Skip to main content
Loading Explore guides…
Introduction to Cognitive Science · Acedra
Home
Explore
Psychology
Introduction to Cognitive Science
Sign in to save
Chat
If YouTube asks you to sign in, open the video on YouTube.
Watch on YouTube
Reload player
Transcript
263 segments
Copy
Hide video
0:00
Hey everyone. Welcome to Introduction to Cognitive Science. I'm Adam, and the goal for our lecture today is to introduce you to the field of cognitive science.
0:12
This summer session we are focusing on the cognitive science of language, so let's begin by thinking about some of the important ways that language figures in our mental and social life.
0:29
So language is central to our mental and social lives. Let's consider and reflect on some of the ways in which language is central to our lives.
0:41
Think about the thoughts that you have about yourself, your friends, and others,
0:47
your memories about the past and goals for the future. Are these articulated in language?
0:55
Think about the most hurtful and heartwarming things that someone has ever said to you. Has someone broken your heart? Or maybe your acceptance letter to CAL, how did that make you feel? Was that in language?
1:11
Think about the socialization processes that we engage in such as gossiping, providing testimony, and sharing social knowledge with each other.
1:23
Think about laws, social regulations, and formal agreements.
1:29
Would things like marriage or graduation ceremonies be possible without language?
1:38
Also consider scientific facts, hypotheses, and the very accumulation of knowledge itself. Can we build knowledge without language?
1:52
Also consider philosophical inquiry, the nature of thought experiments and counterfactual reasoning, reasoning with if-thens.
2:02
Is that possible without language or conditional statements? Also we have prayers, so not only are we talking about our inner mental life and our social life but even aspects of our religious life include language, such as prayer, religious texts, and religious practices.
2:26
Also consider some of the most exemplary works of art, including literature, poetry, and lyrical music. Would our aesthetic experience be as full and rich without language?
2:41
And also bringing the debate to the current age, we have memes, stereotypes, propaganda, and so much more to consider.
2:52
All these things, it seems, draw importantly on our ability to use and understand language.
3:01
By reflecting on the omnipresence and centrality of language in our mental and social lives,
3:11
I'm challenging you to think about how important language is to the field of cognitive science and any inquiry into the nature of mind, more generally. My view is that, by learning more about language and the mind, we can use this to help enrich our appreciation of the central role that language plays in our mental and social lives.
3:38
We read a few articles for today. One of the articles for today is this entry in the Stanford Encyclopedia of Philosophy on Cognitive Science.
3:48
This is published by Thagard 1996. The most important things I wanted us to take away from this first article is just getting an understanding of what cognitive science is as well as some of the historical foundations.
4:06
So what is cognitive science? How is it distinct from other fields or disciplines that also investigate the mind?
4:17
Well, cognitive science is the interdisciplinary study of mind and intelligence or intelligent systems.
4:27
To better understand the mind and intelligence, cognitive science draws upon seven subfields. These include anthropology, artificial intelligence, education linguistics, neuroscience philosophy and psychology.
4:48
Cognitive science is distinct from these subfields because cognitive science is going to draw upon all relevant subdomains or subdisciplines in order to provide a comprehensive understanding of the mind itself or intelligent systems.
5:08
Notice that, whereas neuroscience focuses on an implementation level or, basically, hardware, how the brain operates, philosophy is going to approach the mind from a different level, from the algorithmic or computational level, sort of the problems that the mind has to solve.
5:29
So we see one strength of cognitive science, one strength of this interdisciplinary approach is, whereas philosophy may be excellent at providing certain levels of analysis of the mind, which David Marr will call the computational or algorithmic level, we see that it's going to fail to provide a rich understanding of the neural level, how brains work,
6:00
or what Marr will call the implementation level.
6:04
And vice versa, the opposite is true for neuroscience. Neuroscience is going to be exceptionally well equipped at studying the implementation of cognition, what material properties of the brain, neurons and whatnot, are responsible for cognitive functioning.
6:27
So we see that when we engage in neuroscience, we're focused on a different level, the level of implementation rather than computation or algorithm.
6:39
So whereas these distinct subdomains or subdisciplines will focus on a certain level primarily,
6:49
what we're going to do as cognitive scientists is try to draw upon the best findings, methods, and concepts from these different domains and paint an overall integrative picture of how the mind works, what it is, and how it works.
7:08
I think that this is a beautiful idea.
7:13
It pays due attention to the fact that we can each, within each discipline, sort of be restricted to a certain level of analysis, and by engaging with other cognitive scientists that have different areas of expertise, as well as working to enrich our own breadth of methodological expertise, this will help us have a better, more full understanding of what the mind is. And as I mentioned in the last lecture,
7:47
we can draw upon linguistic anthropology. We can draw upon natural language processing. We can draw upon applied linguistics and language learning. We can draw upon research on phonetics, morphology, syntax, semantics, and pragmatics. We can draw upon research on the cognitive neuroscience of language.
8:11
We can draw upon philosophies or theories from the philosophy of language, as well as empirical research from the psychology of language. And you see, by drawing upon these different methods that focus on different levels of analysis, we can start to paint a well-informed picture of how the mind comprehends and produces language.
8:35
And again, for this summer session, we're going to focus on language just so that we have a clear target of what we're trying to understand as cognitive scientists. As I mentioned in the last lecture, cognitive scientists also pursue other topics such as the emotions and facial expressions, the nature of musical processing, and many other fascinating topics.
8:59
All right, so this is what cognitive science is, how it's different from the subdomains that it includes, and sort of the strengths of this approach. A weakness of this approach is that it's easy, in some sense, to be bad at cognitive science. You can easily lose focus and just sort of learn random things from anthropology, learn random facts from neuroscience, and not do a good job of tying things together.
9:31
So what we want to do, as anyone interested in interdisciplinary research, is we want to make sure that we're always coming back to something central or something topical so that we have a clear purpose in our interdisciplinary investigation, so that we don't get lost.
9:58
So a very interesting question in cognitive science,
10:02
very fascinating for me but also it's been very influential in the field itself, is this question, how do we learn language? It's something that we all do as children.
10:14
We do it rather well. However, it's absolutely puzzling how we're able to do this so well.
10:24
You can imagine, as a child, you're born into the world, and you might have what William James calls a "blooming, buzzing confusion". There's just a lot of phenomenal experience going on, and you-- it seems very hard to make sense of all of this.
10:42
Interestingly enough, children are able to learn words very young and learn words at an incredible rate. It's just, something we'll talk about in more detail later. Here, I'm just using this to open up a question, how do we learn words?
10:59
You might think, as B.F. Skinner thinks, that we learn this through a form of learning theory.
11:10
B.F. Skinner in his very popular work Verbal Behavior, published in 1957, he puts forward this idea that language is acquired through learning theory. And this is a behaviorist approach to psychology,
11:27
so behaviorism and learning theory go together.
11:31
And this is the idea that language acquisition is due to learning from experience-based training and conditioning. Okay, so this is the idea that learning a language is similar to learning how to skateboard or learning any other skill.
11:53
This is also similar-- this idea's similar to how animals, nonhuman animals, learn certain complex tasks. If you want to train a chimpanzee to perform some sort of physical behavior, maybe you want them to tap a certain thing or to produce certain gestures, well, you're going to use reward and punishment to try to modify the animal's behavior.
12:23
And through a series of punishments and rewards, you can steer the animal's behavior largely to engaging in the actions that you want it to engage in.
12:34
Of course, this is within limits, but the general idea here is--
12:41
and one of the major contributions at this time is the idea of conditioning. So we know that Pavlov's dogs, they demonstrated conditioning to a stimulus. At first, they salivate only to food, but when a bell is paired with the food-- so you ring a bell before you present the food--
13:08
the dogs will start to salivate at the presentation of the bell
13:14
because the bell, they've learned, that the bell is tied to the food, that, in some sense, the bell predicts that food is coming up.
13:23
So through this early work, which led to a Nobel Prize in physiology, we see that learning theory and behaviorism was a very important idea during this time.
13:40
And we can-- we want to contextualize the contribution here because learning theory and behaviorism was a response to Freudian theory, where we posit sort of an ego and an id and a superego, and these are sort of like concepts within our minds. It's not observable, and it's hard to do science this way. So you see that the move from this sort of unconscious Freudian psychology to behaviorism, this is leading to the scientific credibility
14:22
of psychology. It's sort of aligning psychology with physiology, which is a very important step at this time.
14:31
So since we can explain learning nonlinguistic behavior through behaviorism and learning theory, which we've done to explain basic physiological responses in animals, the idea here, what Skinner tried to argue in Verbal Behavior, is that learning language is just like this. Okay, so maybe I see an apple, and I call it an apple. And then my mom is going to say, "good job, Adam",
15:00
and I'm going to get a reward, like maybe verbal praise.
15:03
And maybe as a child, when I see an apple, I accidentally call it something different, like "mama". Well, that's wrong, so then my mom will correct me, "no, that's not mama. That's an apple". And then through a punishment-- maybe she'll give me a--
15:26
she'll express an attitude that is indicative of negative emotions, or maybe I'll-- she'll leave me alone, and that might be a form of punishment through-- the idea here is that, through rewards and punishment,
15:42
children also learn language just like they might learn anything else, like how to perform a gesture or tap things on a board correctly.
15:54
So we can now appreciate behaviorism and learning theory as a response to Freudian psychology. However, we now have also another approach to answering this question of how we learn language,
16:10
and this is provided from Noam Chomsky, who's one of the most cited authors of all time. And importantly, he did foundational work on syntax and the cognitive science of language.
16:25
So in 1959, Chomsky published a review of BF Skinner's Verbal Behavior, which, in some sense, undermined this approach and showed the flaw in trying to apply behaviorism and learning theory to language acquisition.
16:44
So it may be the case that much of what we learn is attributable to some sort of learning theory. However, Chomsky is going to point out that certain aspects of cognition cannot be grounded in pure behaviorist theory.
17:06
Language is not learned based on rewards and punishments. It's not learned like how we learn to skateboard. Noam Chomsky is going to argue against that view and put forth a very different, diametrically opposed view of language acquisition.
17:28
So the opposite of behaviorism and learning theory is nativism and Universal Grammar. This takes just an opposite approach to how we learn language.
17:42
On this view, language acquisition is due to the development of an innate cognitive module with rules for language.
17:54
So whereas behaviorist-- the behaviorist approach to language learning thought that we learn language just like we learn how to skateboard,
18:03
this view, the nativist view of language learning, is that, rather than learning this as a skill from the outside, language is actually a module, a cognitive module, that we grow within us, that we develop within us. Just like we develop a heart, we develop lungs, we develop different sort of physiological modules, if you will. Just we develop these modules, we similarly develop cognitive modules. And on this view, it's not that we learn language
18:37
from the outside, but rather, we develop or mature a language module from the inside. And that's why we call it a nativist approach. It's based on the idea that language is really innate to us as humans. There's something about the way we are as human beings that language is really intricate to our human nature, if you will.
19:03
So we see that these are very different views of how we learn a language. In some sense, the behaviorist view is a tabula rasa, empiricist view. The mind's a blank slate, and whatever I want my mind to have in terms of skills or rules, I have to bring it from the outside. It has to be learned from the world and put into my mind.
19:27
The other view is that-- rejects the idea that the mind is a blank slate, but rather, the mind contains modules, cognitive modules. And that means we're born with some basic knowledge of rules for grammar, grammatical rules. And one such rule may be all sentences, insofar as they're wellformed sentences, must contain a noun phrase and a verb phrase. That might be such a rule.
19:57
All right, so we will-- now that we see these two different views of language, we can better appreciate each view within its historical context, behaviorism as a response to Freudian psychology, cognitivism, or cognitive science, as a response to behaviorism. So a lot of the course, what we're going to do-- you can be a psychologist and be a behaviorist.
20:24
You can also be a psychologist or a physiologist and sort of focus on the physiological basis of behavior but not really focus too much on cognitive, higher-order cognitive aspects like what's going on with language.
20:43
So again, just notice that with cognitive science, as we start to entertain the arguments and the implications of Chomsky, photo and others, we are focusing on cognitive phenomena, not just physiological or sensory phenomenon but what's truly cognitive.
21:07
So language, I think, is a great example. I'll show you some more--
21:13
in the next few slides, we'll talk more about why language is such a great example of cognitive phenomena, but also I want to check in with you and just make sure that you're thinking now-- when you think about the different aspects of mental life, ask yourself, is this cognitive? Or is this something maybe lower level? Or how are you making these distinctions in certain domain areas? Like emotions, is that cognitive? Or is it purely physiological? I want you to be asking these questions
21:46
as we go through different content in the course.
21:50
Okay, all right, so now that we have some different views on the table of how we learn a language-- we have the behaviorist view. We also have the nativist view-- now, I just want to offer you some slides to push the argument.
22:08
Why might we believe one view over the other? What's the heart of Chomsky's argument against behaviorist psychology and learning theory?
22:19
Okay, so Chomsky provides some nice examples in his book. I have provided a few of my own, which I think are really fun.
22:30
So one thing that's important to understand about language is that language is not just a string of words next to each other.
22:39
So we see right here, I have the structure of language. I have four words printed right here.
22:47
But language is not just stringing together independent pieces sort of in linear fashion,
22:55
but rather, language has hierarchical structure.
23:02
And that's what I'm going to show you through the next few slides is that we can't understand language as pure pairings of stimulus and response or things that are tied through local contingencies. We'll see that, due to the hierarchical structure of language, that a lot of the operations that we perform linguistically, like forming questions,
23:27
have to deal with structural movements.
23:30
And in order for account for structural movements in language, we have to think of language in terms of structure,
23:39
in terms of phrases, noun phrases and verb phrases, rather than just words, words, words, words, words, words.
23:48
So this is very important because what this means is, in order for something to have hierarchical structure, we have to be able to assign distinct categorical labels to things, like this is an NP, or a noun phrase, and this is a VP, or a verb phrase. And then we also have to have rules for how those parts go together.
24:08
Just like there's rules for how LEGOs can go together and there's rules for how moves in chess can be performed, we also have grammatical rules or rules for well-formed syntactic constructions in language. And what's really cool is that we seem to have a really amazing grasp of these rules at an incredibly young age.
24:36
We're going to see an empirical study tomorrow from Burko that tests this empirically on young children. We're going to give them a nonsense word, ask them to generate a plural of the nonsense word, and they're going to apply rules to do that. So we'll see tomorrow from the article by Burko that their knowledge of how to generate that--
25:02
the plural form of the nonsense word can't be based off of prior experience or just from conditioning, stimulus response conditioning. It can't be based on behaviorism because children never were exposed to those nonsense words before. So their ability to generate correct plural forms and other forms shows that they're applying a rule, an abstract rule, to that novel instance.
25:30
So here, what we see in this fun example is that one of the rules that we can apply in language is recursion.
25:42
This is one of the truly amazing properties of natural language because what recursion allows us to do is demonstrate this property of productivity of language. And what we mean by productivity of language is that I can create novel sentences. In fact, I can create infinitely novel sentences by just adding something new to a previous sentence.
26:12
And we can do this in a rule-governed way, and I show you how to do this in a rule-governed way with this example of recursion with-- by using the VP, under the VP what that clause is. In the next slide, I'll show you that we can also apply recursion under the NP, but in this first slide, I want to show you that we can apply this rule to the structure of sentences to create longer and longer sentences.
26:44
And we see fun examples of this in shows like Friends and when we're talking to each other. This is a central part of theory of mind, knowing what other people are knowing, knowing that they know what you know, and so on.
27:01
Here from Friends, we have the example, "They don't know that we know they know we know."
27:08
And what we see here is that, in principle, the, "We know they know we know," that can go back and forth. We can reiterate that process, in principle, infinitely many times. Of course, in practice, we have memory constraints, and we have life constraints. So in practice, we can't generate infinitely long sentences. But in terms of principles, linguistic principles, we see that we have a rule that allows us to apply increasingly more that clauses under the VP.
27:46
And when I teach syntax-- I taught syntax last semester, but if I teach syntax here at Berkeley, I'll show you how to-- all the rules for how to construct these sentences. But here, the main point is that notice here we have a simple sentence, "Batman survived."
28:06
We can create a more complex sentence, "Darkseid knew that Batman survived." And the way that we do that is with the rule that, under the VP-- the VP can-- the verb phrase can include just a verb like survived.
28:23
The verb phrase can also include what's called an S bar or something that takes the that clause.
28:30
And then with the S bar, I can have a that clause and a sentence, and then the rules for sentences is that sentences take noun phrases and verb phrases.
28:40
So we see that, with a certain set of rules, we can use those rules to create infinitely long sentences. So I showed you how to create the sentence, "Darkseid knew that Batman survived." But we can just do that again to create the even longer sentence, "Barbados knew that Darkseid knew that Batman survived." And I can get really ridiculous with this and have, "The Joker knew that Barbados knew that Darkseid knew that Batman survived." And I can have another one, "Catwoman
29:15
knew that Joker knew that Barbados knew that Darkseid knew that Batman survived." And you understand what that means. In the most recent example, where I started with Catwoman, you know that this claim is really about the epistemic state of Catwoman. Catwoman knows.
29:32
So even though you've never heard those specific particular sentences before, you easily understand them. Why? Because you have these rules, at least implicitly, in your mind. As linguists, our job is to explicate them to make sense of them and to provide sort of like an economical configuration for those rules. But the important thing is, implicitly, we have those rules in our minds, not just as adults but as children.
30:06
This is how we are able to learn words and learn language so quickly and robustly is because, in some sense, language learning is already in our minds and the only job of experience is to turn that system on and to set certain parameters for that system. Just like for our retina, we're going to need certain stimulus to get vision going, but once we have vision going, we don't need more experience. As the sort of module, information processing module,
30:45
vision is sort of good to go on its own.
30:48
The idea is that we have a language module that's sort of like that, too, in that it needs something to turn it on and to set parameters.
30:59
This is an English language we're learning as opposed to a Japanese language, so we might switch word order or tune the parameters for the specific language. But in some sense, we have a universal set of rules, and then we can just, for each rule, turn it, set it to a certain dial.
31:21
For feature A, I turn it left. For feature B, I turn it right. And that just is how experience tunes the parameters. But the rules are innate to our mind, at least according to as Chomsky argues.
31:39
All right, now that we have an example of recursion and we see how recursion can be reiteratively applied under the VP using that clauses, I'll show you another example.
31:52
And this has to do with recursion under the NP, or under the noun phrase, with prepositional phrases.
32:00
This is fun for me to show you just to point out that recursion doesn't require, necessarily,
32:08
embeddings or recursion under the VP only. We can also create increasingly complex structure under the NP, and that's sort of the fun of this example here.
32:23
So here, in this example, what we have is, "The snow in winter in Gotham falls slowly."
32:31
And if we want to break this down in terms of structure,
32:35
we have an NP, which is "The snow in winter in Gotham," and then we have a VP, "Fall slowly."
32:43
Notice that, as an NP, we can switch this whole NP with another NP.
32:53
So another example of a noun phrase that's simpler than this one is just snow.
33:01
Snow is also a noun phrase, and we can have the sentence, "Snow falls slowly." So you see, the role of the NP is to allow instances of that categorical type. And this property is recalled-- is referred to as systematicity.
33:22
Language is systematic because we have categorical labels, like NP, in which I can switch out NPs with other end NPs in a systematic way. I can trade, "The snow in winter in Gotham," which is an NP, for another type of NP.
33:42
Snow or rain falls slowly. And so often we'll use a substitution practice to make sure that our trees have been done correctly.
33:55
Notice also what's really fascinating and illuminating from this example is that, in terms of locality, if you were to put this sentence linear-- so all the words are in linear order next to each other-- notice that Gotham is closest to falls. So the word Gotham is closest to the word falls. So you might think, if we're understanding language purely in terms of local contingencies like, what word occurs before fall, it would seem like, "Gotham falls slowly," would be the natural reading
34:30
for this because Gotham is closest to falls. So if we understood language purely in terms of linear order, we might think that Gotham falls slowly is how to interpret the sentence since Gotham and falls are right next to each other. But we know that that's not the case. When I say, "The snow in winter in Gotham falls slowly," and I ask you, what's falling slowly? Your answer's not going to be Gotham. It's going to be the snow.
35:00
It's the snow that's falling slowly or the snow in winter in Gotham that's falling slowly, not Gotham that's falling slowly. So again, this is to point out the importance of structure, not linear order, for language processing.
35:18
Again, we know that it's this whole unit that falls slowly, not just this individual item.
35:26
And for us to grasp that, for us to draw upon syntactic rules in order to know that, we, in some sense, have to have these rules in our mind is the argument because I know that NPs and PPs go together and those have to tie together with some VP that I interpret this linear order in a structured way.
35:51
So the important point is, although we see sentences in linear order, we don't interpret them that way. We interpret them in terms of their hierarchical structure.
36:05
And in order for things to be cognizant structurally-- hopefully, you're following with me now. In order for us to colonize things structurally, this is a truly cognitive action or thing that we're engaged in. This can't be pure stimulus response. This has to do with categorical systematic replacements and rule following.
36:32
Awesome. Let's keep it going.
36:35
Not only do we have recursion-- recursion is a nice way to show the productive nature of language.
36:47
And the replacement of NPs with other NPs and the replacement of VPs with other VPs is a great example of the systematicity of language.
36:57
When you grasp the form of sentences-- excuse me-- you can now understand novel sentences that you've never experienced before. If you get the structure of this sentence, "The snow in winter in Gotham falls slowly," you can now apply that rule-based structure to understand other sentences like "The clan of ninjas in Tibet fights fiercely," or something like that. I just made that up-- made that example up. But you see that even though you've never
37:34
heard that structure before, you know that it's not Tibet that falls or fights fiercely. It's the clan of ninjas from Tibet that fights fiercely.
37:47
All right. As you can tell, I've been watching some Batman.
37:51
Okay. In addition to recursion-- this is-- one of the key examples that Chomsky points out for the rule-like nature of language is question formation. We have a really cool ability to take declarative sentences-- sentences of the form, "The Joker was laughing," and forming questions from that statement.
38:18
So this is the tree structure for "The Joker is laughing" And we have a whole day devoted to syntax, so don't worry if this seems like it's moving a little bit quickly. The goal for today is not to memorize tree diagram structure. That's for another week. Today is just to get an idea that language is not just linear ordering of stimuli. Okay.
38:44
So another case in which we see the clear rule-like nature of language has to do with question information. And the way that we form polar questions-- polar questions are yes/no questions-- is we apply a movement operation to the statement.
39:02
So we take a statement like "The Joker is laughing," and we're going to perform one movement operation to turn that statement into a polar question.
39:13
So here's the tree diagram for "The Joker is laughing." And notice here that this right here is the same as this right here except for one thing that I added, or a few things that I added. One is that I turned-- I took the "is" from the auxiliary verb position-- because here "laughing" is the verb; "is" is the auxiliary verb. So to form a polar question, we take the auxiliary verb, the "is," and we're going to front it. We're going to move it to the front of the sentence.
39:51
And we'll attach that with an S-- using an S-bar. We'll use-- I'll just call it a clause here.
40:00
We can get into the details and syntax, but for convenience we'll just call this the clause and this the sentence. And we see that by using an S-bar connecting a clause in the sentence that now we have the question version of the statement here.
40:18
And we see now that this is a general way to form polar questions from statements. And that's pretty awesome because now what's incredibly important here is that this rule--
40:35
the rule that we apply in this instance-- is a general rule. It not only helps us understand this particular utterance, but it's something that will help us understand any utterance of this form.
40:49
And, again, this is what's helping us to have a potentially infinite number-- an unlimited limit of ideas that we can contemplate, which I think is truly fascinating as humans-- is we can think of anything in terms of creative structures. And this is one of the reasons why we can do that.
41:13
Cool. So we see that here "Is the Joker laughing?" which is the sentence we're going for. It's formed by performing a movement operation on the sentence right here. Okay.
41:26
We see that-- we also have rules for forming more complicated questions. So, in addition to polar questions-- which are questions in the form of yes/no-- in response to any polar question like, "Is the Joker laughing?" you can respond yes or no. "Yes, he's laughing." "No, he's not."
41:45
For Wh-questions, the question is not-- the answer to this question is not yes or no but requires some more contentful answer.
41:57
We call it a Wh-question because usually these are of the form like "Who did this?" or "Where was this?" or "When was this?" And these tend to start with W-H.
42:12
So, here we have two movement operations to form Wh-questions.
42:19
So I have the statement on the right on the slide. We see that the statement is "Batman is searching for Bane." So that's a factual declarative statement. Okay. We can now take this statement and turn it into a question. "Who is Batman searching for?" And we see that in order to do that, we're going to perform two movement operations.
42:49
We're going to move "Bane" over to here-- who. You see that-- it's almost like what becomes null or empty here--
43:06
we're going to attach this null operator right here. This is just creating a blank position
43:13
and moving that over here in the front position.
43:18
So when we want to create a who question-- like "Who did this?"-- well, the answer to the who is a person, like Bane or Batman.
43:27
So in order to create that who, we've got to find the Batman or the Bane-- that person-- and we're going to replace it with a null operator and move it over here. And we'll now turn this into who.
43:46
So "who" is in the front and what was "Bane" is now replaced with a null operator. Okay. We also have to do one more movement operation. And that has to do with the "is" because the question is, "Who is Batman searching for?" So what we do is we move the "is" from the auxiliary verb position-- we move it here.
44:11
And by making-- by those two movement operations,
44:16
we now have a more complex question-- a complex question where the answer would be "Bane."
44:24
"Who is Batman searching for?" And that's the tree structure for that question that's derived from this statement.
44:34
And notice, what's really cool about this example is that the movement operations here are not local. But there's an inner one that goes from the auxiliary verb here to the clause position here. And there's another movement that goes from the noun here to the clause position here. So in order for you to track movements and understand how to create relevant questions from statements, you have to have something like a rule system for language in your mind.
45:10
And, again, this is-- the argument is-- this is impossible to explain via stimulus response conditioning or through pure behaviorism. Behaviorism, in some sense, doesn't have the explanatory power to get you things like interpreting sentences of potentially infinitely length due to recursive operations because we've never experienced those. We were never conditioned for those, as of yet, unexperienced possibilities.
45:41
All right.
45:43
In syntax, I have like so many examples. I think these are so fun.
45:50
But I think that is sufficient to motivate the argument for right now. Okay.
45:56
All right. So now that we have a grasp on the rule-like nature of language, we, in some sense, have to have a rule system of language. This is what allows language to be systematic and productive. And this systematic and productive nature of language is what enables us to be such fascinating linguistic creatures. Okay. Now let's go into some of the explanatory positions for accounting for this view of language as perhaps based on a cognitive module that is innately consisting of rules.
46:34
So one idea-- well, two important ideas from Jerry Fodor-- one has to do with the modularity of mind, and the other has to do with the language of thought hypothesis. So I'm going to go ahead and talk about these in turn.
46:48
First, let's talk about the modularity of mind idea.
46:53
This has to do with the idea that the mind consists of mental modules or mental faculties. So, again, this is not the tabula rasa, not the idea that the mind's a blank slate, but the other view that the mind is actually contained with many modules. And each module is sort of like an information processing unit that specializes in a particular processing task. So the language module would be specialized for language processing, whereas the vision module would be specialized for visual processing.
47:28
And the idea here is that we have modules devoted to specific tasks cognitively, just like physiologically we have different modules specialized for different tasks. The heart is specialized for pumping blood through the body. And the liver is specialized for other things, like cleansing your system, and so on. They are distinct. And they perform different actions.
47:53
The idea here-- at least what I want to highlight for today-- is that as cognitive modules--
48:02
what we want from cognitive modules is for them to have certain properties. So what do we mean by a cognitive module? Well, in the Fodorian sense of cognitive module, modules have these properties--
48:16
Fodor includes more properties than this, but these are the most important properties for us to consider today.
48:25
One is that cognitive modules are domain specific. For example, the vision module is specific for the domain of vision and not for audition or olfaction.
48:40
Two, cognitive modules are informationally encapsulated. This means information from visual processing is distinct and encapsulated from other kinds of information processing, such as olfaction.
48:56
And third, cognitive modules are innately endowed. All normally functioning humans are naturally endowed with this kind of information processing module. You have vision. You have audition. You have language as humans, as natural instances of the humankind.
49:19
And these are some of the most important properties of cognitive modules. So in this view, a language module-- if Chomsky is right, that we must think of the mind as consisting-- as having a language module in order to account for language acquisition, then these are the properties that that language module must have.
49:44
Further, another important contribution from Fodor is the language of thought hypothesis.
49:53
The idea here is that language is not just an external thing. It's not even primarily an external thing. On this view-- on the language of thought hypothesis-- excuse me-- it's our thoughts that's sort of the most important thing. And our thoughts are structured, just like language is structured in a hierarchical organization so we can produce new utterances. We can produce new utterances and understand new utterances. This is even more so with internal thought
50:35
is that our thoughts are structured. Our thoughts are not just independent parts, or independent points, but our thoughts are structured.
50:48
And this is why we call thoughts--
50:52
well, this is why we apply the term "language of thought" to our thoughts. It's not that we think in English or that we think in French. That's not the view here. Although it might be true that we do, the view here is not that we think in those specific natural languages. The view here by Fodor is that the structure of our thoughts are structured like language is structured. The emphasis here is on structure, not on that it's English, or that it's French, or whatever.
51:26
So to make this super clear, we often use the term mentalese to refer to the language of thought. And this is just to make it very clear that when I learn,
51:39
I have an innate mentalese. And then when I learn a language like English, I'm in some sense mapping English onto mentalese. So there's already an internal structure in my mind, and I learn a natural language by mapping that natural language into my mentalese.
52:02
So the important thing is that on this view-- the language of thought hypothesis-- a central aspect of our cognition
52:10
is that it's structured like language.
52:14
And notice, if you're a cognitive scientist that believes that cognition is structured in mentalese, then of course it's very clear why studying language even at the syntactic level is very important for the field of cognitive science. It may be that the nature of thought as such-- that the nature of cognition as such is structured in a way similar to how language is structured.
52:41
For example, there's some really interesting recent work on exploring the idea that maybe how we understand events--
52:51
our cognition of events is based on something like a syntax.
53:01
All right. Now that we've considered Jerry Fodor's idea of the language of thought and his criteria for cognitive modules, next Let's go ahead
53:15
and gain a clear understanding of the distinction between behaviorist psychology, cognitive science, evolutionary psychology, and dynamical systems. We wrote another article for today
53:30
by Kenrick, entitled "Evolutionary Psychology, Cognitive Science, and Dynamical Systems."
53:40
So we just discussed some of this information. But just to be super clear, this is all provided on one slide. So what is behaviorist psychology? Who was a major contributor of this approach in psychology? Well, we know BF Skinner.
53:59
And this approach has focused on observable stimuli and observable behavioral responses and dismissed all talk of consciousness and mental representations.
54:12
Cognitive science, on the other hand-- as championed by Noam Chomsky, Jerry Fodor, and others--
54:20
has focused on common processes underlying intelligent problem solving in computers and in humans.
54:31
Evolutionary psychology-- this approach, we'll look at research by Tooby and Cosmides. We'll also look at some work by Buss and some other evolutionary psychologists-- Eckman, for example, on facial expressions. And we'll see that their focus is on particular cognitive mechanisms that are adapted to particular ecological contexts.
55:01
And, finally, we have the field, or the approach, of dynamical systems. And this focuses on how complexity or complex systems,
55:12
complex behaviors emerge from a few variables interacting in accord with a few basic rules.
55:26
All right. The last thing I want to cover today
55:30
has to do with levels of analysis. And we briefly touched upon this earlier in the lecture when we're understanding the strengths of cognitive science as opposed to the other subfields. And we saw that whereas other subfields focus on certain levels of analysis, cognitive science can try to draw upon research at different levels and offer an integrative understanding of the mind-- for example, how language functions in the mind. Okay. Marr-- David Marr--
56:05
he published this book Vision-- A Computational Investigation into the Human Representation and Processing of Visual Information. But in this book, what's important is not just his contributions to vision, but also to his approach to understanding the different levels of analysis.
56:26
So as I mentioned, cognitive science is an excellent approach to studying the mind and intelligence systems since it draws upon subdisciplines that focus on distinct levels of analysis.
56:39
Example-- philosophy and neuroscience both focus on the mind in some way but from different levels of analysis.
56:48
Marr proposed a useful way for understanding these distinct levels of analysis.
56:55
Let's take a look.
56:57
This is a nice figure from his book Vision. In this figure from chapter 1, we see that there are three levels at which any machine-- biological or non-biological-- carries out an information processing task.
57:14
The three levels at which that information processing tasks can be understood are in terms of the computational level, the algorithmic level, and the implementation level.
57:27
Or as provided here in the three columns, we have the computational theory. And what this is about is--
57:36
it focuses on this question. What is the goal of the computation? Why is it appropriate? And what is the logic of the strategy by which it can be carried out? So what's the goal of computation? This might be, like, the ecological problem that you're trying to solve. Maybe you need to fly. Maybe you need to run. Maybe you need to avoid a predator. It's sort of the goal of the computation.
58:05
And philosophers can be good at thinking about this as well as evolutionary psychologists.
58:10
The next level is the algorithmic level.
58:15
And this focuses on the questions-- how can this computational theory, be implemented? In particular, what is the representation for the input and output? And what is the algorithm for the transformation?
58:32
And so this has to do with more, like, what's sort of the program? What's the representational content like?
58:40
Are we operating off of a Turing machine, off of a neural network? What's the algorithm that can be used at an abstract level to perform or achieve the goal at the computational level?
58:56
So you might have a goal for your computation to perform, like I need to fly.
59:05
That's going to be the computational level. The algorithmic level is, in an abstract sense, what do I need to do to accomplish flying? And notice that this might be represented at a level that is somewhat abstract.
59:23
Then the third level is the hardware implementation level. So the algorithm is more like the software implementation. Here, the hardware implementation is more like, okay, now that we know that there's some general rules for how to fly, how do we implement flying in this particular creature or this particular machine? There may be some general rules for how to fly. But given the type of bird or the type of aircraft that you happen to be, how do we make
59:54
it work for you in terms of the particular hardware that you have?
1:00:01
So we see that these are different levels.
1:00:04
Computation-- problem or goal; representation or software type of problem-- how to solve the problem via the program; and then hardware level-- how do we implement that software or that program in a physical machine?
1:00:23
And as cognitive scientists, we're not satisfied with just any one answer. We want to have a systematic understanding and, ideally, integrate all three levels.
1:00:34
This is a very nice picture that helps to clarify the distinct levels of understanding as well as sort of the epistemology and the methodology behind it.
1:00:49
So to make this clear, what's the computational level of understanding?
1:00:56
Well, we know that this has to do with a problem to be solved cognitively.
1:01:03
Cognitively, this bird needs to fly. That's sort of the goal-- the ecological objective that has been imposed on this entity.
1:01:13
So cool. We know that what it needs to do is this. It needs to fly.
1:01:21
Now that we know what this thing needs to do, there's got to be some rules for flying. So maybe we're going to-- we need to flap at a certain rate, or we need to do some sort of flapping behavior to maintain lift or elevation.
1:01:40
Cool. So to do this, we got to do this.
1:01:45
Excellent Now that we know flapping must be accomplished in order to accomplish lift and flight-- great-- how do we implement it in this type of entity? Well, this type of entity happens to have feathers. So that's what's going to be used to flap.
1:02:04
And we see that these three levels focus on different things but tie together at this entity-- for this entity. For us to have comprehensive deep understanding of this entity and its cognitive abilities, its helpful for us to understand its computations to be performed; its algorithms to perform that-- achieve that computation; and how those algorithms are implemented in the hardware, or in our biology, or in the mechanics of a robot.
1:02:38
Cool.
1:02:39
Also notice that this provides us with a system of explanation and understanding.
1:02:46
So here, in B right here, we see that one, the problem, or the computation, suggests two. A problem suggests rules. So I need to fly? Okay, well, that suggests some sort of flapping behavior.
1:03:05
Two, the rules predict the implementation. So now that I know that flapping must be accomplished in order to fly, I've got to look for some part of the machine that can do that--
1:03:17
that can implement that procedure. And boom, I see that the feathers can serve as the material instantiation or implementation for that rule-based program. So sweet. Two predicts three.
1:03:35
Next, what we can do-- and see how now we're going from something like a philosophy and neuroscience-- going back this way, we can use three to confirm two.
1:03:48
So by implementing flapping--
1:04:00
for example, by building a robot with feathers, we can now see if feathers actually allows this rule to be carried out. So remember before I talked about how artificial intelligence and modeling can be very helpful for us to understand intelligence. And this is where I'm talking about.
1:04:24
At the top, we saw how we can go from a problem to the implementation. And here, we see how we can go from implementation to further confirming our theory. So we might have--
1:04:39
in order to develop a firm understanding, we can build a system that has feathers. And that can confirm for us that this entity with feathers--
1:04:56
it does provide the sufficient implementation components to enable the rules-- to enable the flapping to occur. So that's what we mean by three confirms two.
1:05:12
Remember this was at a more abstract level, like at a software-rule level. And so we want to use physical models to confirm that that's true.
1:05:24
And then we use two to explain one. Now that we've created a robot that has feathers and we see that that robot can indeed fly,
1:05:38
we can now use that to explain its ability to solve a computational problem, which is the problem of achieving flight.
1:05:49
So, again, to be very clear, the computational problem suggests an algorithm or rule for solving it. The algorithm or rule for solving a problem predicts some sort of implementation.
1:06:06
Then, we can build such a system with the components and see if it does actually accomplish those rules--
1:06:19
the algorithm of flight or flapping.
1:06:22
And, if so, then that explains how this computational problem can be solved.
1:06:32
Finally, we see right here-- this Is very easy to explain. It's just that we have an epistemological bias towards three-- the implementation level. Like, we love neuroscience. When we see, in scientific articles, pictures of the brain, that's very convincing for us. So this is just to point out that we have an epistemological bias for that with which we can manipulate. And that which is at a more--
1:07:03
at the level of understanding rather than physical manipulation, we have less of a bias towards those.
1:07:10
So this is in the realm of philosophers and theorists. And that's not to discount this level. As we see here, that level is actually an essential component. We use theory to help drive ideas for appropriate algorithms and appropriate implementations for those algorithms. So we see that theory is very important for driving research. We just have a bias towards what's manipulable, what's at the implementation level.
1:07:43
And although this is very easy to understand for this toy model right here, a bird flying, the important thing is this is to help us understand any type of cognition--
1:07:58
whether it's language, whether it's the cognition of a robot-- is that we can understand cognition information processing in these three levels.
1:08:10
And I hope by seeing this, this helps to reinforce the point from cognitive science that's very important and useful to draw upon all of the subfields to paint the best picture can of how the mind performs some sort of cognitive task-- the one that's dear to me, of course, language, but also music, emotions, and facial expressions, which we'll study throughout the course. So as we journey through the course, ask yourself questions. What's the role of cognition in the particular task?
1:08:49
And then, how can I think about cognition in terms of these three levels?
1:09:01
The last slide for today--
1:09:10
I'm going to pause one second. I'm running out of battery.
1:09:15
All right. So my final point for the lecture today is that philosophy may be a useful approach for thinking about the mind at the computational or algorithmic levels but does not necessarily offer insight into the implementation level of understanding.
1:09:35
Neuroscience, on the other hand, may be a useful approach for investigating the implementation level, but this does not necessarily offer us insight into the computational level of understanding. So, by drawing upon the theoretical-computational strengths of philosophy and the empirical-implementation strengths of neuroscience, we can advance our understanding of the mind.
1:10:00
So that covers our lecture today on Introduction to Cognitive Science-- what the field is about, how it differs from some of the subfields. Hopefully you've also gained some clarity on what it means for something to be cognitive and the different levels of understanding what a cognitive function consists of.
1:10:23
Thanks for your attention. I hope you enjoyed the lecture. I look forward to seeing you tomorrow for our next lecture.
Here to help you learn
Sign in to keep a private conversation with your course tutor.
Ask me anything about the material…
Sign in to chat