1 00:00:00,000 --> 00:00:02,960 Welcome back to the Deep Dive. Today, we're tackling 2 00:00:02,960 --> 00:00:06,500 a huge figure, Noam Chomsky. Now, lots of people 3 00:00:06,500 --> 00:00:08,519 know him for his political work, obviously, which 4 00:00:08,519 --> 00:00:10,960 is, you know, still incredibly relevant. Definitely. 5 00:00:11,039 --> 00:00:13,339 His political analysis as well. It's everywhere. 6 00:00:13,519 --> 00:00:16,420 Right. But our focus today, our mission is really 7 00:00:16,420 --> 00:00:20,269 zeroing in on his linguistics. The ideas that 8 00:00:20,269 --> 00:00:24,089 just completely shook up how we study human language. 9 00:00:24,230 --> 00:00:26,489 We're going deep into that revolution. Yeah, 10 00:00:26,530 --> 00:00:28,350 it's a fascinating story because Chomsky's work 11 00:00:28,350 --> 00:00:30,890 wasn't just like a better grammar book for English. 12 00:00:30,949 --> 00:00:32,990 It was a fundamental break. We want to walk you 13 00:00:32,990 --> 00:00:36,229 through how his theories developed from, you 14 00:00:36,229 --> 00:00:38,409 know, the early days of generative grammar, which 15 00:00:38,409 --> 00:00:41,329 was maybe a bit messy. Yeah, complex. Well, yeah, 16 00:00:41,369 --> 00:00:43,369 complex, but brilliant. Yeah. All the way to 17 00:00:43,369 --> 00:00:47,789 the much leaner minimalist program. And the key 18 00:00:47,789 --> 00:00:49,630 thing, the narrative we're following is this 19 00:00:49,630 --> 00:00:52,149 shift he forced, moving the study of languages, 20 00:00:52,429 --> 00:00:55,630 you know, English, Swahili, whatever, away from 21 00:00:55,630 --> 00:00:57,710 being just these external things, these cultural 22 00:00:57,710 --> 00:00:59,829 objects. And putting it where? Squarely into 23 00:00:59,829 --> 00:01:02,719 biology, into the mind. Cognitive science, essentially. 24 00:01:02,920 --> 00:01:05,099 So it wasn't just about sentence diagrams anymore. 25 00:01:05,260 --> 00:01:07,159 It was about the brain, about what makes us human. 26 00:01:07,939 --> 00:01:10,040 Studying language like you'd study vision or 27 00:01:10,040 --> 00:01:13,060 something, a natural object. Exactly. That was 28 00:01:13,060 --> 00:01:16,439 the radical part. Before Chomsky, you had behaviorism 29 00:01:16,439 --> 00:01:19,019 dominating. Language was seen as just learned 30 00:01:19,019 --> 00:01:22,420 habits, stimulus response, things you could observe 31 00:01:22,420 --> 00:01:24,959 externally. Like learning multiplication tables 32 00:01:24,959 --> 00:01:27,560 or something. Sort of, yeah. Chomsky said no. 33 00:01:28,040 --> 00:01:30,700 That misses the whole point. We need a theory 34 00:01:30,700 --> 00:01:34,060 of the human linguistic capacity itself, an internal 35 00:01:34,060 --> 00:01:36,900 thing, a mental faculty that grows, kind of like 36 00:01:36,900 --> 00:01:39,180 an organ following a biological script. Okay, 37 00:01:39,219 --> 00:01:41,640 and the core of this biological thing, the absolute 38 00:01:41,640 --> 00:01:44,379 unique feature of human language, is what he 39 00:01:44,379 --> 00:01:48,099 called the basic property. Yes, that's the bedrock, 40 00:01:48,180 --> 00:01:50,620 the basic property. It sounds simple, but it's 41 00:01:50,620 --> 00:01:53,159 profound. It says, human language is a finite 42 00:01:53,159 --> 00:01:55,959 system. We have a finite number of words, a finite 43 00:01:55,959 --> 00:01:59,000 brain. Right. Can't store infinite things. Exactly. 44 00:01:59,140 --> 00:02:02,599 And here's the magic. It has infinite generative 45 00:02:02,599 --> 00:02:05,159 capacity. It can create a limitless number of 46 00:02:05,159 --> 00:02:07,700 sentences, sentences never heard before. And 47 00:02:07,700 --> 00:02:09,800 crucially, these sentences aren't just strings 48 00:02:09,800 --> 00:02:13,060 of words. They have hierarchical structure. Structure. 49 00:02:13,099 --> 00:02:15,620 And each one gets interpreted. Right. Each structured 50 00:02:15,620 --> 00:02:18,759 expression gets shipped off to two interface 51 00:02:18,759 --> 00:02:21,400 systems. Think of them as output channels, kind 52 00:02:21,400 --> 00:02:23,840 of. What are they? One is the sensory motor system, 53 00:02:24,099 --> 00:02:27,139 the SM interface. That's the part that deals 54 00:02:27,139 --> 00:02:30,500 with actually producing the language, the sounds 55 00:02:30,500 --> 00:02:33,620 of speech, or the signs in sign language, the 56 00:02:33,620 --> 00:02:36,759 physical output. Got it. Externalizing. Externalizing, 57 00:02:36,759 --> 00:02:38,699 exactly. The other is the conceptual intentional 58 00:02:38,699 --> 00:02:41,939 system, this CI interface. This is where meaning 59 00:02:41,939 --> 00:02:45,219 happens, thought, planning, understanding. So 60 00:02:45,219 --> 00:02:47,300 the whole language machine, whatever it is, has 61 00:02:47,300 --> 00:02:49,539 to satisfy both sound and meaning. Precisely. 62 00:02:49,539 --> 00:02:51,240 Those are the constraints it works under. It 63 00:02:51,240 --> 00:02:52,879 has to generate structures that both systems 64 00:02:52,879 --> 00:02:55,020 can interpret. Which immediately makes you think 65 00:02:55,020 --> 00:02:57,219 of it as, yeah, a biological computation, not 66 00:02:57,219 --> 00:02:59,919 just a handy tool we invented. Okay, so for this 67 00:02:59,919 --> 00:03:02,819 deep dive, we're drawing on analyses that cover, 68 00:03:02,979 --> 00:03:06,000 well, the whole history. The syntax models, minimalism, 69 00:03:06,120 --> 00:03:08,979 the biology, even the philosophical debates. 70 00:03:09,199 --> 00:03:12,400 Right, let's get into it. Okay, so where to start? 71 00:03:12,740 --> 00:03:15,860 Probably with the foundations. Generative grammar, 72 00:03:16,139 --> 00:03:20,199 aspects of the theory of syntax, 1965. That was 73 00:03:20,199 --> 00:03:22,560 the big one. Right. And he kicked off by pointing 74 00:03:22,560 --> 00:03:24,400 out things about language that the older theories 75 00:03:24,400 --> 00:03:27,139 just couldn't handle, puzzles they couldn't solve. 76 00:03:27,300 --> 00:03:30,520 Exactly. Three really striking properties. They 77 00:03:30,520 --> 00:03:33,120 showed why the old data collecting behaviorist 78 00:03:33,120 --> 00:03:35,419 approach just wasn't enough. Okay, what's the 79 00:03:35,419 --> 00:03:38,199 first one? Infinitude. Simple, but think about 80 00:03:38,199 --> 00:03:41,919 it. There's no longest sentence ever in any language. 81 00:03:42,120 --> 00:03:44,060 Right. You can always add something. Mary thinks 82 00:03:44,060 --> 00:03:46,000 that John knows that. Sarah hopes that Peter 83 00:03:46,000 --> 00:03:48,219 believes. Yeah, you can go on forever. It seems 84 00:03:48,219 --> 00:03:50,240 trivial, but it means language isn't a finite 85 00:03:50,240 --> 00:03:52,800 list of sentences you memorize. It's a system 86 00:03:52,800 --> 00:03:55,219 for creating them. We understand sentences we've 87 00:03:55,219 --> 00:03:57,759 never encountered before, constantly. Okay, infinitude. 88 00:03:57,840 --> 00:04:01,300 Makes sense. What's number two? Structure. This 89 00:04:01,300 --> 00:04:03,580 is maybe less obvious, but absolutely crucial. 90 00:04:03,919 --> 00:04:06,120 Words aren't just like beads on a string, one 91 00:04:06,120 --> 00:04:08,560 after another. They clump together in meaningful 92 00:04:08,560 --> 00:04:12,180 groups. Hierarchies. Constituents, phrases. Exactly. 93 00:04:12,599 --> 00:04:14,740 Constituents. Take a phrase like unlockable. 94 00:04:15,000 --> 00:04:17,240 Is that unlockable, meaning not able to be locked 95 00:04:17,240 --> 00:04:21,519 or unable to be unlocked? The string of sounds 96 00:04:21,519 --> 00:04:24,279 is the same. But the structure, how you group 97 00:04:24,279 --> 00:04:26,000 the parts, gives you two different meanings. 98 00:04:26,360 --> 00:04:29,399 Precisely. Or his example, old men and women. 99 00:04:29,579 --> 00:04:32,579 Is it old men and women? Old men and women. Linear 100 00:04:32,579 --> 00:04:34,560 order tells you nothing. The structure is everything. 101 00:04:34,939 --> 00:04:38,079 Okay. Infinitude, structure. And the third one. 102 00:04:38,519 --> 00:04:41,240 This sounds like the trickiest. It is, in many 103 00:04:41,240 --> 00:04:44,779 ways. Displacement. This is a weird one. Words 104 00:04:44,779 --> 00:04:47,639 or whole phrases. They often feel like they belong 105 00:04:47,639 --> 00:04:49,300 somewhere else in the sentence for the meaning 106 00:04:49,300 --> 00:04:51,120 to make sense, but we actually say them somewhere 107 00:04:51,120 --> 00:04:53,259 different. Can you give an example? Sure. Think 108 00:04:53,259 --> 00:04:56,360 about a simple question. What did you see? The 109 00:04:56,360 --> 00:04:59,420 word what is clearly the object of see, right? 110 00:04:59,519 --> 00:05:01,860 You saw something. Yeah. Its meaning relates 111 00:05:01,860 --> 00:05:05,100 to the verb see. Right. Its interpretive position, 112 00:05:05,279 --> 00:05:07,899 where it gets its role, is after see, like in 113 00:05:07,899 --> 00:05:09,860 the statement you saw what. But in the question, 114 00:05:09,959 --> 00:05:11,980 we pronounce it way out at the front. What did 115 00:05:11,980 --> 00:05:15,089 you see? It's been displaced. Okay, I see. So 116 00:05:15,089 --> 00:05:17,569 the early model, generative grammar, how did 117 00:05:17,569 --> 00:05:20,110 it tackle structure first, the hierarchy thing? 118 00:05:20,310 --> 00:05:23,379 It used phrase structure rules. PS rules. These 119 00:05:23,379 --> 00:05:25,560 were simple rewrite rules, like recipes for building 120 00:05:25,560 --> 00:05:28,540 sentences. The most basic one was maybe S right 121 00:05:28,540 --> 00:05:32,680 arrow, NPVP. Sentence becomes noun phrase followed 122 00:05:32,680 --> 00:05:35,560 by verb phrase, subject and predicate. Exactly. 123 00:05:35,720 --> 00:05:38,240 Captures that fundamental division. And you could 124 00:05:38,240 --> 00:05:40,699 draw trees based on these rules. So for the very 125 00:05:40,699 --> 00:05:43,180 old man left, the rules would show that the very 126 00:05:43,180 --> 00:05:46,420 old man all hangs together as one unit, an NP. 127 00:05:46,680 --> 00:05:48,899 And that explains why rules apply to the whole 128 00:05:48,899 --> 00:05:51,339 chunk, not just the individual words. He left, 129 00:05:51,420 --> 00:05:53,120 the man left, but the very old man who lives 130 00:05:53,120 --> 00:05:55,680 down the lane left. The rule applies to the NP 131 00:05:55,680 --> 00:05:58,060 unit. Correct. That's constituency. It showed 132 00:05:58,060 --> 00:06:00,439 structure mattered. But the real breakthrough 133 00:06:00,439 --> 00:06:04,899 for handling infinitude was recursion. Ah, recursion. 134 00:06:05,079 --> 00:06:07,680 Embedding things inside themselves. Like Russian 135 00:06:07,680 --> 00:06:10,639 dolls. Exactly like Russian dolls. That's the 136 00:06:10,639 --> 00:06:14,740 engine of infinity. You add a PS rule like VPWriteROVSS, 137 00:06:14,879 --> 00:06:17,839 a verb phrase can contain a verb and a whole 138 00:06:17,839 --> 00:06:21,279 sentence. So... The woman knows as the man left. 139 00:06:21,439 --> 00:06:24,139 Right. And then you can do it again. The child 140 00:06:24,139 --> 00:06:27,379 thinks as the woman knows as the man left. You 141 00:06:27,379 --> 00:06:29,360 can just keep embedding sentences within sentences 142 00:06:29,360 --> 00:06:32,180 using that one recursive rule. So a finite set 143 00:06:32,180 --> 00:06:34,420 of rules, one of which loops back on itself, 144 00:06:34,620 --> 00:06:36,779 it gives you infinite possible structures. That 145 00:06:36,779 --> 00:06:39,560 was the insight. Finite means infinite output. 146 00:06:39,660 --> 00:06:42,139 It was elegant. It showed how a biological system 147 00:06:42,139 --> 00:06:44,899 with limited resources could, in principle, achieve 148 00:06:44,899 --> 00:06:48,360 unbounded creativity. Okay, so PS rules plus 149 00:06:48,360 --> 00:06:50,300 recursion, handle structure, and infinitude. 150 00:06:50,500 --> 00:06:53,000 What about displacement, the words moving around? 151 00:06:53,279 --> 00:06:55,420 That required a different kind of rule, transformational 152 00:06:55,420 --> 00:06:57,519 movement rules. The idea was you had two levels. 153 00:06:57,660 --> 00:07:00,379 First, the PS rules generated an underlying structure, 154 00:07:00,560 --> 00:07:02,720 the deep structure, where everything was in its 155 00:07:02,720 --> 00:07:05,060 logical or interpretive position. So you did 156 00:07:05,060 --> 00:07:07,339 see what would be the deep structure for the 157 00:07:07,339 --> 00:07:09,959 question. Kind of, yeah. That's where what gets 158 00:07:09,959 --> 00:07:13,100 its meaning as the object. then a transformational 159 00:07:13,100 --> 00:07:15,519 rule would move what to the front to create the 160 00:07:15,519 --> 00:07:17,920 surface structure? The sentence we actually say. 161 00:07:18,060 --> 00:07:20,519 What did you see? And back then, in the 60s, 162 00:07:20,519 --> 00:07:22,399 it seemed like this movement could go quite far. 163 00:07:22,639 --> 00:07:24,680 It did look like that initially. People called 164 00:07:24,680 --> 00:07:26,860 it unbounded movement. You could have sentences 165 00:07:26,860 --> 00:07:29,620 like, Which problemy did you say that Mary claimed 166 00:07:29,620 --> 00:07:32,899 that Bill thinks that John will solve? The which 167 00:07:32,899 --> 00:07:35,839 problem seems to leap across multiple clauses 168 00:07:35,839 --> 00:07:38,560 from its original spot after solves. That sounds 169 00:07:38,560 --> 00:07:41,579 computationally messy. If things can just jump 170 00:07:41,579 --> 00:07:43,920 anywhere, how does the system keep track? And 171 00:07:43,920 --> 00:07:45,740 surely not all possible movements are allowed. 172 00:07:46,160 --> 00:07:48,980 Exactly right. That intuition led to one of the 173 00:07:48,980 --> 00:07:51,720 most important discoveries in the field, locality. 174 00:07:51,980 --> 00:07:54,860 Movement isn't unbounded at all. It's heavily 175 00:07:54,860 --> 00:07:57,800 restricted. There are barriers, or islands, as 176 00:07:57,800 --> 00:08:00,379 John Ross famously called them in 67. Islands, 177 00:08:00,480 --> 00:08:02,439 like structures you can't move things out of. 178 00:08:02,699 --> 00:08:04,779 Precisely. They're configurations that block 179 00:08:04,779 --> 00:08:08,040 movement. For instance, the complex NP constraint. 180 00:08:08,879 --> 00:08:11,540 You can't pull a who word out of a clause that's 181 00:08:11,540 --> 00:08:14,519 modifying a noun. Give me an example. Okay. You 182 00:08:14,519 --> 00:08:17,139 can say, Bill believes the claim that John solved 183 00:08:17,139 --> 00:08:20,220 the problem. Fine. But try to question the problem. 184 00:08:20,540 --> 00:08:22,759 Which problem does Bill believe the claim that 185 00:08:22,759 --> 00:08:26,029 John solved? Sounds terrible, right? Yeah, definitely 186 00:08:26,029 --> 00:08:28,149 ungrammatical. That's because the phrase, the 187 00:08:28,149 --> 00:08:30,589 claim that John solved the problem, is a complex 188 00:08:30,589 --> 00:08:33,470 NP, and it acts like an island. You can't extract 189 00:08:33,470 --> 00:08:35,830 from it. Discovering these islands showed that 190 00:08:35,830 --> 00:08:38,759 movement, while real, operates under very strict 191 00:08:38,759 --> 00:08:41,919 locality conditions. The big question became 192 00:08:41,919 --> 00:08:45,320 why those specific constraints exist, what deeper 193 00:08:45,320 --> 00:08:47,580 principle explains them. Okay, that makes sense. 194 00:08:47,639 --> 00:08:49,980 Constraints point to underlying principles. Now, 195 00:08:50,000 --> 00:08:51,700 before we jump ahead to minimalism, you mentioned 196 00:08:51,700 --> 00:08:54,559 the architecture changed, specifically the role 197 00:08:54,559 --> 00:08:57,059 of the lexicon, the dictionary. This was a big 198 00:08:57,059 --> 00:09:00,200 shift from aspects to that 1970 paper, Remarks 199 00:09:00,200 --> 00:09:03,120 on Nominalization. Huge shift. And it follows 200 00:09:03,120 --> 00:09:05,639 the same logic. Yeah. Push complexity out of 201 00:09:05,639 --> 00:09:08,179 the syntax. Keep the computational rules simple 202 00:09:08,179 --> 00:09:10,600 and general. The lexicon is where you store all 203 00:09:10,600 --> 00:09:12,559 the messy, unpredictable stuff about individual 204 00:09:12,559 --> 00:09:15,340 words. Like what? Like which verb takes an object. 205 00:09:15,419 --> 00:09:17,899 Transitive, which doesn't end transitive. Which 206 00:09:17,899 --> 00:09:19,860 needs two objects. Which needs a prepositional 207 00:09:19,860 --> 00:09:22,539 phrase. All those quirky requirements. Before 208 00:09:22,539 --> 00:09:25,419 aspects, grammars had incredibly complex PS rules 209 00:09:25,419 --> 00:09:28,620 with subclasses for verbs like V transitive, 210 00:09:28,700 --> 00:09:30,779 V intransitive. Right, cluttering up the syntax. 211 00:09:31,100 --> 00:09:33,179 Totally. The move was to simplify the syntax 212 00:09:33,179 --> 00:09:36,840 rules, maybe just VP, right arrow, dollar, NPS, 213 00:09:36,960 --> 00:09:40,120 allowing for an optional noun phrase object and 214 00:09:40,120 --> 00:09:42,820 an optional sentence object. And then you store 215 00:09:42,820 --> 00:09:45,899 the specific needs of each verb in its lexical 216 00:09:45,899 --> 00:09:49,440 entry. Hit needs an NP. Sleep needs nothing. 217 00:09:49,600 --> 00:09:52,549 Put needs an NP and a PP. So the syntax is the 218 00:09:52,549 --> 00:09:54,610 general blueprint. The lexicon holds the specific 219 00:09:54,610 --> 00:09:57,289 instructions for each word. Much cleaner. Much 220 00:09:57,289 --> 00:10:00,470 cleaner. And remarks on nominalization push this 221 00:10:00,470 --> 00:10:03,309 lexicalist approach even further. It looked at 222 00:10:03,309 --> 00:10:05,769 pairs like the verb destroy and the noun destruction 223 00:10:05,769 --> 00:10:08,409 or recite and recital. And argued they weren't 224 00:10:08,409 --> 00:10:10,509 related by a transformational rule. Like you 225 00:10:10,509 --> 00:10:13,049 don't derive destruction from destroy syntactically. 226 00:10:13,190 --> 00:10:16,409 Exactly. The argument was that the relationship 227 00:10:16,409 --> 00:10:19,080 had to be lexical. They're listed separately 228 00:10:19,080 --> 00:10:21,700 in the lexicon, maybe connected by a rule within 229 00:10:21,700 --> 00:10:24,360 the lexicon, but not by a syntactic transformation. 230 00:10:25,019 --> 00:10:27,779 Why? Because deriving the noun from the verb 231 00:10:27,779 --> 00:10:31,659 phrase messes things up too much. How so? Well, 232 00:10:31,720 --> 00:10:33,940 look it. The enemy destroyed the city. Clear 233 00:10:33,940 --> 00:10:36,440 agent, clear patient. But in the enemy's destruction 234 00:10:36,440 --> 00:10:38,820 of the city, the grammatical relations change, 235 00:10:39,080 --> 00:10:41,159 the verbiness is gone, the meaning isn't always 236 00:10:41,159 --> 00:10:44,460 exactly parallel. Trying to derive one from the 237 00:10:44,460 --> 00:10:46,919 other with a transformation would violate fundamental 238 00:10:46,919 --> 00:10:49,200 principles that were emerging. Principles like? 239 00:10:49,299 --> 00:10:51,059 Like the projection principle. The idea that 240 00:10:51,059 --> 00:10:54,059 lexical properties, like needing an object, must 241 00:10:54,059 --> 00:10:57,120 be represented at every level of syntax. Transforming 242 00:10:57,120 --> 00:10:59,639 a verb into a noun inherently changes those properties. 243 00:10:59,879 --> 00:11:02,639 Or the inclusiveness condition, which later became 244 00:11:02,639 --> 00:11:05,200 key in minimalism. What's that? Basically, the 245 00:11:05,200 --> 00:11:07,899 syntax can't just invent new features or categories 246 00:11:07,899 --> 00:11:10,700 out of thin air. It can only combine and rearrange 247 00:11:10,700 --> 00:11:13,519 features that come from the lexicon. Deriving 248 00:11:13,519 --> 00:11:17,919 destruction, a noun, from destroy, a verb, seems 249 00:11:17,919 --> 00:11:20,059 to violate that you're adding noun -ness somehow. 250 00:11:20,299 --> 00:11:22,879 So the conclusion was, these relationships are 251 00:11:22,879 --> 00:11:26,159 stored, not computed by the main syntax. Keep 252 00:11:26,159 --> 00:11:29,000 the syntax pure and general. Right. Offload the 253 00:11:29,000 --> 00:11:30,960 irregularities to the lexicon. Okay, so that 254 00:11:30,960 --> 00:11:33,850 sets the stage for the next big leap. Towards 255 00:11:33,850 --> 00:11:36,610 even more simplicity. Exactly. Towards minimalism. 256 00:11:36,610 --> 00:11:39,090 Even with these refinements, the model that developed 257 00:11:39,090 --> 00:11:41,750 government and binding theory, or GB, in the 258 00:11:41,750 --> 00:11:43,929 80s was still pretty intricate. Lots of different 259 00:11:43,929 --> 00:11:47,110 modules, levels of representation, specific conditions. 260 00:11:47,409 --> 00:11:49,809 A bit Rube Goldberg -esque. Ah, maybe a little. 261 00:11:49,909 --> 00:11:52,350 Lots of moving parts. And that's where the minimalist 262 00:11:52,350 --> 00:11:55,190 program MP comes in, starting in the early 90s. 263 00:11:55,230 --> 00:11:57,090 It wasn't just about finding a better theory. 264 00:11:57,230 --> 00:11:59,690 It was driven by this deeper, almost philosophical 265 00:11:59,690 --> 00:12:03,070 goal. Chomsky called it the Galilean style. The 266 00:12:03,070 --> 00:12:06,269 Galilean style, seeking perfection, like finding 267 00:12:06,269 --> 00:12:09,149 the simplest mathematical forms in nature. Precisely. 268 00:12:09,289 --> 00:12:12,870 The MP asks, assuming the language faculty is 269 00:12:12,870 --> 00:12:15,029 a product of natural law, like other biological 270 00:12:15,029 --> 00:12:18,470 systems, could it be a perfect solution? Meaning, 271 00:12:18,669 --> 00:12:21,750 does it use only the absolute minimum conceptual 272 00:12:21,750 --> 00:12:24,809 machinery necessary to do its job? The job being? 273 00:12:24,970 --> 00:12:27,129 Connecting sound and meaning. Yeah. The SM and 274 00:12:27,129 --> 00:12:29,789 CI interfaces. Right. So the guiding idea became 275 00:12:29,789 --> 00:12:33,440 the strong minimalist thesis, SMT. This is the 276 00:12:33,440 --> 00:12:36,120 bold hypothesis that language is essentially 277 00:12:36,120 --> 00:12:39,000 a perfect system. It has only what's needed to 278 00:12:39,000 --> 00:12:41,320 satisfy the interface conditions and nothing 279 00:12:41,320 --> 00:12:44,059 extra. No redundancy, no weird artifacts of the 280 00:12:44,059 --> 00:12:45,759 theory itself. OK, that's a high bar. If you 281 00:12:45,759 --> 00:12:48,000 want absolute perfection, absolute minimum machinery, 282 00:12:48,340 --> 00:12:50,480 what's the core operation? What is the actual 283 00:12:50,480 --> 00:12:52,919 building? The candidate for that is merge, a 284 00:12:52,919 --> 00:12:55,139 single, incredibly simple operation. Merge takes 285 00:12:55,139 --> 00:12:57,120 two syntactic objects. Let's call them X and 286 00:12:57,120 --> 00:13:00,340 Y. combines them. It forms a new object, which 287 00:13:00,340 --> 00:13:02,200 is simply the set containing those two things, 288 00:13:02,320 --> 00:13:04,639 x, y. A set, not an ordered pair. Crucially, 289 00:13:04,759 --> 00:13:07,919 a set, x, y, is the same as y. Order isn't inherent 290 00:13:07,919 --> 00:13:10,320 in the basic operation. Okay, simple enough. 291 00:13:10,759 --> 00:13:12,960 But how does this one operation do everything? 292 00:13:13,580 --> 00:13:16,539 Didn't the old system need PS rules for building 293 00:13:16,539 --> 00:13:19,220 structure and movement rules for displacement? 294 00:13:19,620 --> 00:13:22,000 Yes, and that's the beauty. The grand collapse, 295 00:13:22,299 --> 00:13:25,179 as it's sometimes called. Merge, in its simplicity, 296 00:13:25,519 --> 00:13:28,740 handles both. It unifies the entire generative 297 00:13:28,740 --> 00:13:31,860 engine. How? There are basically two ways merge 298 00:13:31,860 --> 00:13:35,980 can apply. External merge, EM. This is when you 299 00:13:35,980 --> 00:13:38,200 take two separate items, maybe from the lexicon 300 00:13:38,200 --> 00:13:40,879 or already built chunks, and merge them together 301 00:13:40,879 --> 00:13:43,080 to make a bigger chunk. Like merging the and 302 00:13:43,080 --> 00:13:45,440 king to get the king, merging that with of and 303 00:13:45,440 --> 00:13:48,000 Georgia. That's how you build structure from 304 00:13:48,000 --> 00:13:50,259 the bottom up. It replaces the old PS rules. 305 00:13:50,440 --> 00:13:52,080 Okay, structure building. What about movement? 306 00:13:52,470 --> 00:13:55,330 That's internal merge, IM. This is where merge 307 00:13:55,330 --> 00:13:57,629 takes an object that's already inside the structure 308 00:13:57,629 --> 00:14:00,049 you're building and merges it again higher up 309 00:14:00,049 --> 00:14:02,190 in the structure. Ah, so you're merging something 310 00:14:02,190 --> 00:14:04,610 with a structure that already contains it. Exactly. 311 00:14:04,649 --> 00:14:07,789 And that single operation, internal merge, accounts 312 00:14:07,789 --> 00:14:09,889 for all displacement phenomena, like the Wii 313 00:14:09,889 --> 00:14:11,850 movement we talked about. What did you see? It 314 00:14:11,850 --> 00:14:14,129 replaces all the old separate transformational 315 00:14:14,129 --> 00:14:17,649 rules. Wow. Okay. One operation, two functions 316 00:14:17,649 --> 00:14:19,730 based on where the pieces come from. That is 317 00:14:19,730 --> 00:14:23,360 elegant. But... If you just reemerge something 318 00:14:23,360 --> 00:14:25,559 that's already there, what happens to the original 319 00:14:25,559 --> 00:14:27,879 position? Does it leave a gap? This is where 320 00:14:27,879 --> 00:14:31,019 merge gives you something for free, a major consequence. 321 00:14:31,480 --> 00:14:34,940 The copy theory of movement, CTM. Since internal 322 00:14:34,940 --> 00:14:37,360 merge is just taking an existing element and 323 00:14:37,360 --> 00:14:40,179 urging it again, it doesn't move it in the sense 324 00:14:40,179 --> 00:14:42,899 of leaving a hole. It copies it. So the structure 325 00:14:42,899 --> 00:14:45,779 for what did you see literally contains two copies 326 00:14:45,779 --> 00:14:48,179 of what? Like what copy, ignoring details for 327 00:14:48,179 --> 00:14:51,299 simplicity. That's the core idea, yes. Syntactically, 328 00:14:51,379 --> 00:14:53,299 both copies are there. But wait, then why don't 329 00:14:53,299 --> 00:14:55,379 we say, what did you see what? Good question. 330 00:14:55,919 --> 00:14:57,820 The theory is that when the structure is sent 331 00:14:57,820 --> 00:15:00,759 to the SM interface for pronunciation, only one 332 00:15:00,759 --> 00:15:02,720 of the copies, usually the structurally highest 333 00:15:02,720 --> 00:15:05,080 one, is actually pronounced. The lower copies 334 00:15:05,080 --> 00:15:08,340 are silent. Okay. Silent copies. Why is that 335 00:15:08,340 --> 00:15:10,860 better than the old way? Didn't GB theory have 336 00:15:10,860 --> 00:15:14,519 something like traces left behind? It did. GB 337 00:15:14,519 --> 00:15:17,440 used traces, basically. Empty placeholders, which 338 00:15:17,440 --> 00:15:19,419 were stipulated to be left behind when something 339 00:15:19,419 --> 00:15:21,730 moved. They needed to make sure the meaning was 340 00:15:21,730 --> 00:15:24,190 interpreted correctly, linking back to the original 341 00:15:24,190 --> 00:15:26,330 position. So what was wrong with traces? They 342 00:15:26,330 --> 00:15:28,289 were theoretical artifacts. They weren't part 343 00:15:28,289 --> 00:15:30,129 of the original lexical items you started with. 344 00:15:30,190 --> 00:15:31,710 You had to add them into the structure during 345 00:15:31,710 --> 00:15:34,669 the derivation. Ah, and that violates the minimalist 346 00:15:34,669 --> 00:15:39,289 desire for perfection, simplicity. Exactly. It 347 00:15:39,289 --> 00:15:41,470 violates the inclusiveness condition. I see. 348 00:15:41,649 --> 00:15:44,049 Remember that. The principle that the computation 349 00:15:44,049 --> 00:15:46,789 can't add anything new. It can only work with 350 00:15:46,789 --> 00:15:48,809 the features and objects provided by the lexicon. 351 00:15:49,230 --> 00:15:52,090 Merge, by only combining existing things, even 352 00:15:52,090 --> 00:15:54,710 if it means copying internally, naturally adheres 353 00:15:54,710 --> 00:15:58,490 to IC. Traces violated it. So CTM is seen as 354 00:15:58,490 --> 00:16:00,610 conceptually superior. It emerges naturally from 355 00:16:00,610 --> 00:16:03,590 merge without adding extra baggage. Okay, CTM 356 00:16:03,590 --> 00:16:05,549 is a bonus for merge. What else falls out? You 357 00:16:05,549 --> 00:16:08,049 mentioned merge creates sets. Right. And sets 358 00:16:08,049 --> 00:16:11,309 XY are inherently unordered. This has a massive 359 00:16:11,309 --> 00:16:14,360 consequence. Structured dependence. Since the 360 00:16:14,360 --> 00:16:17,059 core operation, merge, doesn't care about linear 361 00:16:17,059 --> 00:16:19,360 order, left to right, none of the grammatical 362 00:16:19,360 --> 00:16:21,220 rules that operate on these structures can refer 363 00:16:21,220 --> 00:16:24,340 to linear order either. So a rule can't say move 364 00:16:24,340 --> 00:16:26,960 the third word or agree with the word to the 365 00:16:26,960 --> 00:16:30,039 left? Never. Rules have to refer to structural 366 00:16:30,039 --> 00:16:32,620 relationships, like move the MP in the subject 367 00:16:32,620 --> 00:16:34,740 position or agree with the head of the phrase. 368 00:16:35,039 --> 00:16:37,700 This explains a universal fact about human language. 369 00:16:38,500 --> 00:16:41,039 Syntax is always structure -dependent, never 370 00:16:41,039 --> 00:16:43,960 purely linear. The job of actually ordering the 371 00:16:43,960 --> 00:16:46,179 words for pronunciation is pushed entirely off 372 00:16:46,179 --> 00:16:48,940 to the SM interface, the phonology part. Syntax 373 00:16:48,940 --> 00:16:52,049 itself is purely hierarchical. It separates the 374 00:16:52,049 --> 00:16:54,110 abstract computation from the messy business 375 00:16:54,110 --> 00:16:57,549 of speaking or signing. But if merge is so simple 376 00:16:57,549 --> 00:16:59,769 and elegant, why do we still have those locality 377 00:16:59,769 --> 00:17:02,570 constraints, the islands? Why can't internal 378 00:17:02,570 --> 00:17:05,009 merge just copy something from anywhere to anywhere? 379 00:17:05,230 --> 00:17:07,490 Where does locality come from in minimalism? 380 00:17:07,690 --> 00:17:09,190 Right. The system still needs to be efficient. 381 00:17:09,309 --> 00:17:12,809 It can't keep the entire infinite structure active 382 00:17:12,809 --> 00:17:15,130 in its computational workspace all the time. 383 00:17:15,650 --> 00:17:18,829 Locality comes from the idea of phases and the 384 00:17:18,829 --> 00:17:21,750 phase impenetrability condition. This replaces 385 00:17:21,750 --> 00:17:24,549 the older, more complex barrier theories. Phases, 386 00:17:24,650 --> 00:17:26,990 like stages of construction. Kind of. Think of 387 00:17:26,990 --> 00:17:29,269 them as specific structural chunks that correspond 388 00:17:29,269 --> 00:17:33,009 roughly to propositions, CPs, full clauses, and 389 00:17:33,009 --> 00:17:35,710 certain types of verb phrases, called via PV 390 00:17:35,710 --> 00:17:38,309 star P. When the computation completes a phase, 391 00:17:38,589 --> 00:17:41,309 it's like packaging it up. Packaging it up. The 392 00:17:41,309 --> 00:17:44,289 idea is multiple spell out. Once a phase is complete, 393 00:17:44,589 --> 00:17:47,369 its contents, its internal structure, are sent 394 00:17:47,369 --> 00:17:49,609 off to the SM and CI interfaces for interpretation. 395 00:17:50,089 --> 00:17:52,690 It's done. Locked down. And the PIC is what locks 396 00:17:52,690 --> 00:17:55,509 it. Exactly. The PIC says that once a phase, 397 00:17:55,670 --> 00:17:57,730 let's say phase alpha headed by H is completed, 398 00:17:57,869 --> 00:18:00,710 the internal domain of H, everything inside it 399 00:18:00,710 --> 00:18:02,869 except the head itself and its edge, becomes 400 00:18:02,869 --> 00:18:05,410 inaccessible to operations outside of alpha. 401 00:18:05,529 --> 00:18:07,490 The edge. What's the edge? Think of it as a little 402 00:18:07,490 --> 00:18:09,730 landing spot, a buffer zone at the periphery 403 00:18:09,730 --> 00:18:11,849 of the phase. If something inside the phase needs 404 00:18:11,849 --> 00:18:14,210 to eventually move out of the phase, it first 405 00:18:14,210 --> 00:18:16,349 has to move to the edge position before the phase 406 00:18:16,349 --> 00:18:19,079 is completed and sealed off. Okay, so it's like 407 00:18:19,079 --> 00:18:21,500 putting a file tab on the folder before you close 408 00:18:21,500 --> 00:18:23,539 it so you can still grab it later. That's a great 409 00:18:23,539 --> 00:18:26,140 analogy, yeah. If an element doesn't move to 410 00:18:26,140 --> 00:18:28,839 the edge, once the phase is complete, it's stuck 411 00:18:28,839 --> 00:18:32,480 inside forever. This forces movement to be successive 412 00:18:32,480 --> 00:18:34,619 cyclic, moving step by step through the edges 413 00:18:34,619 --> 00:18:37,299 of phases, and it naturally explains why islands 414 00:18:37,299 --> 00:18:40,019 exist. Islands are essentially phase boundaries 415 00:18:40,019 --> 00:18:43,000 that weren't properly evacuated via the edge. 416 00:18:43,220 --> 00:18:46,420 Wow. So the PIC enforces locality as a consequence 417 00:18:46,420 --> 00:18:48,779 of efficient... chunk -by -chunk computation. 418 00:18:49,079 --> 00:18:51,519 It's not an arbitrary list of constraints anymore. 419 00:18:51,740 --> 00:18:54,119 That's the goal. It derives locality from the 420 00:18:54,119 --> 00:18:56,680 need to manage computational resources. Okay, 421 00:18:56,740 --> 00:18:59,259 so we have this elegant, minimal computational 422 00:18:59,259 --> 00:19:03,599 system merge operating in phases. Let's connect 423 00:19:03,599 --> 00:19:05,940 this back to the bigger picture biology and evolution. 424 00:19:06,400 --> 00:19:09,460 Right. The whole minimalist approach reinforces 425 00:19:09,460 --> 00:19:11,539 the internalist stance in cognitive science. 426 00:19:11,940 --> 00:19:15,420 The idea that to do serious science You have 427 00:19:15,420 --> 00:19:17,640 to focus on the internal structures of the mind, 428 00:19:17,759 --> 00:19:20,519 these mental organs or modules. And language 429 00:19:20,519 --> 00:19:22,799 is one of those organs? Yes, the language capacity 430 00:19:22,799 --> 00:19:26,259 or LC. It's seen as an innate part of our biological 431 00:19:26,259 --> 00:19:28,920 endowment, developing along a genetically guided 432 00:19:28,920 --> 00:19:32,220 path, just needing environmental input to trigger 433 00:19:32,220 --> 00:19:34,319 and shape it within its predetermined limits. 434 00:19:34,559 --> 00:19:37,400 And this innate blueprint is universal grammar, 435 00:19:37,680 --> 00:19:40,730 UG. Correct. UG is the initial state of the language 436 00:19:40,730 --> 00:19:42,950 faculty. It's what all humans are born with. 437 00:19:43,029 --> 00:19:45,930 It's species -specific, unique to us, and believed 438 00:19:45,930 --> 00:19:48,269 to be remarkably uniform across the human population, 439 00:19:48,549 --> 00:19:51,009 basically fixed since our species emerged and 440 00:19:51,009 --> 00:19:52,950 migrated out of Africa. The big argument for 441 00:19:52,950 --> 00:19:55,130 UG always comes back to the input problem, right? 442 00:19:55,230 --> 00:19:58,069 The poverty of stimulus. POS. It's still the 443 00:19:58,069 --> 00:20:00,509 core argument, yes. Children learn language incredibly 444 00:20:00,509 --> 00:20:03,150 fast, incredibly well, and reach basically the 445 00:20:03,150 --> 00:20:06,309 same complex grammar, regardless of huge variations 446 00:20:06,309 --> 00:20:08,230 in the quality and quantity of the language they 447 00:20:08,230 --> 00:20:10,630 hear. And often the input is messy, incomplete, 448 00:20:10,930 --> 00:20:14,670 full of errors. Exactly. And crucially, it lacks 449 00:20:14,670 --> 00:20:16,869 the kind of explicit negative evidence needed 450 00:20:16,869 --> 00:20:19,670 to rule out incorrect grammatical hypotheses. 451 00:20:20,470 --> 00:20:22,849 Take structure dependence again. The rule about 452 00:20:22,849 --> 00:20:25,589 moving the auxiliary verb in questions. Yeah. 453 00:20:25,769 --> 00:20:28,809 How does a child learn that the man who is tall 454 00:20:28,809 --> 00:20:31,650 is happy? You form the question by moving the 455 00:20:31,650 --> 00:20:35,029 second. The second ends the man who is tall happy, 456 00:20:35,210 --> 00:20:37,970 not the first one. is the man who tall is happy. 457 00:20:38,069 --> 00:20:40,970 They probably never hear sentences complex enough 458 00:20:40,970 --> 00:20:43,890 often enough or get corrected specifically on 459 00:20:43,890 --> 00:20:45,809 that point to figure it out just from the data. 460 00:20:46,069 --> 00:20:48,490 Precisely. The input underdetermines the grammar 461 00:20:48,490 --> 00:20:50,990 they acquire. The POS argument says there must 462 00:20:50,990 --> 00:20:54,410 be rich, innate structure UG guiding them, telling 463 00:20:54,410 --> 00:20:56,970 them that rules operate on hierarchies, not linear 464 00:20:56,970 --> 00:20:59,309 strings, that knowledge isn't learned. It's built 465 00:20:59,309 --> 00:21:01,890 in. OK, so UG is the innate state. Now, this 466 00:21:01,890 --> 00:21:03,309 is where it gets really interesting, trying to 467 00:21:03,309 --> 00:21:05,170 figure out when this whole system, this faculty 468 00:21:05,170 --> 00:21:07,910 of language, FL, built around merge, actually 469 00:21:07,910 --> 00:21:10,410 appeared in humans. Yeah, the evolutionary question. 470 00:21:11,099 --> 00:21:14,299 If minimalism is right and the core is just merge. 471 00:21:14,940 --> 00:21:17,500 Maybe it wasn't a long, gradual process. Maybe 472 00:21:17,500 --> 00:21:20,099 it was a relatively sudden event. A single mutation, 473 00:21:20,359 --> 00:21:23,079 perhaps, that wired merge into the brain. A single 474 00:21:23,079 --> 00:21:25,740 mutation creating infinite thought? That sounds 475 00:21:25,740 --> 00:21:28,660 traumatic. It does, but it fits the logic of 476 00:21:28,660 --> 00:21:32,119 trying to find the simplest explanation. So researchers 477 00:21:32,119 --> 00:21:34,380 try to pin down the timeline using evidence from 478 00:21:34,380 --> 00:21:37,500 paleoanthropology, archaeology, genomics. What's 479 00:21:37,500 --> 00:21:39,579 the rough window? Well, it has to be after our 480 00:21:39,579 --> 00:21:42,269 lineage split from other apes, obviously. And 481 00:21:42,269 --> 00:21:44,910 likely after the appearance of anatomically modern 482 00:21:44,910 --> 00:21:48,009 humans, AMH, around, say, 200 ,000 years ago. 483 00:21:48,089 --> 00:21:51,369 But probably before major signs of symbolic behavior 484 00:21:51,369 --> 00:21:54,109 become widespread. Like the Blombo's cave findings, 485 00:21:54,390 --> 00:21:56,569 the engraved ochre. Exactly. Those date back 486 00:21:56,569 --> 00:21:59,990 to maybe 80 ,000, 90 ,000 years ago. So that 487 00:21:59,990 --> 00:22:02,250 suggests the capacity for complex thought, presumably 488 00:22:02,250 --> 00:22:04,470 language -based, was there by then. Okay, so 489 00:22:04,470 --> 00:22:06,869 somewhere between 200 ,000 and 80 ,000 years 490 00:22:06,869 --> 00:22:09,839 ago. Can we narrow it more? Genomics helps. Studies 491 00:22:09,839 --> 00:22:11,700 looking at population divergence, particularly 492 00:22:11,700 --> 00:22:13,779 of the Khoisan peoples in southern Africa. Who 493 00:22:13,779 --> 00:22:16,200 speak languages with unique click sounds. Right. 494 00:22:16,500 --> 00:22:18,960 Those studies suggest the sand population lineage 495 00:22:18,960 --> 00:22:21,539 diverged from other human populations somewhere 496 00:22:21,539 --> 00:22:24,960 around 108 ,000 to 157 ,000 years ago. Since 497 00:22:24,960 --> 00:22:27,380 they have language like all other humans, the 498 00:22:27,380 --> 00:22:29,299 faculty must have been in place before that split. 499 00:22:29,480 --> 00:22:32,000 Pushing the date back significantly. It does. 500 00:22:32,470 --> 00:22:34,170 And then there's the really deep time question. 501 00:22:34,369 --> 00:22:37,190 The split between our ancestors and the Neanderthals' 502 00:22:37,210 --> 00:22:40,029 Denisovans was much earlier, maybe 500 ,000 to 503 00:22:40,029 --> 00:22:43,089 700 ,000 years ago. Could FL have emerged before 504 00:22:43,089 --> 00:22:45,529 that split? It's debated. But the minimalist 505 00:22:45,529 --> 00:22:47,730 focus on a single simple mutation like merge 506 00:22:47,730 --> 00:22:50,190 makes a more recent origin, maybe around the 507 00:22:50,190 --> 00:22:53,329 time of AMH, seem plausible. And if merge arose 508 00:22:53,329 --> 00:22:56,089 relatively recently and maybe suddenly, what 509 00:22:56,089 --> 00:22:58,750 does that imply about its primary function? This 510 00:22:58,750 --> 00:23:01,029 is a crucial and controversial conclusion of 511 00:23:01,029 --> 00:23:04,069 the program. FL seems to be fundamentally a system 512 00:23:04,069 --> 00:23:06,450 for internal thought, an internal language of 513 00:23:06,450 --> 00:23:09,029 thought. Not primarily for communication. Communication 514 00:23:09,029 --> 00:23:12,029 appears to be a secondary use. The process of 515 00:23:12,029 --> 00:23:13,890 externalization, getting the thoughts out into 516 00:23:13,890 --> 00:23:16,930 sound or sign via the SM interface, is complex, 517 00:23:17,250 --> 00:23:19,869 messy, and likely evolved later, possibly under 518 00:23:19,869 --> 00:23:22,089 different selection pressures. So merge gives 519 00:23:22,089 --> 00:23:24,390 us the power for infinite structured thought, 520 00:23:24,549 --> 00:23:26,190 and then we figured out how to talk about it. 521 00:23:26,269 --> 00:23:29,130 That's the picture, yeah. Things like the unique 522 00:23:29,130 --> 00:23:31,950 click consonants in Khoisan languages could be 523 00:23:31,950 --> 00:23:35,150 seen as evidence for this. If the internal system, 524 00:23:35,349 --> 00:23:38,470 merge, was universal, but externalization was 525 00:23:38,470 --> 00:23:41,190 a later, perhaps harder problem to solve, you 526 00:23:41,190 --> 00:23:43,190 might expect different populations to develop 527 00:23:43,190 --> 00:23:46,009 slightly different peripheral solutions for outputting 528 00:23:46,009 --> 00:23:48,769 the structures. Clicks might be one such solution 529 00:23:48,769 --> 00:23:51,369 that arose and stabilized in that population 530 00:23:51,369 --> 00:23:54,009 group. That really flips the common sense view 531 00:23:54,009 --> 00:23:56,269 of language on its head. Okay, this leads us 532 00:23:56,269 --> 00:23:59,490 to another controversial area. Meaning, if language 533 00:23:59,490 --> 00:24:02,349 is internal, biological, how does it connect 534 00:24:02,349 --> 00:24:05,509 to the world? Right. This brings us to Chomsky's 535 00:24:05,509 --> 00:24:08,309 semantic internalism. Have a pretty radical departure 536 00:24:08,309 --> 00:24:10,210 from how many philosophers, and frankly, most 537 00:24:10,210 --> 00:24:13,430 people, think about meaning. He essentially denies 538 00:24:13,430 --> 00:24:15,789 that words reliably pick out specific things 539 00:24:15,789 --> 00:24:17,990 or categories in the external objective world. 540 00:24:18,289 --> 00:24:21,630 Wait, so water doesn't refer to H -O? Dog doesn't 541 00:24:21,630 --> 00:24:23,930 refer to dogs? Not in the way we usually think. 542 00:24:24,089 --> 00:24:26,690 He argues against externalist theories of meaning, 543 00:24:26,869 --> 00:24:29,710 like Putnam's idea that meaning just ain't in 544 00:24:29,710 --> 00:24:32,529 the head. Chomsky says, for ordinary language 545 00:24:32,529 --> 00:24:35,769 concepts, meaning is very much in the head, and 546 00:24:35,769 --> 00:24:38,170 it's shaped by our human perspective. What do 547 00:24:38,170 --> 00:24:40,130 you mean human perspective? Our common sense 548 00:24:40,130 --> 00:24:44,109 concepts, water, house, river, even language 549 00:24:44,109 --> 00:24:47,299 itself, are inherently anthropocentric. They 550 00:24:47,299 --> 00:24:50,460 reflect human interests, goals, intended functions, 551 00:24:50,799 --> 00:24:52,720 how we interact with things. They aren't the 552 00:24:52,720 --> 00:24:54,920 objective categories of natural science. Okay, 553 00:24:54,960 --> 00:24:57,180 I need an example here. The famous water tea 554 00:24:57,180 --> 00:24:59,319 one. Yes, that illustrates it perfectly. Take 555 00:24:59,319 --> 00:25:01,460 a glass of tap water, we call it water. Now, 556 00:25:01,500 --> 00:25:04,019 dip a teabag in it for a second. Chemically, 557 00:25:04,079 --> 00:25:06,140 it's almost identical, maybe a few trace molecules 558 00:25:06,140 --> 00:25:08,440 added. But we instantly call it tea. Exactly. 559 00:25:08,579 --> 00:25:11,480 Why? Because of a human action, dipping the bag. 560 00:25:11,799 --> 00:25:16,119 And intended use to drink is tea. The concept 561 00:25:16,119 --> 00:25:18,700 shifts based on our goals, not the underlying 562 00:25:18,700 --> 00:25:21,480 physics or chemistry. And the reverse. If you 563 00:25:21,480 --> 00:25:25,279 had, say, some leftover tea, and you poured it 564 00:25:25,279 --> 00:25:27,740 through an industrial purification system designed 565 00:25:27,740 --> 00:25:31,400 to produce potable water for a town, the output, 566 00:25:31,579 --> 00:25:34,039 even if it chemically resembled weak tea more 567 00:25:34,039 --> 00:25:36,960 than pure HO, would likely be called water by 568 00:25:36,960 --> 00:25:39,519 the engineers because its intended function is 569 00:25:39,519 --> 00:25:41,539 to be part of the water supply. So the label 570 00:25:41,539 --> 00:25:44,500 follows human purpose, not objective composition. 571 00:25:44,799 --> 00:25:47,000 Right. And Chomsky loves pointing out things 572 00:25:47,000 --> 00:25:49,359 like Diet Coke and Bud Light actually have a 573 00:25:49,359 --> 00:25:51,779 higher percentage of HO molecules than seawater 574 00:25:51,779 --> 00:25:54,559 does. If our concept water simply meant HO plus 575 00:25:54,559 --> 00:25:57,299 or minus impurities, then diet soda should be 576 00:25:57,299 --> 00:25:59,680 a better example of water than the ocean. Which 577 00:25:59,680 --> 00:26:02,420 is absurd. OK, so our everyday concepts are kind 578 00:26:02,420 --> 00:26:05,539 of perspectival. geared for human interaction. 579 00:26:05,779 --> 00:26:08,140 Exactly. They're tools for navigating our world, 580 00:26:08,160 --> 00:26:11,019 not for objective science. So if words don't 581 00:26:11,019 --> 00:26:13,440 refer directly to external things, what are meanings 582 00:26:13,440 --> 00:26:15,799 internally? Tomsey talks about I -meanings, internal 583 00:26:15,799 --> 00:26:17,680 meanings. And what do they do? They function 584 00:26:17,680 --> 00:26:19,759 more like assembly instructions for the conceptual 585 00:26:19,759 --> 00:26:21,819 intentional system. They provide perspectives, 586 00:26:22,200 --> 00:26:24,559 pointers, ways to categorize and relate things 587 00:26:24,559 --> 00:26:26,559 within our thoughts. They help us build internal 588 00:26:26,559 --> 00:26:29,430 representations. query our encyclopedic knowledge, 589 00:26:29,630 --> 00:26:33,029 make inferences, but they don't establish a fixed, 590 00:26:33,089 --> 00:26:36,009 rigid link between a word and a specific external 591 00:26:36,009 --> 00:26:38,609 thing or set of things. So meaning is more about 592 00:26:38,609 --> 00:26:41,150 activating internal conceptual resources than 593 00:26:41,150 --> 00:26:43,170 pointing outward. That's a good way to put it. 594 00:26:43,210 --> 00:26:46,289 It allows for the flexibility and nuance we see 595 00:26:46,289 --> 00:26:49,410 in word meaning, the polysemy, the context dependence. 596 00:26:49,690 --> 00:26:51,769 We have multiple ways of thinking about book, 597 00:26:51,970 --> 00:26:55,089 physical object, information content, or London. 598 00:26:55,680 --> 00:26:58,759 geographical area, population, government. I 599 00:26:58,759 --> 00:27:01,279 meanings let us access these different facets. 600 00:27:01,579 --> 00:27:03,500 Okay, that's a very different view of semantics. 601 00:27:03,660 --> 00:27:05,539 Before we wrap up, there's one more connection 602 00:27:05,539 --> 00:27:07,960 you mentioned linking this view of language to 603 00:27:07,960 --> 00:27:10,500 morality. Yes, it's an interesting parallel he 604 00:27:10,500 --> 00:27:13,460 draws. Just as he argues for an innate universal 605 00:27:13,460 --> 00:27:16,019 grammar for language, he suggests there might 606 00:27:16,019 --> 00:27:18,420 be an innate generative basis for moral competence 607 00:27:18,420 --> 00:27:21,220 too, a kind of universal moral grammar. Meaning 608 00:27:21,220 --> 00:27:24,180 we don't just learn moral rules by rote. Right. 609 00:27:24,240 --> 00:27:27,079 The idea is that we possess an innate faculty 610 00:27:27,079 --> 00:27:30,079 that allows us to generate moral judgments about 611 00:27:30,079 --> 00:27:33,039 entirely new situations, situations we've never 612 00:27:33,039 --> 00:27:35,839 faced or been taught about explicitly. It exhibits 613 00:27:35,839 --> 00:27:39,140 the same infinite use of finite means, property, 614 00:27:39,299 --> 00:27:42,259 as language. We apply underlying principles to 615 00:27:42,259 --> 00:27:44,980 new cases. Exactly. We have finite principles, 616 00:27:45,200 --> 00:27:48,059 perhaps related to fairness, harm, intent, and 617 00:27:48,059 --> 00:27:50,799 so on, and we can apply them generatively to 618 00:27:50,799 --> 00:27:53,420 an unbounded range of complex ethical dilemmas. 619 00:27:53,460 --> 00:27:56,380 For Chomsky, this reflects a biological endowment 620 00:27:56,380 --> 00:27:59,319 for sociability, cooperation, and even freedom, 621 00:27:59,440 --> 00:28:02,359 the capacity to make reasoned judgments, rather 622 00:28:02,359 --> 00:28:04,839 than just react instinctively or based on conditioning. 623 00:28:05,410 --> 00:28:07,849 So the same computational principles might underlie 624 00:28:07,849 --> 00:28:10,430 both language and our moral sense. That's a powerful 625 00:28:10,430 --> 00:28:13,130 idea. It is. It paints a picture of human nature 626 00:28:13,130 --> 00:28:15,210 rooted in these innate generative capacities. 627 00:28:15,670 --> 00:28:17,549 Okay, let's try to bring this all together. When 628 00:28:17,549 --> 00:28:19,769 you look back at Chomsky's decades of work in 629 00:28:19,769 --> 00:28:22,130 linguistics, what's the absolute core contribution? 630 00:28:22,549 --> 00:28:24,710 I think it has to be fundamentally shifting the 631 00:28:24,710 --> 00:28:27,609 object of study, anchoring language firmly in 632 00:28:27,609 --> 00:28:30,569 biology in the mind, treating it as a natural 633 00:28:30,569 --> 00:28:33,789 object, a mental organ. subject to scientific 634 00:28:33,789 --> 00:28:36,329 inquiry like any other biological system. And 635 00:28:36,329 --> 00:28:39,910 that shift forced a particular kind of theorizing. 636 00:28:40,089 --> 00:28:42,869 Yes. It drove the relentless pursuit of simplicity 637 00:28:42,869 --> 00:28:46,769 and explanatory depth. Constantly asking, what's 638 00:28:46,769 --> 00:28:49,549 the simplest mechanism that could possibly account 639 00:28:49,549 --> 00:28:52,480 for these complex phenomena? This led from the 640 00:28:52,480 --> 00:28:54,880 complexities of early transformations to the 641 00:28:54,880 --> 00:28:57,960 radical minimalism of merge. It forced the field 642 00:28:57,960 --> 00:29:00,519 to confront deep puzzles, infinitude, structure 643 00:29:00,519 --> 00:29:03,799 dependence, displacement, locality, and propose 644 00:29:03,799 --> 00:29:06,319 computationally explicit solutions. Solutions 645 00:29:06,319 --> 00:29:09,480 like recursion, merge, copies, phases. Exactly. 646 00:29:09,519 --> 00:29:11,920 And in doing so, it challenged deeply ingrained 647 00:29:11,920 --> 00:29:14,240 beliefs that language is primarily for communication, 648 00:29:14,619 --> 00:29:16,759 that meaning is about reference, that learning 649 00:29:16,759 --> 00:29:19,160 is just habit formation. It opened up new ways 650 00:29:19,160 --> 00:29:21,160 to think about thought itself. For listeners 651 00:29:21,160 --> 00:29:22,960 who want to dive deeper into these ideas, the 652 00:29:22,960 --> 00:29:25,079 source material is obviously vast, but where 653 00:29:25,079 --> 00:29:27,119 are some good starting points? Well, for the 654 00:29:27,119 --> 00:29:29,619 linguistics itself, aspects of the theory of 655 00:29:29,619 --> 00:29:32,819 syntax is the classic, though dense. For a broader 656 00:29:32,819 --> 00:29:35,359 overview connecting language, mind, and even 657 00:29:35,359 --> 00:29:38,400 as politics, James McGilver's book Chomsky, Language, 658 00:29:38,519 --> 00:29:40,980 Mind, and Politics, is excellent. And if people 659 00:29:40,980 --> 00:29:43,160 are interested in how these ideas connect to 660 00:29:43,160 --> 00:29:46,359 his more famous political analyses, then Manufacturing 661 00:29:46,359 --> 00:29:48,559 Consent, co -authored with Edward Herman, is 662 00:29:48,559 --> 00:29:50,559 essential reading, showing how media structures 663 00:29:50,559 --> 00:29:52,819 can shape perception within the bounds of that 664 00:29:52,819 --> 00:29:55,839 assumed shared rationality. And for a deep dive 665 00:29:55,839 --> 00:29:58,819 into a specific geopolitical analysis, the fateful 666 00:29:58,819 --> 00:30:01,519 triangle on the Israel -Palestine conflict is 667 00:30:01,519 --> 00:30:04,180 a major work. Again, informed by structural analysis. 668 00:30:04,519 --> 00:30:06,920 Great suggestions. OK, time for the final provocative 669 00:30:06,920 --> 00:30:08,960 thought. All right. So if we take the minimalist 670 00:30:08,960 --> 00:30:12,460 picture seriously, if the core of human language, 671 00:30:12,599 --> 00:30:15,640 merge, arose potentially as a single, simple 672 00:30:15,640 --> 00:30:18,619 evolutionary step, and if his primary function 673 00:30:18,619 --> 00:30:21,710 internally is to generate thought. with externalization 674 00:30:21,710 --> 00:30:24,410 for communication being a more complex, secondary 675 00:30:24,410 --> 00:30:27,549 adaptation. Then why do we, in our everyday thinking 676 00:30:27,549 --> 00:30:30,049 and even in many scientific approaches outside 677 00:30:30,049 --> 00:30:32,670 of this specific framework, still instinctively 678 00:30:32,670 --> 00:30:35,049 treat language as if it is primarily communication, 679 00:30:35,450 --> 00:30:38,490 as if speech or writing is the language? Are 680 00:30:38,490 --> 00:30:40,450 we systematically overlooking our most unique 681 00:30:40,450 --> 00:30:43,299 cognitive gift? unbounded recursive thought, 682 00:30:43,380 --> 00:30:45,579 perhaps because its external manifestation is 683 00:30:45,579 --> 00:30:47,599 just so much more obvious, even though it might 684 00:30:47,599 --> 00:30:50,059 be derivative, it makes you wonder if we're still 685 00:30:50,059 --> 00:30:51,900 missing the true nature of our own minds.