1 00:00:00,000 --> 00:00:02,740 Welcome to the Deep Dive. Today, we're really 2 00:00:02,740 --> 00:00:04,839 getting into the weeds, looking at the deep structure 3 00:00:04,839 --> 00:00:07,559 of how humans communicate. That's right. We're 4 00:00:07,559 --> 00:00:09,779 tackling the relationship between sound itself, 5 00:00:10,199 --> 00:00:13,039 the acoustics, and how we actually build meaning 6 00:00:13,039 --> 00:00:16,300 from it, the linguistics, the cognition. And 7 00:00:16,300 --> 00:00:18,399 our main source material comes from a pretty 8 00:00:18,399 --> 00:00:20,679 significant event, right? The Parole et Musique 9 00:00:20,679 --> 00:00:23,500 Colloquium at the Collège de France back in 2008. 10 00:00:23,879 --> 00:00:26,859 Yeah, a real powerhouse gathering. Thinkers from 11 00:00:26,859 --> 00:00:29,179 acoustics, neuroscience, linguistics, philosophy 12 00:00:29,179 --> 00:00:31,780 all focused on this puzzle. The puzzle being 13 00:00:31,780 --> 00:00:35,840 how speech and music, which seem built from similar 14 00:00:35,840 --> 00:00:38,880 stuff, sound waves, organized hierarchically. 15 00:00:38,979 --> 00:00:41,020 You know, simple elements like notes or phonemes 16 00:00:41,020 --> 00:00:43,380 combining into complex things like phrases or 17 00:00:43,380 --> 00:00:45,039 melodies. How did they end up conveying such 18 00:00:45,039 --> 00:00:47,700 different kinds of meaning? Or maybe do they? 19 00:00:47,899 --> 00:00:49,640 Well, that's the core question, isn't it? It 20 00:00:49,640 --> 00:00:52,000 gets into some classic debates. Is musical ability 21 00:00:52,000 --> 00:00:54,719 just sort of piggybacking on our language faculty? 22 00:00:55,039 --> 00:00:58,359 Like an evolutionary add -on? Maybe. Or is it, 23 00:00:58,380 --> 00:01:00,619 as someone like Steven Pinker famously suggested, 24 00:01:00,920 --> 00:01:03,859 more of an artifice cultural, a cultural invention? 25 00:01:04,400 --> 00:01:07,340 Ah, yes, the auditory cheesecake argument. Exactly. 26 00:01:07,359 --> 00:01:09,599 Something we invented because it tickles our 27 00:01:09,599 --> 00:01:12,640 cognitive pleasure centers. But it isn't fundamental 28 00:01:12,640 --> 00:01:14,739 in the way language is. That's a pretty stark 29 00:01:14,739 --> 00:01:18,310 contrast. Cheesecake versus, well, a... A core 30 00:01:18,310 --> 00:01:20,769 cognitive ability. It is. And then you've got 31 00:01:20,769 --> 00:01:23,090 the artist's perspective, maybe best summed up 32 00:01:23,090 --> 00:01:26,129 by Wagner. Music begins where the power of words 33 00:01:26,129 --> 00:01:29,769 stops. Right. The idea that music taps into emotions 34 00:01:29,769 --> 00:01:32,269 and structures that language just can't reach, 35 00:01:32,310 --> 00:01:34,969 while words handle the precise propositional 36 00:01:34,969 --> 00:01:37,469 stuff. So our mission today for you listening 37 00:01:37,469 --> 00:01:40,030 is to unpack that. We're going deep into the 38 00:01:40,030 --> 00:01:43,680 physiology, the structure, the cognition. to 39 00:01:43,680 --> 00:01:46,019 really trace how sound waves hitting our ears 40 00:01:46,019 --> 00:01:49,079 get transformed into the complex tapestry of 41 00:01:49,079 --> 00:01:51,500 human meaning, whether it's a sentence or a symphony. 42 00:01:51,640 --> 00:01:53,560 Okay, let's start at the very beginning. The 43 00:01:53,560 --> 00:01:55,579 ear, we often think of it as a microphone, right? 44 00:01:55,659 --> 00:01:58,420 Just recording sound. Well, yes and no. Christine 45 00:01:58,420 --> 00:02:00,959 Petty's work, for instance, emphasizes that it's 46 00:02:00,959 --> 00:02:03,939 not a high -fidelity recorder. It's an acoustico 47 00:02:03,939 --> 00:02:06,620 -electric system. Meaning it converts sound waves 48 00:02:06,620 --> 00:02:09,539 to electrical signals for the brain? Yes, but 49 00:02:09,539 --> 00:02:12,759 crucially... It actively shapes that signal along 50 00:02:12,759 --> 00:02:16,120 the way. It's not passive. How so? I mean, isn't 51 00:02:16,120 --> 00:02:18,919 the goal accuracy? The goal is useful information, 52 00:02:19,219 --> 00:02:21,599 which isn't always the same as perfect fidelity. 53 00:02:21,960 --> 00:02:25,139 For starters, human hearing has incredible temporal 54 00:02:25,139 --> 00:02:28,319 resolution. We can detect timing differences 55 00:02:28,319 --> 00:02:31,180 far smaller than needed just for understanding 56 00:02:31,180 --> 00:02:34,289 speech and quiet. Okay, so it's highly sensitive, 57 00:02:34,430 --> 00:02:36,969 but you said it shapes the sound. It distorts 58 00:02:36,969 --> 00:02:39,129 it. It does, and this is fascinating. It's actually 59 00:02:39,129 --> 00:02:42,389 an advantage. The system is highly nonlinear. 60 00:02:42,610 --> 00:02:46,009 Nonlinear meaning what? It means the output isn't 61 00:02:46,009 --> 00:02:48,509 directly proportional to the input. For one thing, 62 00:02:48,629 --> 00:02:51,250 it amplifies quiet sounds much more than loud 63 00:02:51,250 --> 00:02:54,270 sounds. This compresses the dynamic range, helping 64 00:02:54,270 --> 00:02:56,550 us hear both whispers and shouts. Okay, that 65 00:02:56,550 --> 00:02:58,770 makes sense. But actual distortion, adding things 66 00:02:58,770 --> 00:03:01,830 that aren't there. Yes. This goes way back. In 67 00:03:01,830 --> 00:03:04,469 1754, the violinist Giuseppe Tartini noticed 68 00:03:04,469 --> 00:03:07,189 something odd. When he played two notes strongly, 69 00:03:07,430 --> 00:03:10,349 people heard a third lower note that wasn't actually 70 00:03:10,349 --> 00:03:13,150 being played. He called it the terzo suono, the 71 00:03:13,150 --> 00:03:16,150 third sound. Exactly. We now call these combination 72 00:03:16,150 --> 00:03:20,250 tones, or more technically, products, de -distortion, 73 00:03:20,250 --> 00:03:23,650 distortion products. The ear literally creates 74 00:03:23,650 --> 00:03:26,620 them. How? What's the mechanism in the ear? It 75 00:03:26,620 --> 00:03:28,599 happens primarily in the cochlea, the spiral 76 00:03:28,599 --> 00:03:30,680 -shaped part of the inner ear. Specifically, 77 00:03:30,680 --> 00:03:34,159 tiny cells called outer hair cells play a key 78 00:03:34,159 --> 00:03:37,099 role. They don't just sense vibration, they actively 79 00:03:37,099 --> 00:03:40,219 move. They push and pull on the surrounding structures. 80 00:03:40,500 --> 00:03:42,919 They physically change shape. Yes, driven by 81 00:03:42,919 --> 00:03:45,520 the sound input. This active mechanical process 82 00:03:45,520 --> 00:03:48,360 amplifies the signal, especially weak signals. 83 00:03:48,719 --> 00:03:51,280 But it also introduces these nonlinearities, 84 00:03:51,319 --> 00:03:53,939 these distortions, adding those combination tones 85 00:03:53,939 --> 00:03:56,379 to the signal sent to the brain. Wow. And this 86 00:03:56,379 --> 00:03:59,039 is useful. It's thought to help sharpen frequency 87 00:03:59,039 --> 00:04:02,180 tuning and enhance sensitivity. And we can even 88 00:04:02,180 --> 00:04:04,139 measure it. You know those newborn hearing screening 89 00:04:04,139 --> 00:04:06,960 tests? They often measure otoacoustic emissions. 90 00:04:07,560 --> 00:04:10,159 Those are faint sounds produced by the cochlea 91 00:04:10,159 --> 00:04:12,759 itself, generated by the activity of those outer 92 00:04:12,759 --> 00:04:15,430 hair cells. If you can't detect them, it suggests 93 00:04:15,430 --> 00:04:17,850 that amplification and distortion mechanism isn't 94 00:04:17,850 --> 00:04:20,829 working properly. So the ear isn't just a passive 95 00:04:20,829 --> 00:04:24,149 receiver. It's an active, slightly distorting 96 00:04:24,149 --> 00:04:27,350 amplifier tuned for maximum clarity. That's a 97 00:04:27,350 --> 00:04:28,850 great way to put it. It prepares the signal. 98 00:04:29,029 --> 00:04:31,110 Okay, so the ear processes the incoming wave. 99 00:04:31,310 --> 00:04:33,290 Now let's talk about the structure of the sound 100 00:04:33,290 --> 00:04:36,329 itself. Jacques Bouvras points us towards Hermann 101 00:04:36,329 --> 00:04:41,569 von Helmholtz. Ah, Helmholtz. A giant. His théorie 102 00:04:41,569 --> 00:04:43,850 physiologique de la musique was groundbreaking 103 00:04:43,850 --> 00:04:46,610 in the 19th century. He really wanted to put 104 00:04:46,610 --> 00:04:48,550 the understanding of music and sound perception 105 00:04:48,550 --> 00:04:51,370 on a solid scientific footing, moving away from, 106 00:04:51,470 --> 00:04:53,870 well, more speculative ideas. And he focused 107 00:04:53,870 --> 00:04:56,470 on the physical composition of sound. Precisely. 108 00:04:56,470 --> 00:04:58,670 He analyzed what makes up most natural sounds 109 00:04:58,670 --> 00:05:01,009 like a vibrating string or a voice. It's not 110 00:05:01,009 --> 00:05:02,970 just one single frequency. There's the main pitch, 111 00:05:03,029 --> 00:05:06,360 the fundamental frequency, or... F0. S0. S0, 112 00:05:06,560 --> 00:05:09,120 yes. But on top of that, there are harmonics. 113 00:05:09,120 --> 00:05:10,860 These are frequencies that are exact integer 114 00:05:10,860 --> 00:05:13,920 multiples of the F0. So if F0 is 100 hertz, the 115 00:05:13,920 --> 00:05:17,399 harmonics are at 200 hertz, 300 hertz, 400 hertz, 116 00:05:17,399 --> 00:05:19,600 and so on. And the combination of these harmonics, 117 00:05:19,620 --> 00:05:21,439 their relative strength, that's what gives a 118 00:05:21,439 --> 00:05:25,439 sound its character. It's timbam. Exactly. Timbra 119 00:05:25,439 --> 00:05:27,879 is why a violin and a flute playing the same 120 00:05:27,879 --> 00:05:30,199 note at the same loudness for the same duration 121 00:05:30,199 --> 00:05:33,600 still sound completely different. Helmholtz realized 122 00:05:33,600 --> 00:05:36,180 timbre is all about the relative intensity of 123 00:05:36,180 --> 00:05:39,040 those harmonics. So one instrument might boost 124 00:05:39,040 --> 00:05:41,660 the second and fourth harmonic, another the third 125 00:05:41,660 --> 00:05:44,439 and fifth. Right. And to perceive this difference, 126 00:05:44,660 --> 00:05:47,079 Helmholtz theorized that the ear must somehow 127 00:05:47,079 --> 00:05:50,480 act like a frequency analyzer. It has to decompose 128 00:05:50,480 --> 00:05:53,339 that complex incoming sound wave back into its 129 00:05:53,339 --> 00:05:55,939 basic sinusoidal components. Like performing 130 00:05:55,939 --> 00:05:58,980 a Fourier analysis. That was his insight. He 131 00:05:58,980 --> 00:06:00,980 proposed that the basilar membrane within the 132 00:06:00,980 --> 00:06:03,480 cochlea essentially does this. Different parts 133 00:06:03,480 --> 00:06:05,560 of the membrane vibrate more strongly in response 134 00:06:05,560 --> 00:06:07,459 to different frequencies. Creating a kind of 135 00:06:07,459 --> 00:06:09,560 map. It's a spatial map of frequency along its 136 00:06:09,560 --> 00:06:12,620 length. It's called the tonotopic map. High frequencies 137 00:06:12,620 --> 00:06:14,779 stimulate one end, low frequencies the other. 138 00:06:14,860 --> 00:06:17,120 So the ear physically separates the frequency 139 00:06:17,120 --> 00:06:19,339 components of the sound before sending that information 140 00:06:19,339 --> 00:06:22,480 onwards. Okay, so the ear actively processes. 141 00:06:22,579 --> 00:06:25,360 It analyzes the harmonic structure. But then 142 00:06:25,360 --> 00:06:27,519 the brain takes it another step further, doesn't 143 00:06:27,519 --> 00:06:29,519 it, with this idea of the missing fundamental. 144 00:06:29,959 --> 00:06:34,079 Ah, yes. This is truly remarkable. It shows perception 145 00:06:34,079 --> 00:06:36,699 isn't just about receiving what's there. It's 146 00:06:36,699 --> 00:06:39,819 about constructing what should be there. Explain 147 00:06:39,819 --> 00:06:42,300 that. How can we perceive a fundamental frequency 148 00:06:42,300 --> 00:06:45,720 if it's missing from the sound? Your brain, or 149 00:06:45,720 --> 00:06:48,560 rather your auditory system, calculates it based 150 00:06:48,560 --> 00:06:51,100 on the harmonics it does receive. If the sound 151 00:06:51,100 --> 00:06:54,420 contains strong energy at, say, 600 hertz, 800 152 00:06:54,420 --> 00:06:56,660 hertz and 1 ,000 hertz. Which are multiples of 153 00:06:56,660 --> 00:06:59,120 200 hertz. Exactly. They are the third, fourth, 154 00:06:59,199 --> 00:07:01,019 and fifth harmonics of a 200 hertz fundamental. 155 00:07:01,300 --> 00:07:03,459 Your auditory system finds the greatest common 156 00:07:03,459 --> 00:07:05,240 divisor of the harmonic frequencies present, 157 00:07:05,379 --> 00:07:08,399 and you perceive the pitch as 200 hertz, even 158 00:07:08,399 --> 00:07:10,939 if there's zero acoustic energy at 200 hertz 159 00:07:10,939 --> 00:07:13,160 in the signal hitting your ear. That's wild. 160 00:07:13,420 --> 00:07:15,699 Does this happen in everyday life? All the time. 161 00:07:15,740 --> 00:07:18,620 The classic example is the telephone. Old analog 162 00:07:18,620 --> 00:07:20,800 phone lines had limited bandwidth. They often 163 00:07:20,800 --> 00:07:22,699 couldn't transmit the low fundamental frequencies 164 00:07:22,699 --> 00:07:25,620 of, say, a deep male voice. But you could still 165 00:07:25,620 --> 00:07:27,699 tell the person's pitch was low. Absolutely. 166 00:07:27,759 --> 00:07:30,699 Your brain just reconstructed the missing F0 167 00:07:30,699 --> 00:07:33,259 from the higher harmonics that did get through. 168 00:07:33,379 --> 00:07:35,879 The voice didn't sound unnaturally high -pitched. 169 00:07:35,939 --> 00:07:37,779 It sounded like the person. And this isn't just 170 00:07:37,779 --> 00:07:40,420 a human thing. You mentioned goldfish. Believe 171 00:07:40,420 --> 00:07:43,160 it or not, yes. Yeah. This ability to perceive 172 00:07:43,160 --> 00:07:45,699 the pitch. Based on harmonic structure, even 173 00:07:45,699 --> 00:07:48,339 with a missing fundamental, is an ancient auditory 174 00:07:48,339 --> 00:07:50,139 function. It's been traced back surprisingly 175 00:07:50,139 --> 00:07:53,240 far in evolutionary terms. Why? What's the advantage? 176 00:07:53,339 --> 00:07:56,600 Why evolve this complex reconstruction ability? 177 00:07:57,000 --> 00:08:00,560 Several key advantages. First, grouping. It helps 178 00:08:00,560 --> 00:08:02,600 bundle all the harmonics coming from a single 179 00:08:02,600 --> 00:08:05,480 sound source together. Imagine two people talking 180 00:08:05,480 --> 00:08:07,839 at once. Their harmonics are all mixed in the 181 00:08:07,839 --> 00:08:10,459 air. Reconstructing the F0 for each helps your 182 00:08:10,459 --> 00:08:12,120 brain assign the right harmonics to the right 183 00:08:12,120 --> 00:08:14,600 voice. Ah, source segregation, the cocktail party 184 00:08:14,600 --> 00:08:17,360 problem again. Exactly. Second, estimating size. 185 00:08:17,600 --> 00:08:20,019 The fundamental frequency is related to the size 186 00:08:20,019 --> 00:08:22,279 of the sound source. Think vocal cord length. 187 00:08:22,720 --> 00:08:25,879 Lower F0 generally means a larger animal. Useful 188 00:08:25,879 --> 00:08:27,579 for knowing if that sound in the bushes is a 189 00:08:27,579 --> 00:08:29,939 mouse or a bear. Okay, survival value there. 190 00:08:30,139 --> 00:08:33,320 Definitely. And third, as we touched on, robustness. 191 00:08:33,820 --> 00:08:36,519 It makes pitch precession incredibly resilient 192 00:08:36,519 --> 00:08:38,909 to noise or filtering. like the telephone example. 193 00:08:39,169 --> 00:08:41,830 It allows us to extract that essential pitch 194 00:08:41,830 --> 00:08:45,120 information even from degraded signals. So the 195 00:08:45,120 --> 00:08:47,559 auditory system is this incredible biological 196 00:08:47,559 --> 00:08:51,120 engine, distorting, analyzing, and even constructing 197 00:08:51,120 --> 00:08:54,259 information to give us a stable perception. Precisely. 198 00:08:54,259 --> 00:08:56,379 It lays the foundation for everything that follows. 199 00:08:56,600 --> 00:08:58,740 Okay, let's build on that foundation now. How 200 00:08:58,740 --> 00:09:01,259 does this sophisticated auditory processing connect 201 00:09:01,259 --> 00:09:04,120 specifically to producing and understanding speech? 202 00:09:04,399 --> 00:09:07,100 We need to talk about formants. Right. So F0 203 00:09:07,100 --> 00:09:09,500 gives us the overall pitch or melody of the voice, 204 00:09:09,639 --> 00:09:12,549 the prosody. But the specific speech sounds, 205 00:09:12,629 --> 00:09:14,669 the vowels and consonants, depend on something 206 00:09:14,669 --> 00:09:17,549 else. How the vocal tract shapes the sound produced 207 00:09:17,549 --> 00:09:20,330 by the vocal cords. Our mouth, throat, nasal 208 00:09:20,330 --> 00:09:23,330 cavity, acting as filters or resonators. Exactly. 209 00:09:23,529 --> 00:09:26,509 The vocal tract is like a configurable tube with 210 00:09:26,509 --> 00:09:28,850 about 17 different points where we can change 211 00:09:28,850 --> 00:09:32,769 its shape, moving the tongue, lips, jaw, vellum. 212 00:09:33,360 --> 00:09:36,000 As the sound wave from the vocal cords, rich 213 00:09:36,000 --> 00:09:38,539 in harmonics, travels through this tube, certain 214 00:09:38,539 --> 00:09:40,580 frequencies get amplified because they resonate 215 00:09:40,580 --> 00:09:43,039 in the cavities we form. And those peaks of resonance, 216 00:09:43,320 --> 00:09:45,960 those are the formants. That's it. Formants are 217 00:09:45,960 --> 00:09:48,200 the characteristic frequency peaks in the spectrum 218 00:09:48,200 --> 00:09:50,820 of a speech sound. Different shapes of the vocal 219 00:09:50,820 --> 00:09:53,580 tract create different formant patterns. It's 220 00:09:53,580 --> 00:09:55,559 the specific frequencies of the first two or 221 00:09:55,559 --> 00:09:57,620 three formants, and crucially how they change 222 00:09:57,620 --> 00:10:00,259 over time, that largely define which vowel or 223 00:10:00,259 --> 00:10:03,120 consonant we perceive. So while music might focus 224 00:10:03,120 --> 00:10:05,539 more on the stability and relationship of F0 225 00:10:05,539 --> 00:10:08,519 pitch and harmonic fusion, language relies heavily 226 00:10:08,519 --> 00:10:10,820 on the rapid dynamic shifts in these formant 227 00:10:10,820 --> 00:10:12,840 frequencies, occurring at a pretty sustained 228 00:10:12,840 --> 00:10:16,200 rhythm to create its distinct sound units. F0 229 00:10:16,200 --> 00:10:19,620 for melody, formants for the code. Okay. This 230 00:10:19,620 --> 00:10:21,320 code, these phonemes, they then get assembled 231 00:10:21,320 --> 00:10:23,799 into words and sentences using grammar. This 232 00:10:23,799 --> 00:10:26,019 brings us to the abstract rules, the syntax, 233 00:10:26,360 --> 00:10:29,039 Luigi Rizzi's work, drawing on Chomsky's framework. 234 00:10:29,659 --> 00:10:31,700 Yes, the principles and parameters framework. 235 00:10:32,000 --> 00:10:35,299 This is a major theoretical approach in linguistics, 236 00:10:35,440 --> 00:10:38,000 trying to explain a fundamental observation. 237 00:10:39,330 --> 00:10:42,250 Human languages seem incredibly diverse on the 238 00:10:42,250 --> 00:10:44,450 surface. Different word orders, different sounds, 239 00:10:44,570 --> 00:10:47,230 different structures. Yet any child can learn 240 00:10:47,230 --> 00:10:50,129 any language they're exposed to, suggesting there 241 00:10:50,129 --> 00:10:53,879 must be some underlying shared foundation. Universal 242 00:10:53,879 --> 00:10:56,779 grammar. The principles part of the theory. Right. 243 00:10:57,049 --> 00:10:59,710 Universal grammar proposes a set of invariant 244 00:10:59,710 --> 00:11:02,389 principles common to all human languages that 245 00:11:02,389 --> 00:11:05,519 constitute our innate language faculty. This 246 00:11:05,519 --> 00:11:07,799 provides the basic toolkit for building language. 247 00:11:08,059 --> 00:11:09,820 But how do we get the diversity then? That's 248 00:11:09,820 --> 00:11:12,279 the parameters part. Exactly. Within this universal 249 00:11:12,279 --> 00:11:14,399 structure, there are specific points of variation, 250 00:11:14,700 --> 00:11:16,480 the parameters, which are typically conceived 251 00:11:16,480 --> 00:11:19,539 as binary switches. A language sets each parameter 252 00:11:19,539 --> 00:11:21,620 one way or the other during the acquisition process. 253 00:11:22,039 --> 00:11:25,110 Like a switch for word order. Verb, object, or 254 00:11:25,110 --> 00:11:27,590 object, verb. That's a classic example. Some 255 00:11:27,590 --> 00:11:31,429 languages are VO, like English, eat apples. Others 256 00:11:31,429 --> 00:11:35,610 are OV, like Japanese, Ringo Otaburu, Apple's 257 00:11:35,610 --> 00:11:39,809 ACC8. This choice is a parameter setting. The 258 00:11:39,809 --> 00:11:42,330 key idea is that the vast complexity of language 259 00:11:42,330 --> 00:11:45,210 arises from a fifth set of universal principles 260 00:11:45,210 --> 00:11:47,490 interacting with a limited number of these parameter 261 00:11:47,490 --> 00:11:50,960 choices. So what are the truly fundamental operations 262 00:11:50,960 --> 00:11:54,379 universal grammar provides? What allows us to 263 00:11:54,379 --> 00:11:57,580 build complex sentences at all? The theory highlights 264 00:11:57,580 --> 00:12:00,240 a couple of absolutely core operations. The most 265 00:12:00,240 --> 00:12:03,159 basic one is often called fusion or merge. It's 266 00:12:03,159 --> 00:12:05,519 incredibly simple. Take two linguistic objects, 267 00:12:05,700 --> 00:12:09,000 say the and cat, and combine them to make a new 268 00:12:09,000 --> 00:12:11,279 larger object, the cat. And you can just keep 269 00:12:11,279 --> 00:12:13,220 doing this, combining the new object with another 270 00:12:13,220 --> 00:12:15,860 one? Recursively, yes. Combine the cat with sat, 271 00:12:16,000 --> 00:12:18,179 then that with on the mat, and so on. The simple 272 00:12:18,179 --> 00:12:20,299 operation is what gives language its power to 273 00:12:20,299 --> 00:12:22,580 generate infinitely complex, hierarchically structured 274 00:12:22,580 --> 00:12:25,440 sentences. No known human language is limited 275 00:12:25,440 --> 00:12:28,039 to just simple atomic expressions. Okay. Fusion 276 00:12:28,039 --> 00:12:30,519 builds the structure. What's the other key operation? 277 00:12:30,720 --> 00:12:32,960 You mentioned movement earlier. Right. Movement. 278 00:12:33,179 --> 00:12:36,820 This is needed because very often where a word 279 00:12:36,820 --> 00:12:39,460 or phrase is logically interpreted in a sentence 280 00:12:39,460 --> 00:12:41,620 isn't where it's actually pronounced. Give us 281 00:12:41,620 --> 00:12:43,919 that question example again. Sure. Take what 282 00:12:43,919 --> 00:12:47,029 did Mary buy? Logically, what? is the object 283 00:12:47,029 --> 00:12:49,549 of by. The underlying structure is something 284 00:12:49,549 --> 00:12:52,990 like Mary bought what. But to form the question 285 00:12:52,990 --> 00:12:55,730 in English, the word what has to move to the 286 00:12:55,730 --> 00:12:58,690 beginning of the sentence. So movement rearranges 287 00:12:58,690 --> 00:13:01,110 the pieces built by fusion for communicative 288 00:13:01,110 --> 00:13:03,929 purposes, like forming questions or focusing 289 00:13:03,929 --> 00:13:06,590 certain elements. Precisely. And languages differ 290 00:13:06,590 --> 00:13:09,370 in what moves and when it moves. But the operation 291 00:13:09,370 --> 00:13:12,200 itself seems to be universal. This connection 292 00:13:12,200 --> 00:13:14,720 between the parameters and the actual shape of 293 00:13:14,720 --> 00:13:16,960 words, the morphology, is also really tight, 294 00:13:17,059 --> 00:13:18,960 isn't it? Yeah. The verb movement example is 295 00:13:18,960 --> 00:13:21,179 fascinating. It really is. This is where you 296 00:13:21,179 --> 00:13:23,480 see the abstract syntax connecting directly to 297 00:13:23,480 --> 00:13:25,200 the nitty -gritty details of language structure. 298 00:13:25,799 --> 00:13:28,220 The V to T movement parameter is a great case 299 00:13:28,220 --> 00:13:31,240 study. Explain V to T again. Verb moving to tense. 300 00:13:31,620 --> 00:13:34,820 Yes. It concerns whether the main verb in a sentence 301 00:13:34,820 --> 00:13:37,860 has to move syntactically upwards to merge with 302 00:13:37,860 --> 00:13:40,629 the tense marker, T. The trigger seems to be 303 00:13:40,629 --> 00:13:42,429 the morphological status of that tense marker. 304 00:13:42,690 --> 00:13:45,889 So, in a language like French... Where verb endings 305 00:13:45,889 --> 00:13:47,889 for tense are rich and act like affixes they 306 00:13:47,889 --> 00:13:50,730 can't stand alone, the verb root must move up 307 00:13:50,730 --> 00:13:53,509 to attach to them. That's why in French, the 308 00:13:53,509 --> 00:13:55,549 verb typically comes before negation or certain 309 00:13:55,549 --> 00:14:00,110 adverbs. Jean ne mange pas. Jean, N -A -G dates 310 00:14:00,110 --> 00:14:03,110 not. Mange moves past pas. Whereas in modern 311 00:14:03,110 --> 00:14:05,669 English, the verb usually stays put. Right. Our 312 00:14:05,669 --> 00:14:08,110 tense endings, A -D -A -S -S, are considered 313 00:14:08,110 --> 00:14:10,350 morphologically weak. They don't force the verb 314 00:14:10,350 --> 00:14:12,529 to move. If we need something before negation, 315 00:14:12,649 --> 00:14:15,629 we insert an auxiliary verb like dough John does 316 00:14:15,629 --> 00:14:18,370 not eat. The main verb eat stays after the negation. 317 00:14:18,950 --> 00:14:21,289 Modals like will or can are already independent 318 00:14:21,289 --> 00:14:23,509 words in the tense position, so the main verb 319 00:14:23,509 --> 00:14:25,830 also stays put. And the clincher is the historical 320 00:14:25,830 --> 00:14:28,110 evidence from English itself. This is the really 321 00:14:28,110 --> 00:14:30,200 compelling part for the theory. If the richness 322 00:14:30,200 --> 00:14:32,919 of tense morphology causes verb movement, then 323 00:14:32,919 --> 00:14:34,840 if a language loses its rich morphology over 324 00:14:34,840 --> 00:14:37,379 time, it should also lose verb movement. And 325 00:14:37,379 --> 00:14:40,059 that happened in English. Exactly. As English 326 00:14:40,059 --> 00:14:42,559 evolved from Old English through Middle English, 327 00:14:42,779 --> 00:14:45,700 its system of verb conjugations became drastically 328 00:14:45,700 --> 00:14:49,240 simplified, a process called morphological impoverishment. 329 00:14:49,519 --> 00:14:52,379 And right alongside this simplification, English 330 00:14:52,379 --> 00:14:55,580 gradually lost the syntactic rule of V to T movement. 331 00:14:55,840 --> 00:14:58,509 So Shakespeare could write, he loves not. with 332 00:14:58,509 --> 00:15:01,309 the verb before not, but we have to say he does 333 00:15:01,309 --> 00:15:05,129 not love. Precisely. The loss of the rich endings 334 00:15:05,129 --> 00:15:08,090 seems to have directly caused the change in the 335 00:15:08,090 --> 00:15:10,490 syntactic rule. It's a beautiful correlation 336 00:15:10,490 --> 00:15:13,149 between the sound form structure, morphology, 337 00:15:13,289 --> 00:15:15,590 and the abstract sentence structure, syntax. 338 00:15:16,169 --> 00:15:17,789 And you see this cross -linguistically, too, 339 00:15:17,929 --> 00:15:21,009 with Icelandic versus mainland Scandinavia. Yes, 340 00:15:21,129 --> 00:15:23,929 comparative evidence backs it up. Icelandic retains 341 00:15:23,929 --> 00:15:26,929 very rich verb agreement morphology, and it consistently 342 00:15:26,929 --> 00:15:30,269 shows V to T movement. Closely related languages 343 00:15:30,269 --> 00:15:32,070 like Swedish and Danish went through radical 344 00:15:32,070 --> 00:15:34,470 morphological simplification, much like English 345 00:15:34,470 --> 00:15:36,590 did, and they generally lack V to T movement 346 00:15:36,590 --> 00:15:38,870 today. That's powerful evidence for this principles 347 00:15:38,870 --> 00:15:40,970 and parameters approach. If people want to dig 348 00:15:40,970 --> 00:15:43,970 into this, Rizzi's work is key. Absolutely. Luigi 349 00:15:43,970 --> 00:15:46,289 Rizzi's research on the fine structure of the 350 00:15:46,289 --> 00:15:48,929 left periphery of the clause and the idea of 351 00:15:48,929 --> 00:15:51,889 a syntactic cartography is central to understanding 352 00:15:51,889 --> 00:15:54,649 these kinds of detailed syntactic analyses. Okay, 353 00:15:54,710 --> 00:15:56,950 we've covered the ear, sound structure, and the 354 00:15:56,950 --> 00:15:59,750 abstract rules of syntax. But there's an even 355 00:15:59,750 --> 00:16:02,809 more basic question. Where did the units of speech, 356 00:16:03,070 --> 00:16:06,169 sound, the phonemes, come from in the first place? 357 00:16:06,620 --> 00:16:09,659 Why does language use discrete sounds like B 358 00:16:09,659 --> 00:16:12,720 -A -T instead of just continuous whooshes and 359 00:16:12,720 --> 00:16:15,340 cries? That's a huge question in language evolution. 360 00:16:16,059 --> 00:16:18,960 Pierre -Yves Oudia tackles this, arguing that 361 00:16:18,960 --> 00:16:20,980 standard natural selection, while necessary, 362 00:16:21,379 --> 00:16:23,440 might not be sufficient to explain the emergence 363 00:16:23,440 --> 00:16:26,559 of this combinatorial digital system. So what's 364 00:16:26,559 --> 00:16:29,159 the missing piece? He points to auto -organization 365 00:16:29,159 --> 00:16:32,220 or self -organization. This is the idea that 366 00:16:32,220 --> 00:16:34,860 complex, structured patterns can emerge spontaneously 367 00:16:34,860 --> 00:16:37,740 in systems with many interacting parts without 368 00:16:37,740 --> 00:16:39,720 needing a specific blueprint for that pattern. 369 00:16:39,860 --> 00:16:42,139 Think of crystal formation or flocking birds. 370 00:16:42,419 --> 00:16:44,519 Complexity arising from simple interactions. 371 00:16:44,759 --> 00:16:47,679 How does that apply to speech sounds? Udayer 372 00:16:47,679 --> 00:16:50,500 uses computational modeling. He creates virtual 373 00:16:50,500 --> 00:16:53,360 agents, sort of like simple robots, equipped 374 00:16:53,360 --> 00:16:56,039 with simulated vocal tracks and auditory systems 375 00:16:56,039 --> 00:16:59,279 linked by neural networks. Crucially, they're 376 00:16:59,279 --> 00:17:01,360 given the ability to imitate sounds they hear 377 00:17:01,360 --> 00:17:03,700 and to adapt their imitations to get better. 378 00:17:03,759 --> 00:17:05,819 So they babble and listen to each other. Essentially, 379 00:17:05,940 --> 00:17:08,539 yes. They start by exploring their vocal possibilities, 380 00:17:08,920 --> 00:17:11,559 producing a whole range of continuous messy sounds. 381 00:17:11,980 --> 00:17:14,480 But because they are motivated to imitate each 382 00:17:14,480 --> 00:17:17,079 other, and because some vocal tract configurations 383 00:17:17,079 --> 00:17:20,039 are easier to produce consistently and easier 384 00:17:20,039 --> 00:17:22,519 to distinguish perceptually, The system starts 385 00:17:22,519 --> 00:17:25,500 to settle. Exactly. The continuous space of possible 386 00:17:25,500 --> 00:17:29,420 sounds spontaneously discretizes. Certain stable 387 00:17:29,420 --> 00:17:32,099 configurations, effectively, phone -like categories 388 00:17:32,099 --> 00:17:34,940 emerge as a tracker points in the system. The 389 00:17:34,940 --> 00:17:37,339 agents converge on using a shared, limited set 390 00:17:37,339 --> 00:17:39,279 of distinct sounds because those are the most 391 00:17:39,279 --> 00:17:41,299 stable and learnable points in the collective 392 00:17:41,299 --> 00:17:43,819 communication system. It's like the sounds crystallize 393 00:17:43,819 --> 00:17:46,279 out of the continuous noise through social interaction 394 00:17:46,279 --> 00:17:48,740 and imitation. That's a perfect analogy. It's 395 00:17:48,740 --> 00:17:51,670 a crystallization phenomenon. And the implication 396 00:17:51,670 --> 00:17:55,109 is profound. The emergence of a combinatorial 397 00:17:55,109 --> 00:17:57,410 phonemic system, which feels like a massive evolutionary 398 00:17:57,410 --> 00:18:01,430 leap, might not have required complex, genetically 399 00:18:01,430 --> 00:18:04,349 pre -programmed knowledge of phones. Instead, 400 00:18:04,630 --> 00:18:06,769 it might have resulted from a relatively small 401 00:18:06,769 --> 00:18:09,769 biological change, perhaps improving the ability 402 00:18:09,769 --> 00:18:12,589 or motivation for vocal imitation, represented 403 00:18:12,589 --> 00:18:15,829 by a parameter sigma, sigmiflu. sigma -nuttles, 404 00:18:15,869 --> 00:18:18,609 combined with the inherent self -organizing properties 405 00:18:18,609 --> 00:18:21,470 of neural systems interacting within a population. 406 00:18:21,769 --> 00:18:23,950 So the building blocks of language might have 407 00:18:23,950 --> 00:18:26,269 emerged almost as a kind of physical consequence 408 00:18:26,269 --> 00:18:29,269 of having bodies that could make sounds and brains 409 00:18:29,269 --> 00:18:32,029 that could imitate interacting socially. That's 410 00:18:32,029 --> 00:18:34,450 the intriguing possibility Udayar's work raises. 411 00:18:34,630 --> 00:18:37,049 It connects nicely, actually, with earlier functionalist 412 00:18:37,049 --> 00:18:39,750 ideas about phone systems, too. Like Liljen Krantz 413 00:18:39,750 --> 00:18:43,079 and Lindblom's work. Yes. Back in 1972, they 414 00:18:43,079 --> 00:18:45,160 showed that the vowel systems across the world's 415 00:18:45,160 --> 00:18:47,920 languages aren't random. They tend to use vowels 416 00:18:47,920 --> 00:18:50,619 that are maximally distinct acoustically, spreading 417 00:18:50,619 --> 00:18:52,619 out in the available vowel space to make them 418 00:18:52,619 --> 00:18:55,470 easier to tell apart. Udayer's model provides 419 00:18:55,470 --> 00:18:58,309 a potential mechanism for how a population might 420 00:18:58,309 --> 00:19:01,430 spontaneously converge on such an optimal, self 421 00:19:01,430 --> 00:19:03,890 -organized system. Fascinating. It shifts the 422 00:19:03,890 --> 00:19:06,670 focus from just innate blueprints to the dynamics 423 00:19:06,670 --> 00:19:08,890 of interaction and emergence. It really does. 424 00:19:09,089 --> 00:19:11,789 Let's take these ideas about structure, context, 425 00:19:12,049 --> 00:19:14,369 and reconstruction and look at an extreme case. 426 00:19:14,589 --> 00:19:17,390 Sima Ram's work on African drum languages. Yes, 427 00:19:17,410 --> 00:19:20,730 this is a fantastic real -world example of meaning 428 00:19:20,730 --> 00:19:23,250 being conveyed with incredibly reduced acoustic 429 00:19:23,250 --> 00:19:26,009 information. He studied the bandalinda people, 430 00:19:26,150 --> 00:19:29,210 who use a special two -toned wooden drum, the 431 00:19:29,210 --> 00:19:32,069 lenga, to communicate complex messages over long 432 00:19:32,069 --> 00:19:34,250 distances. Two tones, that's it. How can you 433 00:19:34,250 --> 00:19:36,349 send sentences with just two pitches? Well, they 434 00:19:36,349 --> 00:19:38,809 don't just use pitch, they also use rhythm. The 435 00:19:38,809 --> 00:19:41,190 drum language essentially mimics the tonal contour, 436 00:19:41,490 --> 00:19:44,849 the sequence of high and low tones, and the rhythmic 437 00:19:44,849 --> 00:19:47,289 pattern, syllable durations, of the original 438 00:19:47,289 --> 00:19:49,710 spoken utterance in their tonal language. But 439 00:19:49,710 --> 00:19:51,930 they lose all the vowels and consonants, the 440 00:19:51,930 --> 00:19:54,640 actual phonetic substance. Completely gone. All 441 00:19:54,640 --> 00:19:57,200 that's left is the melody and rhythm played on 442 00:19:57,200 --> 00:20:00,039 two discrete pitch levels. As you can imagine, 443 00:20:00,140 --> 00:20:03,599 this leads to massive ambiguity. Huge homophony. 444 00:20:03,660 --> 00:20:06,460 The same rhythm and tone pattern could correspond 445 00:20:06,460 --> 00:20:09,019 to dozens of different phrases with totally different 446 00:20:09,019 --> 00:20:12,839 meanings. How on earth is it intelligible? That's 447 00:20:12,839 --> 00:20:15,319 the key question. Intelligibility relies heavily 448 00:20:15,319 --> 00:20:18,480 on two things. Shared context and the internal 449 00:20:18,480 --> 00:20:20,720 structure of the messages themselves. Context 450 00:20:20,720 --> 00:20:23,539 meaning. The drum language isn't used for casual 451 00:20:23,539 --> 00:20:26,250 chat. It's reserved for a very limited set of 452 00:20:26,250 --> 00:20:29,769 specific, important social situations. Announcing 453 00:20:29,769 --> 00:20:32,849 a birth or death, warning of danger, summoning 454 00:20:32,849 --> 00:20:35,470 people, coordinating hunts, sometimes administrative 455 00:20:35,470 --> 00:20:37,849 things like tax collection. The listener already 456 00:20:37,849 --> 00:20:39,950 knows the likely topic domain. Okay, that narrows 457 00:20:39,950 --> 00:20:42,049 it down. And the message structure. The messages 458 00:20:42,049 --> 00:20:44,730 aren't just the core information. They are heavily 459 00:20:44,730 --> 00:20:47,089 padded with standardized, redundant, general 460 00:20:47,089 --> 00:20:49,809 formulas. These are well -known phrases like, 461 00:20:49,910 --> 00:20:53,799 listen carefully, or have you understood? or 462 00:20:53,799 --> 00:20:56,339 specific ways to name the sender and recipient. 463 00:20:56,619 --> 00:20:59,740 Like boilerplate text? Sort of, but encoded in 464 00:20:59,740 --> 00:21:02,819 tone and rhythm. These formulas are long, their 465 00:21:02,819 --> 00:21:05,460 melody is familiar, and they act as signposts, 466 00:21:05,460 --> 00:21:08,720 framing the crucial, potentially ambiguous, specific 467 00:21:08,720 --> 00:21:11,700 formulas that carry the new information. The 468 00:21:11,700 --> 00:21:13,880 listener recognizes the familiar patterns of 469 00:21:13,880 --> 00:21:16,019 the general formulas, which helps them parse 470 00:21:16,019 --> 00:21:18,490 the message rhythmically. So the listener isn't 471 00:21:18,490 --> 00:21:20,730 decoding phonemes. They're recognizing rhythmic 472 00:21:20,730 --> 00:21:23,210 chunks, using the known formulas as anchors, 473 00:21:23,210 --> 00:21:25,369 and then using the limited social context to 474 00:21:25,369 --> 00:21:27,089 figure out which of the possible meanings for 475 00:21:27,089 --> 00:21:30,289 the ambiguous part fits. Exactly. Aron describes 476 00:21:30,289 --> 00:21:33,549 it as a cognitive process of rhythmic segmentation 477 00:21:33,549 --> 00:21:36,170 and then a selective recombination of possibilities, 478 00:21:36,569 --> 00:21:39,329 heavily constrained by the known formulas and 479 00:21:39,329 --> 00:21:42,420 the shared sociolinguistic situation. It's a 480 00:21:42,420 --> 00:21:44,240 brilliant demonstration that meaning construction 481 00:21:44,240 --> 00:21:47,480 can succeed even with drastically impoverished 482 00:21:47,480 --> 00:21:49,960 acoustic input, as long as there's enough structural 483 00:21:49,960 --> 00:21:52,519 redundancy and shared knowledge to guide the 484 00:21:52,519 --> 00:21:54,619 listener's reconstruction. It really highlights 485 00:21:54,619 --> 00:21:57,079 how much meaning resides in the shared system 486 00:21:57,079 --> 00:21:59,700 and context, not just in the signal itself. Arum 487 00:21:59,700 --> 00:22:01,900 even compares it to a phonological system in 488 00:22:01,900 --> 00:22:04,440 its logic. He does. He sees it as synchronically 489 00:22:04,440 --> 00:22:07,000 closed, using a fixed set of elements and rules, 490 00:22:07,140 --> 00:22:09,970 but diachronically open. able to adapt over time, 491 00:22:10,109 --> 00:22:12,009 much like the phonology of a spoken language. 492 00:22:12,250 --> 00:22:14,329 It shows the fundamental need for structured, 493 00:22:14,470 --> 00:22:16,990 conventionalized systems for complex communication, 494 00:22:17,369 --> 00:22:19,450 whatever the medium. Okay, we've seen how structure 495 00:22:19,450 --> 00:22:22,210 allows meaning, even under duress. But this brings 496 00:22:22,210 --> 00:22:24,130 us back to the music versus language question. 497 00:22:24,650 --> 00:22:27,109 If they share so much hierarchical structure, 498 00:22:27,390 --> 00:22:29,750 reliance on pitch and rhythm, are they processed 499 00:22:29,750 --> 00:22:32,869 by the same brain systems? This is where neuropsychology 500 00:22:32,869 --> 00:22:35,730 comes in, right? Isabelle Peretz, Regine Kalinsky. 501 00:22:35,950 --> 00:22:38,700 Yes, their work. particularly studying patients 502 00:22:38,700 --> 00:22:40,539 with brain damage or developmental conditions 503 00:22:40,539 --> 00:22:43,960 like amusia, provides crucial evidence for dissociations 504 00:22:43,960 --> 00:22:46,700 between language and music abilities. Dissociations 505 00:22:46,700 --> 00:22:49,000 meaning you can lose one without losing the other. 506 00:22:49,200 --> 00:22:52,359 Exactly. If music processing simply relied entirely 507 00:22:52,359 --> 00:22:54,940 on language circuits, damage affecting language 508 00:22:54,940 --> 00:22:57,240 should proportionally affect music and vice versa. 509 00:22:57,420 --> 00:22:59,779 But that's often not what we see. What kind of 510 00:22:59,779 --> 00:23:01,819 dissociations do they find? Well, you can find 511 00:23:01,819 --> 00:23:04,279 patients who, after a stroke, for instance, lose 512 00:23:04,279 --> 00:23:06,380 their ability to perceive or produce melodies. 513 00:23:07,019 --> 00:23:09,559 They can't recognize tunes or sing pitches correctly, 514 00:23:09,839 --> 00:23:12,059 but their perception and production of rhythm 515 00:23:12,059 --> 00:23:15,319 might be perfectly fine. Or the opposite, intact 516 00:23:15,319 --> 00:23:18,839 melody, impaired rhythm. So pitch and rhythm, 517 00:23:18,980 --> 00:23:21,500 key musical elements, can be independently impaired. 518 00:23:21,880 --> 00:23:24,759 Yes, suggesting they rely on at least partially 519 00:23:24,759 --> 00:23:27,849 distinct neural resources. And the classic, really 520 00:23:27,849 --> 00:23:31,089 striking example is the aphasic singer. Absolutely. 521 00:23:31,309 --> 00:23:34,650 Many individuals with severe non -fluent aphasia, 522 00:23:34,670 --> 00:23:36,650 meaning they have profound difficulty producing 523 00:23:36,650 --> 00:23:39,849 spoken language, maybe can only utter a few words, 524 00:23:40,029 --> 00:23:42,950 can often still sing familiar songs quite well 525 00:23:42,950 --> 00:23:46,269 with correct melody and rhythm. Some can even 526 00:23:46,269 --> 00:23:48,549 learn new melodies accurately. Even though they 527 00:23:48,549 --> 00:23:51,390 can barely speak a sentence. Right. This strongly 528 00:23:51,390 --> 00:23:53,910 implies that the neural pathways used for planning 529 00:23:53,910 --> 00:23:56,109 and executing the complex motor sequences of 530 00:23:56,109 --> 00:23:58,710 speech are different from those used for producing 531 00:23:58,710 --> 00:24:01,809 sung melody. So this challenges the idea that 532 00:24:01,809 --> 00:24:04,130 music is just piggybacking on language modules. 533 00:24:04,250 --> 00:24:07,349 If there is some specific music module or specialization, 534 00:24:07,529 --> 00:24:09,750 what does it seem to handle? Peretz's research, 535 00:24:10,009 --> 00:24:12,250 especially on congenital amusia people who are 536 00:24:12,250 --> 00:24:15,950 tone deaf from birth, points towards pitch processing. 537 00:24:16,599 --> 00:24:18,619 specifically the perception of pitch intervals 538 00:24:18,619 --> 00:24:21,220 and contour as being a chord deficit. So people 539 00:24:21,220 --> 00:24:23,400 with amusia often have normal rhythm perception? 540 00:24:23,839 --> 00:24:27,180 Often, yes. Their primary difficulty lies in 541 00:24:27,180 --> 00:24:29,039 telling whether melodies go up or down correctly, 542 00:24:29,279 --> 00:24:32,000 or recognizing tunes they've heard before, because 543 00:24:32,000 --> 00:24:34,059 they can't process the pitch relationships accurately. 544 00:24:35,210 --> 00:24:37,349 This suggests that the fine -grained analysis 545 00:24:37,349 --> 00:24:40,130 of pitch structure, crucial for melody, might 546 00:24:40,130 --> 00:24:42,569 be a capacity more specialized for music than 547 00:24:42,569 --> 00:24:45,890 for language, which relies more on broader pitch 548 00:24:45,890 --> 00:24:48,529 contours for intonation. Okay, that's evidence 549 00:24:48,529 --> 00:24:50,930 from impairments. But what about brain imaging 550 00:24:50,930 --> 00:24:53,589 in healthy people? Don't those studies often 551 00:24:53,589 --> 00:24:55,930 show lots of overlap between language and music 552 00:24:55,930 --> 00:24:57,990 processing? They do show significant overlap, 553 00:24:58,150 --> 00:25:01,509 yes. Functional neuroimaging studies, like fMRI, 554 00:25:01,809 --> 00:25:04,630 often reveal activation in similar brain regions, 555 00:25:04,930 --> 00:25:07,950 auditory cortex, frontal areas, parietal areas 556 00:25:07,950 --> 00:25:10,029 when people listen to speech or music. So how 557 00:25:10,029 --> 00:25:11,890 do we square that with the dissociation evidence? 558 00:25:12,150 --> 00:25:14,509 The overlap isn't surprising. Both language and 559 00:25:14,509 --> 00:25:17,029 music involve complex sound perception, sequencing, 560 00:25:17,250 --> 00:25:20,490 memory, attention, maybe emotion. They naturally 561 00:25:20,490 --> 00:25:22,809 recruit overlapping general cognitive resources 562 00:25:22,809 --> 00:25:25,109 and auditory processing networks distributed 563 00:25:25,109 --> 00:25:28,049 across both brain hemispheres. But it's not total 564 00:25:28,049 --> 00:25:31,930 overlap. No. Crucially, almost all neuroimaging 565 00:25:31,930 --> 00:25:34,390 studies that carefully compare speech and music 566 00:25:34,390 --> 00:25:37,849 also report distinct areas of activation. There 567 00:25:37,849 --> 00:25:40,210 are regions that respond more strongly to speech 568 00:25:40,210 --> 00:25:43,430 than to music, and vice versa. Can you give an 569 00:25:43,430 --> 00:25:45,529 example? One finding is that singing familiar 570 00:25:45,529 --> 00:25:48,529 songs in unison with others tends to activate 571 00:25:48,529 --> 00:25:51,190 certain areas, like parts of the left inferior 572 00:25:51,190 --> 00:25:54,509 parietal lobe, more strongly than speaking familiar 573 00:25:54,509 --> 00:25:56,710 phrases in unison. And that might connect back 574 00:25:56,710 --> 00:26:00,009 to the aphasic singers. Potentially, yes. It 575 00:26:00,009 --> 00:26:02,130 suggests there might be specific neural pathways 576 00:26:02,130 --> 00:26:05,170 facilitated by singing that can sometimes bypass 577 00:26:05,170 --> 00:26:08,210 damaged speech pathways, offering a neural basis 578 00:26:08,210 --> 00:26:10,230 for why singing therapy can sometimes help people 579 00:26:10,230 --> 00:26:12,869 with non -fluid aphasia improve their vocal output. 580 00:26:13,049 --> 00:26:15,049 So the picture is one of shared resources and 581 00:26:15,049 --> 00:26:17,349 specific pathways. It seems the brain is both 582 00:26:17,349 --> 00:26:20,029 integrated and specialized, and technology is 583 00:26:20,029 --> 00:26:21,910 pushing the boundaries of manipulating these 584 00:26:21,910 --> 00:26:24,450 signals even further. Xavier Rodet's work at 585 00:26:24,450 --> 00:26:27,410 IRCM. Ah, yes, IRCM in Paris, the Institute for 586 00:26:27,410 --> 00:26:29,230 Research and Coordination in Acoustics Music. 587 00:26:29,430 --> 00:26:32,009 They've been at the forefront of analyzing, transforming, 588 00:26:32,269 --> 00:26:34,630 and synthesizing sound, especially the human 589 00:26:34,630 --> 00:26:38,049 voice, for decades. Using tools like SuperVP, 590 00:26:38,269 --> 00:26:41,980 IRCM corpus tools, what can they actually do 591 00:26:41,980 --> 00:26:44,720 with voice signals? A remarkable range of things. 592 00:26:44,940 --> 00:26:47,759 Beyond basic analysis, they can perform sophisticated 593 00:26:47,759 --> 00:26:50,509 transformations. They can change perceived speaker 594 00:26:50,509 --> 00:26:53,130 identity, make a female voice sound male or vice 595 00:26:53,130 --> 00:26:55,769 versa, while keeping it sounding relatively natural. 596 00:26:55,990 --> 00:26:58,589 Modify expressivity, too. Make a flat voice sound 597 00:26:58,589 --> 00:27:01,529 emotional. Yes. Manipulating parameters related 598 00:27:01,529 --> 00:27:04,329 to pitch contour, timing, and spectral characteristics 599 00:27:04,329 --> 00:27:07,750 to alter the perceived emotional tone. They've 600 00:27:07,750 --> 00:27:09,630 even used these tools for practical applications 601 00:27:09,630 --> 00:27:11,740 in film. You mentioned correcting pronunciation. 602 00:27:12,319 --> 00:27:14,660 Right, the example of adjusting Gerard Depardieu's 603 00:27:14,660 --> 00:27:17,500 French -accented English for the film Vatel. 604 00:27:17,680 --> 00:27:20,240 They could subtly modify the formant structures 605 00:27:20,240 --> 00:27:23,339 associated with specific phonemes to make his 606 00:27:23,339 --> 00:27:25,799 pronunciation closer to a target native English 607 00:27:25,799 --> 00:27:28,000 speaker without him having to re -record lines 608 00:27:28,000 --> 00:27:30,799 endlessly. And the really famous example is the 609 00:27:30,799 --> 00:27:33,380 film Farinelli, recreating a castrato voice. 610 00:27:33,819 --> 00:27:36,779 Exactly. That was a huge challenge. Castrati 611 00:27:36,779 --> 00:27:40,400 had the lung power of adult males, but... Due 612 00:27:40,400 --> 00:27:43,019 to castration before puberty, the small vocal 613 00:27:43,019 --> 00:27:45,920 folds and larynx of a boy allowing them an incredibly 614 00:27:45,920 --> 00:27:49,039 high range and unique timbre. Since that voice 615 00:27:49,039 --> 00:27:51,400 type doesn't exist today, they had to synthesize 616 00:27:51,400 --> 00:27:54,880 it. Using SuperVP, they meticulously analyzed 617 00:27:54,880 --> 00:27:57,420 recordings of a female soprano for the high notes 618 00:27:57,420 --> 00:28:00,059 and agility and a male countertenor for different 619 00:28:00,059 --> 00:28:03,440 quality and lower range aspects. They then morphed 620 00:28:03,440 --> 00:28:05,359 and blended characteristics of the two voices' 621 00:28:05,480 --> 00:28:08,700 pitch, timbre, spectral envelope, vibrato frame 622 00:28:08,700 --> 00:28:11,500 by frame to create a seamless synthetic voice 623 00:28:11,500 --> 00:28:14,319 that convincingly evoked the legendary sound 624 00:28:14,319 --> 00:28:17,559 of Farinelli across his huge range. Wow. That 625 00:28:17,559 --> 00:28:19,819 shows incredible control over the acoustic components 626 00:28:19,819 --> 00:28:22,380 of voice quality. And they can even turn speech 627 00:28:22,380 --> 00:28:25,140 into song. Yes, that's another fascinating capability. 628 00:28:25,660 --> 00:28:27,660 Techniques like shape invariant modification 629 00:28:27,660 --> 00:28:30,440 in SuperVP allow them to take a recording of 630 00:28:30,440 --> 00:28:32,599 someone speaking and dramatically transpose the 631 00:28:32,599 --> 00:28:34,799 pitch upwards, maybe two or three octaves, and 632 00:28:34,799 --> 00:28:36,559 stretch the durations of syllables, effectively 633 00:28:36,559 --> 00:28:39,450 transforming it. into a sung melody while keeping 634 00:28:39,450 --> 00:28:41,630 the original voice quality to a large extent 635 00:28:41,630 --> 00:28:44,430 yes it demonstrates that the core acoustic components 636 00:28:44,430 --> 00:28:47,650 the fundamental frequency pitch the duration 637 00:28:47,650 --> 00:28:50,130 and spectral envelope related to formance and 638 00:28:50,130 --> 00:28:52,690 timber can be manipulated somewhat independently 639 00:28:52,690 --> 00:28:55,829 you can impose a musical pitch and rhythm onto 640 00:28:55,829 --> 00:28:58,819 a spoken utterance's spectral signature It really 641 00:28:58,819 --> 00:29:01,559 blurs the line, showing how malleable vocal sound 642 00:29:01,559 --> 00:29:04,039 is when you have the right analytical and synthesis 643 00:29:04,039 --> 00:29:07,240 tools. It underscores that both speech and song 644 00:29:07,240 --> 00:29:09,539 are built from the same fundamental acoustic 645 00:29:09,539 --> 00:29:12,380 ingredients, even if our brains and cultures 646 00:29:12,380 --> 00:29:14,799 have specialized them for different communicative 647 00:29:14,799 --> 00:29:17,380 functions. So wrapping this all up, this deep 648 00:29:17,380 --> 00:29:19,579 dive really shows that getting from sound waves 649 00:29:19,579 --> 00:29:22,279 to complex meaning, whether it's the precise 650 00:29:22,279 --> 00:29:24,259 meaning of a sentence or the evocative meaning 651 00:29:24,259 --> 00:29:26,700 of music, is an incredibly intricate process. 652 00:29:27,079 --> 00:29:29,259 It absolutely is. It relies on sophisticated, 653 00:29:29,599 --> 00:29:32,599 likely innate auditory analysis mechanisms like 654 00:29:32,599 --> 00:29:35,500 reconstructing that missing F0. Then you have 655 00:29:35,500 --> 00:29:37,960 these highly structured abstract generative rules 656 00:29:37,960 --> 00:29:40,859 like fusion and movement and language. And layered 657 00:29:40,859 --> 00:29:43,059 on top of that are the shared cultural conventions 658 00:29:43,059 --> 00:29:45,779 and contextual knowledge. like we saw with the 659 00:29:45,779 --> 00:29:48,279 drum language. And the key takeaway seems to 660 00:29:48,279 --> 00:29:50,859 be that our perception isn't just a mirror of 661 00:29:50,859 --> 00:29:53,240 acoustic reality. Not at all. It's an act of 662 00:29:53,240 --> 00:29:56,140 construction. Like Purkinje said about vision 663 00:29:56,140 --> 00:29:59,970 reluder, the areas of the senses are the truths 664 00:29:59,970 --> 00:30:03,230 of perception, the distortions, the constructions, 665 00:30:03,230 --> 00:30:06,109 the filling in. That is how our brain creates 666 00:30:06,109 --> 00:30:09,829 a stable, meaningful, perceptual world from ambiguous 667 00:30:09,829 --> 00:30:13,470 sensory input. The illusion is the reality of 668 00:30:13,470 --> 00:30:15,920 perception. For listeners who found specific 669 00:30:15,920 --> 00:30:18,279 aspects fascinating and want to explore further, 670 00:30:18,539 --> 00:30:20,880 let's quickly recap some of the key theoretical 671 00:30:20,880 --> 00:30:23,500 anchor points and names. Sure, for the absolute 672 00:30:23,500 --> 00:30:25,859 basics of sound, pitch, and timbre perception, 673 00:30:26,180 --> 00:30:28,940 going back to Helmholtz's theory Physiologique 674 00:30:28,940 --> 00:30:31,279 de la Musique, perhaps via Bouveresse's analysis, 675 00:30:31,579 --> 00:30:34,019 is fundamental. For how fern systems might emerge. 676 00:30:34,359 --> 00:30:36,740 Pierre -Yves Zoué's work on auto -organization 677 00:30:36,740 --> 00:30:39,960 is cutting edge, but it builds on earlier functionalist 678 00:30:39,960 --> 00:30:42,319 insights, like those from Liljen Krantz and Lindblom 679 00:30:42,319 --> 00:30:45,349 on maximizing perception. contrast in vowel systems. 680 00:30:45,529 --> 00:30:47,349 And the architecture of grammar. That really 681 00:30:47,349 --> 00:30:49,890 takes you into the Chomsky tradition, specifically 682 00:30:49,890 --> 00:30:52,609 the principles and parameters model, with Luigi 683 00:30:52,609 --> 00:30:54,769 Rithi being a key figure for understanding the 684 00:30:54,769 --> 00:30:57,329 detailed analysis of syntactic structures and 685 00:30:57,329 --> 00:30:59,650 variation, like that V to T movement parameter. 686 00:31:00,069 --> 00:31:03,609 Finally, the music language distinction. Isabel 687 00:31:03,609 --> 00:31:07,210 Peretz's extensive work on amusia and the neuropsychology 688 00:31:07,210 --> 00:31:10,109 of music is indispensable there, really probing 689 00:31:10,109 --> 00:31:12,109 the questions of modularity and specialization. 690 00:31:12,559 --> 00:31:14,900 We started asking if music is auditory cheesecake 691 00:31:14,900 --> 00:31:17,700 or something more profound. The evidence seems 692 00:31:17,700 --> 00:31:21,759 to point towards shared roots, but distinct specializations 693 00:31:21,759 --> 00:31:24,599 in the brain. Yes, they share the basic physics 694 00:31:24,599 --> 00:31:28,400 of sound and early auditory processing, but diverge 695 00:31:28,400 --> 00:31:30,680 in terms of which features they primarily exploit. 696 00:31:31,019 --> 00:31:34,000 Dynamic formants and syntax for language, precise 697 00:31:34,000 --> 00:31:36,750 pitch intervals and contour for music. and likely 698 00:31:36,750 --> 00:31:38,930 rely on partially separate neural pathways for 699 00:31:38,930 --> 00:31:41,349 production and fine -grained perception. But 700 00:31:41,349 --> 00:31:43,170 perhaps the most mind -bending idea comes from 701 00:31:43,170 --> 00:31:45,650 combining the auto -organization concept with 702 00:31:45,650 --> 00:31:47,849 insights about infant cognition, like those from 703 00:31:47,849 --> 00:31:50,710 Dane Lamberts, showing brain readiness for hierarchical 704 00:31:50,710 --> 00:31:53,809 structure even before babbling. Right. It suggests 705 00:31:53,809 --> 00:31:56,250 that maybe the magic ingredient wasn't the evolution 706 00:31:56,250 --> 00:31:59,650 of incredibly complex dedicated language or music 707 00:31:59,650 --> 00:32:01,869 modules from scratch. Maybe it was a relatively 708 00:32:01,869 --> 00:32:05,269 small tweak like enhancing our capacity for vocal 709 00:32:05,269 --> 00:32:07,690 imitation and learning within a social context. 710 00:32:07,970 --> 00:32:10,829 Which was just enough to unleash the brain's 711 00:32:10,829 --> 00:32:15,049 inherent ability to self -organize complex combinatorial 712 00:32:15,049 --> 00:32:17,710 hierarchical systems from the bottom up. It's 713 00:32:17,710 --> 00:32:20,250 a powerful thought. That all this richness of 714 00:32:20,250 --> 00:32:22,569 meaning we create might have bootstrapped itself 715 00:32:22,569 --> 00:32:25,009 from relatively simple changes leveraging the 716 00:32:25,009 --> 00:32:27,170 fundamental properties of neural matter. Which 717 00:32:27,170 --> 00:32:29,930 leaves us, and you the listener, with a final 718 00:32:29,930 --> 00:32:32,970 thought. As we increasingly interact with new 719 00:32:32,970 --> 00:32:36,190 kinds of minds AI systems trained on vast amounts 720 00:32:36,190 --> 00:32:38,890 of our language and cultural data, systems capable 721 00:32:38,890 --> 00:32:41,849 of sophisticated imitation and generation, what 722 00:32:41,849 --> 00:32:44,470 new forms of complexity, perhaps entirely non 723 00:32:44,470 --> 00:32:46,569 -linguistic or non -musical as we currently understand 724 00:32:46,569 --> 00:32:49,589 them, might be starting to self -organize? What 725 00:32:49,589 --> 00:32:51,549 emergent structures are we accidentally kicking 726 00:32:51,549 --> 00:32:54,130 off just by making small adaptive shifts in our 727 00:32:54,130 --> 00:32:56,470 technological architecture? Something to ponder. 728 00:32:57,329 --> 00:33:00,150 Until the next deep dive, keep listening not 729 00:33:00,150 --> 00:33:02,609 just to the sounds but to the structures hidden 730 00:33:02,609 --> 00:33:03,150 within them.