so you have to look at the whole context. 524 00:28:26,040 --> 00:28:30,120 Yes, so we need to look at the signaller, at the recipient, the gesture, 525 00:28:30,120 --> 00:28:33,960 then we need to look at not just one or two but hundreds of cases so we 526 00:28:33,960 --> 00:28:36,320 can see the patterns emerging in the behaviour. 527 00:28:36,320 --> 00:28:39,840 And you've been using the general public to help you with your research as 528 00:28:39,840 --> 00:28:41,320 well, haven't you? Yes, 529 00:28:41,320 --> 00:28:45,280 so we were able over a few years to look at all the other apes but there 530 00:28:45,280 --> 00:28:47,480 was one of them missing, which was us. 531 00:28:47,480 --> 00:28:51,680 And we know we have language but we don't know if we still have access 532 00:28:51,680 --> 00:28:54,680 to some of the communication that the apes also use. 533 00:28:54,680 --> 00:28:57,240 The sort of gestural stuff they are using, yes. 534 00:28:57,240 --> 00:29:00,680 Exactly. So whether or not if we showed a member of the public a 535 00:29:00,680 --> 00:29:03,240 particular gesture, could they guess what it meant. 536 00:29:03,240 --> 00:29:06,720 And what we did was put up lots of videos online and ask everyone to 537 00:29:06,720 --> 00:29:09,400 come along and sort of play the game and have a go. 538 00:29:09,400 --> 00:29:10,760 Can we have a go at that now? 539 00:29:10,760 --> 00:29:13,440 Yes, please. Fantastic, so I think we've got a clip to start with. 540 00:29:13,440 --> 00:29:17,000 We are going to watch this closely because we are going to ask you what you think is going on. 541 00:29:21,080 --> 00:29:22,920 So the gesture there was that kind of arm. 542 00:29:22,920 --> 00:29:25,000 Little arm raise from Charlotte. 543 00:29:25,000 --> 00:29:26,800 OK, and is that a baby? 544 00:29:26,800 --> 00:29:28,960 Yes, she's just a little one. 545 00:29:28,960 --> 00:29:33,920 Do we think that's A - I want that, B - move closer, or C - go away? 546 00:29:36,920 --> 00:29:38,200 ALL: B. 547 00:29:38,200 --> 00:29:41,320 Yes, you guys are all well versed in chimpanzee already. 548 00:29:41,320 --> 00:29:43,080 That was move closer. 549 00:29:43,080 --> 00:29:44,320 OK, can we try another one? 550 00:29:46,080 --> 00:29:49,240 OK, we are just highlighting this because it happens very quickly. 551 00:29:49,240 --> 00:29:52,680 Yes, this is a really subtle one so it's just that little foot movement 552 00:29:52,680 --> 00:29:55,360 as she looks back and gives a little foot wiggle to her son. 553 00:29:57,680 --> 00:29:58,840 What happens next? 554 00:30:00,920 --> 00:30:05,040 Is that play with me, climb on me or come here? 555 00:30:05,040 --> 00:30:06,280 ALL: B. 556 00:30:06,280 --> 00:30:08,840 B, yeah. It's rather sweet, that one. Very, very swift. 557 00:30:08,840 --> 00:30:10,240 Yes. Really subtle. 558 00:30:11,320 --> 00:30:13,000 So this is really fascinating. 559 00:30:13,000 --> 00:30:15,800 It does look like you are seeing a sign language, 560 00:30:15,800 --> 00:30:18,680 you're seeing really gestural use of communication. 561 00:30:18,680 --> 00:30:21,080 They're not just waving their arms around passionately, 562 00:30:21,080 --> 00:30:24,200 there is a very precise meaning being conveyed here. 563 00:30:24,200 --> 00:30:28,800 But of course the real power of human language is that we can combine our 564 00:30:28,800 --> 00:30:34,120 words and symbols into sequences and that gives us a more complex level of meaning. 565 00:30:34,120 --> 00:30:35,760 Can the chimps ever do this? 566 00:30:35,760 --> 00:30:40,000 Do you ever see any kind of examples of sequences? 567 00:30:40,000 --> 00:30:43,680 Definitely. So what we see sometimes are these one-off gestures but they 568 00:30:43,680 --> 00:30:45,800 will also put gestures together. 569 00:30:45,800 --> 00:30:47,600 Sometimes a couple at once, 570 00:30:47,600 --> 00:30:52,360 that was a big scratch and shaking the object and he's actually waiting 571 00:30:52,360 --> 00:30:54,320 there. And whatever that gesture was, 572 00:30:54,320 --> 00:30:57,800 it doesn't seem to have done the trick because Rohara is ignoring him. 573 00:30:57,800 --> 00:31:01,160 He's going to do it again. He's going to do it again. And the arm went up. 574 00:31:01,160 --> 00:31:03,280 So we've got a scratch, a tree wiggle and the arm going up. 575 00:31:03,280 --> 00:31:04,800 Yes. 576 00:31:04,800 --> 00:31:08,200 And it worked. Oh, all right then, I'll get down. 577 00:31:08,200 --> 00:31:09,960 She's a bit of an old lady. 578 00:31:09,960 --> 00:31:14,720 So the tree waving is the sort of "Come here, I'm in charge". 579 00:31:14,720 --> 00:31:18,520 The scratching? Scratching usually means groom, let's groom, 580 00:31:18,520 --> 00:31:20,800 let's get together and groom. 581 00:31:20,800 --> 00:31:24,640 Our big question for us now is when they are putting these gestures into 582 00:31:24,640 --> 00:31:28,920 sequences, if it was human words the order of the sequence would make a 583 00:31:28,920 --> 00:31:32,560 difference to the meaning, and we are trying to work out now if that's the same for the chimps. 584 00:31:32,560 --> 00:31:33,880 Amazing. 585 00:31:33,880 --> 00:31:35,440 Thank you so much, Cat, thank you. 586 00:31:42,920 --> 00:31:46,080 Now, why is this so exciting? 587 00:31:46,080 --> 00:31:50,800 Well, if other apes can also combine symbols into a specific order to 588 00:31:50,800 --> 00:31:53,760 produce something with complex meaning, 589 00:31:53,760 --> 00:31:58,040 it unlocks a great deal more of the power of the language. 590 00:31:58,040 --> 00:32:02,200 Combining symbols into sequences allows us to describe the world in 591 00:32:02,200 --> 00:32:03,800 much more complex ways. 592 00:32:03,800 --> 00:32:07,800 It gives us the power to share much more complex ideas. 593 00:32:07,800 --> 00:32:12,160 It's a really useful skill but it's not in any way a simple skill. 594 00:32:13,320 --> 00:32:17,120 And to look at exactly what kind of thing the chimps are up against in 595 00:32:17,120 --> 00:32:21,560 terms of the difficulty of understanding sequences of symbols and decoding them 596 00:32:21,560 --> 00:32:25,000 accurately, I need two volunteers. 597 00:32:26,200 --> 00:32:29,440 Can I have you there in the Los Angeles... 598 00:32:29,440 --> 00:32:32,640 Can I have you in the Santa - sorry, Rudolph. 599 00:32:32,640 --> 00:32:34,880 Fantastic, thank you very much, thank you... 600 00:32:36,120 --> 00:32:37,160 You come round here. 601 00:32:38,960 --> 00:32:41,040 You come round this side. Thank you very much. 602 00:32:41,040 --> 00:32:43,040 Now, what's your name? 603 00:32:43,040 --> 00:32:45,200 Gracie. Gracie, lovely to meet you, Gracie. 604 00:32:45,200 --> 00:32:46,520 And what's your name? Ryan. 605 00:32:46,520 --> 00:32:48,200 Ryan? Yes. Fantastic. 606 00:32:48,200 --> 00:32:49,360 Now, Ryan and Gracie, 607 00:32:49,360 --> 00:32:53,600 what we need to do is put some general covers on you. 608 00:32:53,600 --> 00:32:56,480 OK, we are just going to, sort of, from top to toe, cover you up. 609 00:32:56,480 --> 00:33:00,320 Thank you. So what we are going to do is I'm going to ask you one at a 610 00:33:00,320 --> 00:33:03,560 time to read some instructions and to mix some things together. 611 00:33:03,560 --> 00:33:05,520 And, Gracie, I'm going to ask you to go first. 612 00:33:05,520 --> 00:33:08,720 So, Ryan, so that you don't see what's going on, 613 00:33:08,720 --> 00:33:12,200 I need you to pop on a blindfold and some ear defenders, OK? 614 00:33:12,200 --> 00:33:15,640 Just so that you don't give away the game. 615 00:33:15,640 --> 00:33:18,600 Right, so in a second I'm going to step forward with you 616 00:33:18,600 --> 00:33:21,040 and we are going to turn over your instructions. 617 00:33:21,040 --> 00:33:23,680 An important thing to remember, when we get to the end 618 00:33:23,680 --> 00:33:25,920 of the instructions, is you step back with me. 619 00:33:25,920 --> 00:33:27,320 Are you all right? 620 00:33:27,320 --> 00:33:28,600 OK, let's start. 621 00:33:39,200 --> 00:33:41,600 I think you might need to... You've got it, you've got it. 622 00:33:41,600 --> 00:33:43,520 Good code-cracking. 623 00:33:44,920 --> 00:33:46,040 And step back. 624 00:33:47,840 --> 00:33:49,760 To see an alarming change of colour. 625 00:33:49,760 --> 00:33:51,000 Fantastic, Gracie. 626 00:33:51,000 --> 00:33:52,480 OK, now don't go anywhere. 627 00:33:52,480 --> 00:33:54,360 We are now going to turn to Ryan. 628 00:33:54,360 --> 00:33:57,400 OK, so what we are going to do, we are going to turn over your instructions 629 00:33:57,400 --> 00:33:58,760 in just a second and I want 630 00:33:58,760 --> 00:34:01,840 you to follow them with these chemicals here. OK? 631 00:34:01,840 --> 00:34:05,960 Yes. First mix A into B. 632 00:34:05,960 --> 00:34:07,520 A into B. 633 00:34:07,520 --> 00:34:09,560 So I just put... A into B. 634 00:34:09,560 --> 00:34:11,640 So I just put them like this? 635 00:34:11,640 --> 00:34:12,600 Yes, very good. 636 00:34:19,520 --> 00:34:21,440 OK. Add C. 637 00:34:22,760 --> 00:34:23,640 And then... 638 00:34:26,200 --> 00:34:28,080 you step back. 639 00:34:28,080 --> 00:34:30,160 LAUGHTER 640 00:34:33,360 --> 00:34:35,920 Are you OK, there? 641 00:34:35,920 --> 00:34:36,840 Amazing. 642 00:34:38,400 --> 00:34:39,600 Well done. 643 00:34:40,960 --> 00:34:42,920 Thank you so much. 644 00:34:42,920 --> 00:34:45,200 Now, we had two different outcomes there. 645 00:34:45,200 --> 00:34:49,400 We had either a dramatic change of colour or an actual foam explosion, 646 00:34:49,400 --> 00:34:51,960 depending on the order in which the chemicals were mixed. 647 00:34:51,960 --> 00:34:55,600 And the only reason why Gracie and Ryan mixed them in different orders 648 00:34:55,600 --> 00:34:58,120 was just how we'd punctuated the instructions. 649 00:34:58,120 --> 00:34:59,720 So they both have the same instructions, 650 00:34:59,720 --> 00:35:02,840 and they were just understanding them differently based on where we 651 00:35:02,840 --> 00:35:06,000 put the full stops. Thank you very much, Gracie, thank you very much, Ryan. 652 00:35:10,160 --> 00:35:12,120 So you can see that the exact grammar, 653 00:35:12,120 --> 00:35:14,400 in this case the punctuation of the sentences, 654 00:35:14,400 --> 00:35:17,560 has completely changed the meaning of those sentences 655 00:35:17,560 --> 00:35:22,120 and that's why this tiny glimpse of the ability to combine gestures 656 00:35:22,120 --> 00:35:24,560 in the apes is so exciting. 657 00:35:24,560 --> 00:35:28,720 So if we think about the animals that we have talked about this evening, 658 00:35:28,720 --> 00:35:32,040 we've seen birds who are able to produce very complex sounds 659 00:35:32,040 --> 00:35:36,480 and they can mimic many other sounds, but they don't seem to understand the meaning of 660 00:35:36,480 --> 00:35:39,240 those words. We've seen dogs, 661 00:35:39,240 --> 00:35:42,720 now dogs don't talk but dogs are incredibly good at working out 662 00:35:42,720 --> 00:35:46,080 what human words mean and what they refer to. 663 00:35:46,080 --> 00:35:48,000 And we've seen the chimpanzees, 664 00:35:48,000 --> 00:35:52,840 they are both using gestures to have quite precise meanings and they are 665 00:35:52,840 --> 00:35:58,760 also using sequences of those gestures to have more sentence-like 666 00:35:58,760 --> 00:36:02,920 structures. But they are not the same as the kind of codes humans use. 667 00:36:02,920 --> 00:36:06,960 Why are there these distinctions between humans and other animals? 668 00:36:06,960 --> 00:36:12,080 Well, I think a big difference is likely to do with our brains. 669 00:36:12,080 --> 00:36:15,160 I just want you to look at these different ones here. 670 00:36:15,160 --> 00:36:18,760 So here we've got a bird brain, that's actually a chicken brain. 671 00:36:18,760 --> 00:36:20,080 That's a dog's brain. 672 00:36:21,320 --> 00:36:26,680 This is a replica chimpanzee brain, and this is a human brain. 673 00:36:27,920 --> 00:36:29,640 Now, they look different. 674 00:36:29,640 --> 00:36:34,320 One obvious difference is size, and the human brain is enormous, 675 00:36:34,320 --> 00:36:37,640 much larger than any of the other brains on the table. 676 00:36:37,640 --> 00:36:40,080 But big brains aren't everything. 677 00:36:40,080 --> 00:36:42,000 To make an analogy to computers, 678 00:36:42,000 --> 00:36:45,520 it's not perhaps just about the raw processing power of the brain, 679 00:36:45,520 --> 00:36:49,040 maybe we need to think about the operating system as well. 680 00:36:49,040 --> 00:36:52,640 And we seem to have a very efficient operating system 681 00:36:52,640 --> 00:36:55,600 for dealing with symbol-based codes. 682 00:36:57,040 --> 00:37:01,200 I always wanted the superpower of being able to talk to animals, 683 00:37:01,200 --> 00:37:06,320 but maybe our mismatching brains mean that I never will. 684 00:37:06,320 --> 00:37:11,200 Maybe there's something else in our world with which we communicate far more frequently, 685 00:37:11,200 --> 00:37:14,240 and maybe I need to stop thinking so much about how I could have 686 00:37:14,240 --> 00:37:18,240 conversations with other animals and maybe think a bit more about talking 687 00:37:18,240 --> 00:37:19,240 to machines. 688 00:37:20,800 --> 00:37:24,000 This year's big Christmas present gadgets are machines that we can talk to, 689 00:37:24,000 --> 00:37:25,600 digital assistants. 690 00:37:25,600 --> 00:37:29,320 A quarter of us now either command our phones via our voice 691 00:37:29,320 --> 00:37:31,800 or use devices like these around the home. 692 00:37:33,600 --> 00:37:38,240 Hey, computer, please tell me what the weather will be like in London tomorrow. 693 00:37:38,240 --> 00:37:42,400 There will be showers there tomorrow with a high of nine and a low of three. 694 00:37:42,400 --> 00:37:43,760 OK. 695 00:37:43,760 --> 00:37:47,920 Now that's nowhere near being anything like a human brain, but when I was doing my PhD 696 00:37:47,920 --> 00:37:50,760 on speech processing 25 years ago, 697 00:37:50,760 --> 00:37:54,320 I would have never believed that one day we would be walking around with 698 00:37:54,320 --> 00:37:57,400 phones in our pockets that we could talk to. 699 00:37:57,400 --> 00:37:59,760 It's incredibly hard to overestimate 700 00:37:59,760 --> 00:38:01,960 how quickly this field has developed. 701 00:38:01,960 --> 00:38:04,440 What's happening inside that box? 702 00:38:04,440 --> 00:38:07,200 How are computers managing to interact with us? 703 00:38:07,200 --> 00:38:10,480 And is it ever having a proper conversation with us? 704 00:38:10,480 --> 00:38:13,160 Are we ever going to have a meaningful dialogue with a computer? 705 00:38:13,160 --> 00:38:16,200 Will it ever understand the jokes we tell? 706 00:38:16,200 --> 00:38:20,240 Now this is an incredibly complex area and we could easily fill three 707 00:38:20,240 --> 00:38:23,800 lectures just on this topic, so I'm going to give you a tiny insight 708 00:38:23,800 --> 00:38:25,720 into aspects of how this works. 709 00:38:25,720 --> 00:38:27,960 But first, let's break down the question. 710 00:38:27,960 --> 00:38:31,640 What do computers need to do to be able to understand us? 711 00:38:31,640 --> 00:38:35,960 And the first thing is it needs to take the sounds that we make and 712 00:38:35,960 --> 00:38:38,280 decode that into words. 713 00:38:38,280 --> 00:38:41,880 This is called speech recognition and it's what is known in science as 714 00:38:41,880 --> 00:38:44,000 a ridiculously difficult question. 715 00:38:44,000 --> 00:38:47,880 And that's because speech is hard and speech is complex. 716 00:38:47,880 --> 00:38:51,440 I'm going to play you a sentence spoken in Estonian 717 00:38:51,440 --> 00:38:55,080 and what I want you to do is count the words you can hear. 718 00:38:55,080 --> 00:38:57,640 OK? So just listen out for words and see how many there are. 719 00:38:57,640 --> 00:39:01,040 VOICE SPEAKS IN ESTONIAN 720 00:39:08,080 --> 00:39:10,080 OK, how many words did you think were there? 721 00:39:11,240 --> 00:39:13,760 Sorry? 13, that's a good guess. 722 00:39:13,760 --> 00:39:15,520 Pink top, glasses. 723 00:39:15,520 --> 00:39:19,560 32? 32, OK, a big jump there, about twice as many. 724 00:39:19,560 --> 00:39:22,160 OK. Let's see the actual sentences. 725 00:39:22,160 --> 00:39:27,080 So we've got 22 words, so actually it's a meaningless question to ask you, 726 00:39:27,080 --> 00:39:31,480 because if you don't speak Estonian why would the words stand out to you? 727 00:39:31,480 --> 00:39:34,880 But this is what the computer is confronted with. 728 00:39:34,880 --> 00:39:39,880 It's hearing this continuous flow of sound and it's got to find some way 729 00:39:39,880 --> 00:39:42,680 of getting a toehold on what those words could be. 730 00:39:42,680 --> 00:39:44,360 Now listen to this sentence. 731 00:39:44,360 --> 00:39:47,160 What I'm going to do is record myself speaking. 732 00:39:49,600 --> 00:39:50,920 Where are the words? 733 00:39:54,080 --> 00:39:58,680 So this is the spectrogram of me saying "Where are the words" that I've just recorded. 734 00:39:58,680 --> 00:40:01,040 But where ARE the words? 735 00:40:01,040 --> 00:40:03,040 There's no individual words there at all. 736 00:40:03,040 --> 00:40:06,480 What you're getting is a sort of smear of energy and what looks like 737 00:40:06,480 --> 00:40:11,440 one thing happening at the end there is actually the "ss" sound at the end of words. 738 00:40:11,440 --> 00:40:15,440 When you hear gaps between words when someone is talking to you, 739 00:40:15,440 --> 00:40:17,960 that's because you understand those words. 740 00:40:17,960 --> 00:40:20,480 If you don't understand the words, you don't hear the gaps, 741 00:40:20,480 --> 00:40:22,920 as we saw with the Estonian. 742 00:40:22,920 --> 00:40:25,960 So how do computers deal with this? 743 00:40:25,960 --> 00:40:30,320 How do computers find the starts and ends of words when there actually 744 00:40:30,320 --> 00:40:32,800 are no physical gaps there necessarily at all? 745 00:40:32,800 --> 00:40:37,560 Well the first thing a computer does is it breaks up the incoming stream 746 00:40:37,560 --> 00:40:40,480 of sound into smaller chunks of sound 747 00:40:40,480 --> 00:40:43,520 and we've got a demonstration of this here. 748 00:40:43,520 --> 00:40:47,800 So this is an incoming sentence and I don't know what this is, 749 00:40:47,800 --> 00:40:51,120 I've just got the spectrogram to work off and I'm a computer for the 750 00:40:51,120 --> 00:40:53,080 purposes of this point. 751 00:40:53,080 --> 00:40:55,640 And what I'm going to do is split this up 752 00:40:55,640 --> 00:40:59,520 into different slices of information. 753 00:40:59,520 --> 00:41:04,840 And what I'm going to do then is take those different slices 754 00:41:04,840 --> 00:41:08,080 and go to look them up in the library that I've got, 755 00:41:08,080 --> 00:41:13,800 which will help me try and get the best estimate of what speech sound I'm probably trying to look at. 756 00:41:13,800 --> 00:41:15,920 And my library has to be very... Has to sort of 757 00:41:15,920 --> 00:41:20,680 have an idealised version of speech sounds because we all talk 758 00:41:20,680 --> 00:41:25,120 differently. So instead of trying to find an exact match for the speech sound, 759 00:41:25,120 --> 00:41:28,200 the computer is looking for almost like a best guess. 760 00:41:28,200 --> 00:41:30,920 So if we do this with the first speech sound in my sequence... 761 00:41:33,840 --> 00:41:36,480 I've got a little slice of sound here. 762 00:41:38,920 --> 00:41:42,560 Now if I look at that in isolation, what I can see 763 00:41:44,160 --> 00:41:48,120 is that there is a broad sort of smear of energy there. 764 00:41:48,120 --> 00:41:50,160 There's no bright hotspots. 765 00:41:50,160 --> 00:41:54,600 Remember when you get a more intense bit of energy in a speech sound or 766 00:41:54,600 --> 00:41:57,640 any sound in the spectrogram, you see it as a brighter colour? 767 00:41:57,640 --> 00:42:01,880 There aren't really any bright colours there, but there's a big stretch of red 768 00:42:01,880 --> 00:42:04,040 and it's going almost the whole length. 769 00:42:04,040 --> 00:42:05,920 If I go over here, 770 00:42:05,920 --> 00:42:11,600 I think I might be looking at a "T" so my best guess for that first sound is 771 00:42:11,600 --> 00:42:15,440 that it is "T". Now who would like to help me guess the next sound? 772 00:42:16,600 --> 00:42:18,640 Can I have you in the blue top, please? 773 00:42:18,640 --> 00:42:20,040 Thank you very much. 774 00:42:22,240 --> 00:42:23,840 Now what's your name? Joe. 775 00:42:23,840 --> 00:42:25,640 Sorry? Joe. Right, Joe, 776 00:42:25,640 --> 00:42:29,120 you're going to help me be a computer processor and work out what the next sound is. 777 00:42:29,120 --> 00:42:30,760 We've got one here. 778 00:42:31,800 --> 00:42:33,760 Now, Joe, 779 00:42:33,760 --> 00:42:38,240 you can see this looks quite different to that sound we were just looking for. 780 00:42:38,240 --> 00:42:41,840 Now can we find anything in our library that looks like that? 781 00:42:41,840 --> 00:42:44,800 I think you're right, 782 00:42:44,800 --> 00:42:46,640 I think we are dealing with "EE." 783 00:42:46,640 --> 00:42:48,440 Can you put that there for me? 784 00:42:48,440 --> 00:42:51,200 Now not to labour the point, if we carry on like this, 785 00:42:51,200 --> 00:42:54,480 you're not getting like really whip quick recognition of the speech, 786 00:42:54,480 --> 00:42:58,520 are you? We are taking our time so what I'm going to do is throw more 787 00:42:58,520 --> 00:43:00,560 processors at the problem, just as a computer would. 788 00:43:00,560 --> 00:43:02,760 I'm going to take a couple more volunteers please, 789 00:43:02,760 --> 00:43:06,520 so can I have you with the glasses, please, 790 00:43:06,520 --> 00:43:09,480 and can I have you with the polo neck sweater. 791 00:43:09,480 --> 00:43:11,800 Thank you very much. Can I have you? 792 00:43:11,800 --> 00:43:14,400 Thank you very much. Got to have a unicorn. 793 00:43:17,760 --> 00:43:21,720 OK, so, Unicorn, what's your name? 794 00:43:21,720 --> 00:43:23,360 Sasha. Sasha, lovely to meet you, Sasha. 795 00:43:23,360 --> 00:43:24,920 Hi. Evie. 796 00:43:24,920 --> 00:43:26,520 Evie. And your name is? 797 00:43:26,520 --> 00:43:27,800 Kit. Kit, excellent. 798 00:43:27,800 --> 00:43:29,880 Now there's four speech sounds left. 799 00:43:29,880 --> 00:43:33,840 I want you to each grab one and see if you can match it up with any of 800 00:43:33,840 --> 00:43:35,520 the speech sounds here, OK? 801 00:43:35,520 --> 00:43:37,200 There you go. 802 00:43:37,200 --> 00:43:38,720 And what have you got here? 803 00:43:38,720 --> 00:43:41,280 Look at this one. Can you see any that have got sloping shapes? 804 00:43:41,280 --> 00:43:43,480 Is it that one? I think you might be right there. 805 00:43:44,560 --> 00:43:46,440 So if you can remember, you came from number four. 806 00:43:48,040 --> 00:43:49,280 I think I have an S sound. 807 00:43:49,280 --> 00:43:51,000 I think you're absolutely right. 808 00:43:51,000 --> 00:43:52,520 You grab an S sound and pop it up. 809 00:43:53,640 --> 00:43:56,000 Now you have got the really difficult one here 810 00:43:56,000 --> 00:44:00,080 because when the speaker said it, they really underarticulated it, 811 00:44:00,080 --> 00:44:04,960 so what you have is this sound here 812 00:44:04,960 --> 00:44:07,560 and that's going to have to be our best guess, OK? 813 00:44:07,560 --> 00:44:10,080 OK, so if you can pop that in, we can see what we've got. 814 00:44:10,080 --> 00:44:11,760 Thank you very much. 815 00:44:11,760 --> 00:44:16,920 It could be team mates, it could be tea mate. 816 00:44:16,920 --> 00:44:20,600 We don't know still what the words are but we've now got a good guess 817 00:44:20,600 --> 00:44:23,600 what our speech sounds are and that takes us to our next stage. 818 00:44:23,600 --> 00:44:25,800 What I want to do first is say thank you very much 819 00:44:25,800 --> 00:44:27,880 to Joe, Sasha, Evie, Kit. 820 00:44:27,880 --> 00:44:29,920 Thank you very much, thank you. 821 00:44:34,800 --> 00:44:36,680 So the addition of processing power 822 00:44:36,680 --> 00:44:38,880 has really helped us be able to speed up this process 823 00:44:38,880 --> 00:44:41,080 and that's why you can talk to your phone without it 824 00:44:41,080 --> 00:44:43,760 taking that amount of time. 825 00:44:43,760 --> 00:44:47,800 And what we are doing here is pulling out what the speech sounds are 826 00:44:47,800 --> 00:44:49,840 but we still don't know what those words are, 827 00:44:49,840 --> 00:44:51,760 we don't know where the edges are. 828 00:44:51,760 --> 00:44:55,520 What we need to do is go to another level and what computers do 829 00:44:55,520 --> 00:44:57,600 is go to what's called a language level 830 00:44:57,600 --> 00:45:02,440 to start to break this stream of sounds up into words and into sentences. 831 00:45:02,440 --> 00:45:07,280 I need two new volunteers to decode a stream of speech for me, just like 832 00:45:07,280 --> 00:45:10,120 you were a computer. Can I have... 833 00:45:10,120 --> 00:45:12,600 you in the middle with the blue T-shirt? 834 00:45:12,600 --> 00:45:15,000 Yes, there you go, fantastic. 835 00:45:15,000 --> 00:45:17,360 Can I have you? Thank you very much, thank you. 836 00:45:22,880 --> 00:45:25,040 Now, what's your name? Max. 837 00:45:25,040 --> 00:45:27,000 Lovely to meet you, Max. And your name is? 838 00:45:27,000 --> 00:45:28,640 Cammy. Cammy, fantastic. 839 00:45:28,640 --> 00:45:31,440 What I'm going to give you is the same task a computer has got when 840 00:45:31,440 --> 00:45:34,520 it's worked out the speech sounds but it doesn't know what the words are. 841 00:45:34,520 --> 00:45:37,840 What you're going to do is see a list of speech sounds and I want you 842 00:45:37,840 --> 00:45:40,280 to try and work out what words could be in there. 843 00:45:40,280 --> 00:45:43,600 Sometimes the computer doesn't know what a sound is at all. 844 00:45:43,600 --> 00:45:46,280 You'll just see a question mark, maybe there was a cough, 845 00:45:46,280 --> 00:45:49,960 maybe there was a noise, so sometimes you're going to have to guess, OK? 846 00:45:49,960 --> 00:45:53,280 So I can just ask you to step over here, Max, thank you very much. 847 00:45:53,280 --> 00:45:56,440 OK and we will see it appear on the screen, OK? 848 00:45:56,440 --> 00:45:58,400 So reading down in that direction, 849 00:45:59,640 --> 00:46:01,480 Try reading that aloud. 850 00:46:01,480 --> 00:46:05,360 THEY READ WORDS ON SCREEN SLOWLY 851 00:46:18,040 --> 00:46:20,920 Now, think about the building that we are in. 852 00:46:20,920 --> 00:46:23,880 The Royal Institution. The Royal Institution, absolutely, 853 00:46:23,880 --> 00:46:25,480 so context can help you. 854 00:46:25,480 --> 00:46:27,600 You go back at that, we are at the Royal Institution. 855 00:46:27,600 --> 00:46:29,720 One last time. 856 00:46:29,720 --> 00:46:32,680 TOGETHER: Drums are really 857 00:46:32,680 --> 00:46:38,280 loud, so we have to 858 00:46:38,280 --> 00:46:42,880 use ear 859 00:46:44,680 --> 00:46:48,440 defenders at the Royal Institution. 860 00:46:48,440 --> 00:46:49,720 Well done. 861 00:46:51,680 --> 00:46:54,280 Thank you. Thank you, Cammy, thank you, Max. 862 00:46:54,280 --> 00:46:56,560 Thank you. 863 00:46:56,560 --> 00:47:00,480 So that's an extreme example, but what you are seeing there was essentially 864 00:47:00,480 --> 00:47:02,520 the same problem a computer is trying to solve. 865 00:47:02,520 --> 00:47:04,200 It's trying to find the edges, 866 00:47:04,200 --> 00:47:06,360 it's trying to work out where the words could be, 867 00:47:06,360 --> 00:47:09,320 and it's helped in this because it knows what the words are. 868 00:47:09,320 --> 00:47:13,320 It has a database of thousands and thousands of words and it also knows 869 00:47:13,320 --> 00:47:16,280 something about how sentences go together. 870 00:47:16,280 --> 00:47:19,880 It knows how probable it is that a word will follow another word. 871 00:47:19,880 --> 00:47:21,760 And if a word becomes more likely, 872 00:47:21,760 --> 00:47:25,120 the probability of it occurring actually changes in the computer 873 00:47:25,120 --> 00:47:27,280 so it's more easily activated. 874 00:47:27,280 --> 00:47:29,480 And, in fact, our brains do something similar. 875 00:47:29,480 --> 00:47:32,760 If you are listening to speech and you hear something which is highly 876 00:47:32,760 --> 00:47:35,320 predictable like "the ship sailed across the bay" 877 00:47:35,320 --> 00:47:38,120 then you will understand that sentence more easily. 878 00:47:38,120 --> 00:47:41,920 So we are seeing actually, in terms of a lot of the code of language, 879 00:47:41,920 --> 00:47:46,960 humans and computers are not as differently matched as we used to be. 880 00:47:46,960 --> 00:47:51,480 Computers are catching up with a lot of our linguistic abilities. 881 00:47:51,480 --> 00:47:54,280 Computers have become huge and very powerful, 882 00:47:54,280 --> 00:47:57,520 they can be searching through very large databases very quickly, 883 00:47:57,520 --> 00:48:01,280 and the internet means a little box like that can be connected to those 884 00:48:01,280 --> 00:48:02,960 large online databases. 885 00:48:02,960 --> 00:48:05,360 It doesn't need to have all the information there. 886 00:48:05,360 --> 00:48:10,880 But, of course, this is not the full story of how human language works. 887 00:48:10,880 --> 00:48:13,600 There is an entire level of communication 888 00:48:13,600 --> 00:48:17,560 that we use all the time that we have barely mentioned. 889 00:48:17,560 --> 00:48:21,000 Rather than just what we say when we're talking, 890 00:48:21,000 --> 00:48:25,200 we are always sending out information in how we say words. 891 00:48:26,760 --> 00:48:32,840 Everything I've talked about so far has been, if we think about brains, 892 00:48:32,840 --> 00:48:36,240 associated with properties of the left side of the brain. 893 00:48:37,920 --> 00:48:42,600 The left side of the brain in humans, for most people, is associated with 894 00:48:42,600 --> 00:48:46,080 how we decode speech, how we control our own voices. 895 00:48:47,240 --> 00:48:51,040 And the right side of the brain is a lot less interested in these 896 00:48:51,040 --> 00:48:54,800 linguistic properties of communication like words and sentences, 897 00:48:54,800 --> 00:48:58,320 and a lot more interested in all the other things that are going on when 898 00:48:58,320 --> 00:48:59,680 we are talking to people: 899 00:48:59,680 --> 00:49:01,240 who we are talking to. 900 00:49:01,240 --> 00:49:02,560 Are they being emotional? 901 00:49:02,560 --> 00:49:04,400 Are they telling a brilliant joke like, 902 00:49:04,400 --> 00:49:07,320 what is round and sounds like a trumpet? 903 00:49:07,320 --> 00:49:09,000 A crumpet. 904 00:49:09,000 --> 00:49:10,360 GROANS 905 00:49:10,360 --> 00:49:14,680 So to have a meaningful conversation with a machine or understand my 906 00:49:14,680 --> 00:49:19,360 brilliant jokes, we need to do something much more difficult. 907 00:49:19,360 --> 00:49:25,200 We need to add in to our computer the missing right half of its brain. 908 00:49:25,200 --> 00:49:28,880 Because actually when we are talking, when you're listening to somebody, 909 00:49:28,880 --> 00:49:32,760 there are very important aspects of our communication which you need to 910 00:49:32,760 --> 00:49:35,280 pick up on to really understand what somebody is saying. 911 00:49:35,280 --> 00:49:37,200 Not just when it's written down. 912 00:49:37,200 --> 00:49:40,320 The information when someone is speaking is very often being expressed as 913 00:49:40,320 --> 00:49:43,120 much by how they are talking as what they are saying. 914 00:49:43,120 --> 00:49:46,120 And this often refers to something called intonation. 915 00:49:46,120 --> 00:49:50,120 Intonation is how we vary the pitch, the speed, 916 00:49:50,120 --> 00:49:52,960 the melody of our voice when we are speaking. 917 00:49:52,960 --> 00:49:57,160 An example would be that we don't talk to each other like that. 918 00:49:57,160 --> 00:49:59,400 We would consider it to be quite strange. 919 00:49:59,400 --> 00:50:02,960 We are always using intonation to clarify, enhance, 920 00:50:02,960 --> 00:50:04,960 put in emphasis and emotion. 921 00:50:04,960 --> 00:50:07,160 In our brains, 922 00:50:07,160 --> 00:50:09,720 intonation is processed very differently 923 00:50:09,720 --> 00:50:11,800 from the words that you're listening to. 924 00:50:11,800 --> 00:50:14,800 On the whole, intonation is strongly found to be something 925 00:50:14,800 --> 00:50:19,080 that the right hemisphere deals with and is interested in. 926 00:50:19,080 --> 00:50:22,880 And often, it can be as important if not more important to the real 927 00:50:22,880 --> 00:50:24,440 meaning of what somebody is saying. 928 00:50:25,920 --> 00:50:29,760 We've got some recordings here of someone speaking in an emotional style 929 00:50:29,760 --> 00:50:33,640 and we have stripped out all the auditory information that tells you about 930 00:50:33,640 --> 00:50:36,360 the words they are saying and it's just leaving you with the intonation. 931 00:50:36,360 --> 00:50:38,280 See if you can guess the emotion. 932 00:50:38,280 --> 00:50:41,240 BUZZING 933 00:50:41,240 --> 00:50:43,040 Any guesses, does that sound happy? 934 00:50:43,040 --> 00:50:45,520 Angry? It sounds annoyed, doesn't it? 935 00:50:45,520 --> 00:50:46,600 Yeah, that was angry. 936 00:50:46,600 --> 00:50:48,000 That was very angry. 937 00:50:48,000 --> 00:50:49,080 Another one. 938 00:50:49,080 --> 00:50:51,800 BUZZING 939 00:50:53,800 --> 00:50:58,320 Yeah, lots of downward inflections. 940 00:50:58,320 --> 00:51:00,480 Sounds softer. 941 00:51:00,480 --> 00:51:04,360 That was a sad voice so you are definitely using intonation to get aspects 942 00:51:04,360 --> 00:51:07,560 of emotion out, but that's by no means the whole story. 943 00:51:07,560 --> 00:51:10,560 And to help me demonstrate why intonation is so very important to 944 00:51:10,560 --> 00:51:14,280 communication, please welcome news presenter and journalist 945 00:51:14,280 --> 00:51:16,600 Krishnan Guru-Murthy. 946 00:51:23,400 --> 00:51:27,120 Krishnan, obviously the words you say are absolutely critical to your job, 947 00:51:27,120 --> 00:51:31,320 but how you are saying them must be something that you're always thinking about. 948 00:51:31,320 --> 00:51:34,600 We use intonation to do all sorts of things on the news, all the 949 00:51:34,600 --> 00:51:36,320 time, and it's very, very complex. 950 00:51:36,320 --> 00:51:39,480 At the beginning of the news, we're saying, "This is urgent, 951 00:51:39,480 --> 00:51:41,160 "it's exciting, you've got to watch, 952 00:51:41,160 --> 00:51:44,360 "something really important happened today and I've got to tell you about 953 00:51:44,360 --> 00:51:46,920 "it." Yeah. And then once we get into the news, 954 00:51:46,920 --> 00:51:50,600 we are trying to use the right intonation for the type of story 955 00:51:50,600 --> 00:51:54,800 we are telling people, so we've got to try and hit the right note, literally. 956 00:51:54,800 --> 00:51:57,600 So if it's something very surprising, 957 00:51:57,600 --> 00:52:00,560 "The new president of the United States is Donald Trump!" 958 00:52:00,560 --> 00:52:02,960 That's quite surprising. That is quite surprising. 959 00:52:02,960 --> 00:52:07,680 If it's very serious or frightening news, something terrible has happened, 960 00:52:07,680 --> 00:52:11,040 "The new president of the United States is Donald Trump." 961 00:52:11,040 --> 00:52:14,200 It's a different thing. 962 00:52:14,200 --> 00:52:17,760 You wouldn't normally do that, but if you were sort of talking about 963 00:52:17,760 --> 00:52:20,240 an attack or a war or something like that, 964 00:52:20,240 --> 00:52:23,280 this is a very serious thing and you're trying to get people in the 965 00:52:23,280 --> 00:52:26,200 right frame of mind. But you're also trying not to frighten them 966 00:52:26,200 --> 00:52:28,280 and you've got to use intonation for that. 967 00:52:28,280 --> 00:52:31,400 Now we've got a very short experiment to do with Krishnan, 968 00:52:31,400 --> 00:52:34,360 who's going to read out some football results. 969 00:52:34,360 --> 00:52:38,000 Just by the intonation on his voice, I want you to... 970 00:52:38,000 --> 00:52:41,080 He's going to stop before he gets to the final score and we're going to 971 00:52:41,080 --> 00:52:45,960 work out if the last team had a higher score or a lower score than 972 00:52:45,960 --> 00:52:48,800 the first team. OK, shall we try one? 973 00:52:48,800 --> 00:52:52,840 Manchester City 2, West Bromwich Albion... 974 00:52:52,840 --> 00:52:53,800 Higher or lower? 975 00:52:55,280 --> 00:52:57,240 Nil. Nil, there you go, lower. 976 00:52:58,560 --> 00:53:01,760 West Ham United one, Tottenham Hotspur... 977 00:53:02,960 --> 00:53:04,360 Higher, yes. Five. 978 00:53:05,600 --> 00:53:07,880 Chelsea one, Liverpool... 979 00:53:07,880 --> 00:53:10,400 One - exactly, it's a draw. 980 00:53:11,640 --> 00:53:13,320 This is absolutely amazing. 981 00:53:13,320 --> 00:53:16,400 What Krishnan is doing, is over the whole course of the sentence, 982 00:53:16,400 --> 00:53:18,520 he's doing a kind of dance with his intonation 983 00:53:18,520 --> 00:53:22,000 that's completely keeping pace with the meaning of what he's saying. 984 00:53:22,000 --> 00:53:25,200 You've got weaving the intonation in and out of the words and you're 985 00:53:25,200 --> 00:53:27,720 picking up on that, you know what that means. 986 00:53:27,720 --> 00:53:31,640 Can you do an example of it incorrectly so we can hear what that sounds like? 987 00:53:33,600 --> 00:53:36,240 Stoke City four, Huddersfield... 988 00:53:37,360 --> 00:53:38,280 five. 989 00:53:39,880 --> 00:53:43,640 And when you say that, it sounds like it's the words that are wrong, whereas it's the 990 00:53:43,640 --> 00:53:46,000 intonation that's wrong, it's really striking. 991 00:53:46,000 --> 00:53:47,840 Thank you very, very much, Krishnan. 992 00:53:47,840 --> 00:53:49,280 Thank you. Thank you, Sophie. 993 00:53:53,800 --> 00:53:57,520 So we are using intonation all the time in regular conversation, 994 00:53:57,520 --> 00:54:02,520 and sometimes we think that that's all we do to pick up on emotion in the voice. 995 00:54:02,520 --> 00:54:06,120 Actually emotion in the voice is incredibly complex. 996 00:54:06,120 --> 00:54:09,440 There's a great deal going on because emotions change how your bodies work 997 00:54:09,440 --> 00:54:12,280 and how they feel. They can change many different aspects of your voice. 998 00:54:12,280 --> 00:54:13,760 Up to now, 999 00:54:13,760 --> 00:54:17,080 computers have really struggled with this kind of information. 1000 00:54:17,080 --> 00:54:20,600 But that's changing. 1001 00:54:20,600 --> 00:54:23,840 And people are starting to make more progress. 1002 00:54:23,840 --> 00:54:25,960 For our last demonstration, 1003 00:54:25,960 --> 00:54:28,560 I'm going to look at something really amazing - 1004 00:54:28,560 --> 00:54:32,480 a computer that can read emotions from human voices. 1005 00:54:32,480 --> 00:54:35,280 Now the acoustics in here are quite reverberant 1006 00:54:35,280 --> 00:54:38,880 and this computer has been built to work in the home, 1007 00:54:38,880 --> 00:54:41,680 and so what I'm going to do is just step outside where there's a lovely 1008 00:54:41,680 --> 00:54:44,400 carpet and it's a little bit more like being in someone's living room. 1009 00:54:44,400 --> 00:54:46,280 Let's go and meet Olly. 1010 00:54:48,240 --> 00:54:51,400 Hi. Hi, Raymond. Hello. 1011 00:54:51,400 --> 00:54:54,920 Now, tell me about Olly. 1012 00:54:54,920 --> 00:54:57,880 So Olly can actually understand what you're saying, as well as 1013 00:54:57,880 --> 00:55:00,200 emotions of your voice, the tone of your voice. 1014 00:55:00,200 --> 00:55:03,320 Excellent, what kind of thing do you do with Olly? 1015 00:55:03,320 --> 00:55:07,040 Olly actually enhances the communication between human and technology. 1016 00:55:07,040 --> 00:55:11,160 So he could be a digital assistant who's really understanding how you 1017 00:55:11,160 --> 00:55:13,040 feel, not just what you are saying? 1018 00:55:13,040 --> 00:55:15,320 Exactly. Yes. Can we find out more about how he works? 1019 00:55:15,320 --> 00:55:16,920 Sure. Is it easiest to demonstrate that? 1020 00:55:16,920 --> 00:55:20,080 Yes, of course. So what you need to do is just stand a sensible distance 1021 00:55:20,080 --> 00:55:22,720 from Olly, speak normal as usual, 1022 00:55:22,720 --> 00:55:26,240 and then maybe you can add some emotion to it when you speak. OK. 1023 00:55:26,240 --> 00:55:27,400 So, are you ready? 1024 00:55:27,400 --> 00:55:28,960 One, two, three. 1025 00:55:30,400 --> 00:55:33,240 SADLY: Hey, Ollie, what's the weather like in London today? 1026 00:55:35,680 --> 00:55:38,480 Oh, there we go, yes. 1027 00:55:41,000 --> 00:55:44,800 So is Ollie getting the emotion out of your voice by the sound of your 1028 00:55:44,800 --> 00:55:48,360 voice or the words you're saying, or is it both? 1029 00:55:48,360 --> 00:55:49,480 Actually it's both. 1030 00:55:49,480 --> 00:55:52,880 It recognises some acoustic components of what you said, 1031 00:55:52,880 --> 00:55:57,320 and it's also using the syntax and semantics of the words you used. 1032 00:55:57,320 --> 00:55:59,320 It's very clever. Thank you very much, 1033 00:55:59,320 --> 00:56:01,240 thank you. Thank you very much. 1034 00:56:04,440 --> 00:56:09,080 Olly is doing an amazing job of decoding emotion as well as words 1035 00:56:09,080 --> 00:56:14,160 from the human voice, and you can see from this just how hard emotion 1036 00:56:14,160 --> 00:56:17,840 is to read and how extremely nuanced it can be. 1037 00:56:17,840 --> 00:56:22,640 We find it very easy, because we've grown up around people using language, 1038 00:56:22,640 --> 00:56:25,800 emotion, social meaning in their voices all the time, 1039 00:56:25,800 --> 00:56:29,880 so it seems easy to us but it's absolutely vital to understanding language. 1040 00:56:35,160 --> 00:56:37,320 All around the world, 1041 00:56:37,320 --> 00:56:42,640 animals are communicating in very simple and very complex ways and we 1042 00:56:42,640 --> 00:56:44,840 have so much more to learn. 1043 00:56:44,840 --> 00:56:48,480 I've touched on just the surface of animal communication today. 1044 00:56:49,960 --> 00:56:56,240 Within this, humans do seem to be extraordinarily complex in their linguistic abilities. 1045 00:56:56,240 --> 00:57:00,000 But we're only really using language with each other. 1046 00:57:00,000 --> 00:57:03,320 Could there ever be another life form that could crack our code - 1047 00:57:03,320 --> 00:57:05,400 perhaps one we haven't even encountered yet? 1048 00:57:07,280 --> 00:57:11,240 Watching Carl Sagan describe his work on the Voyager space probes 1049 00:57:11,240 --> 00:57:14,120 40 years ago in the 1977 Christmas lectures, 1050 00:57:14,120 --> 00:57:17,280 literally set me on course to standing here today - 1051 00:57:17,280 --> 00:57:19,040 that's why I became a scientist. 1052 00:57:20,160 --> 00:57:22,800 The Voyagers contain golden records, 1053 00:57:22,800 --> 00:57:25,000 and the golden records contain the sounds of Earth, 1054 00:57:25,000 --> 00:57:28,760 including greetings in 55 different human languages. 1055 00:57:28,760 --> 00:57:30,480 GREETINGS IN DIFFERENT LANGUAGES 1056 00:57:36,320 --> 00:57:39,560 Now, the sounds of the Earth's languages 1057 00:57:39,560 --> 00:57:41,760 have long since left our solar system. 1058 00:57:41,760 --> 00:57:46,000 The Voyager probes are now 13 billion miles away from Earth. 1059 00:57:46,000 --> 00:57:49,120 They are the most distant objects ever created by humans. 1060 00:57:50,240 --> 00:57:54,080 Now, if those spacecraft ever encounter an alien life form, 1061 00:57:54,080 --> 00:57:57,960 maybe they can use these sounds to start to decode our language. 1062 00:57:57,960 --> 00:58:00,400 And if they have brains that work like ours, 1063 00:58:00,400 --> 00:58:03,760 maybe one day we could have conversations with them. 1064 00:58:03,760 --> 00:58:07,720 This year's lectures have explored our fundamental urge to communicate. 1065 00:58:07,720 --> 00:58:10,960 We've looked at where it comes from and why we're so very good at it, 1066 00:58:10,960 --> 00:58:15,240 and we just can't help but reach out to others. 1067 00:58:15,240 --> 00:58:18,760 The big question is whether there is another form of life out there 1068 00:58:18,760 --> 00:58:20,320 that could crack our code, 1069 00:58:20,320 --> 00:58:24,480 and I hope you've realised that it will need to have an incredible brain - 1070 00:58:24,480 --> 00:58:28,400 at least as incredible as your brains, to do so. 1071 00:58:28,400 --> 00:58:29,800 Thank you.