1. SEL2211: Contexts Lecture 6: The computational theory of mind Humans are conscious in the sense that each of us has an awareness of selfhood. We interact with the world using our senses and our capacity for movement, and on the basis of that interaction we come to have knowledge of and beliefs about the world, which in their turn motivate further interaction. In short, humans have minds. To understand what minds are, and how they work, is to understand what it is to be human. This lecture briefly outlines the history of the study of the mind, then goes on to look at contemporary computational interpretations of mind and language, and concludes with objections to these interpretations.

2. SEL2211: Contexts Lecture 6: The computational theory of mind 1. Understanding the mind: a brief overview From the earliest historical times, and presumably before that, humans have wondered what it is that differentiates them from other animals --what makes them human-- and this has led them to speculate on the nature of what we now call the mind. Evidence that this was so comes from all over the world. For the purposes of this discussion, however, we will concentrate on the Western tradition for the simple reason that it leads directly to contemporary cognitive science, which is our object of interest.

3. SEL2211: Contexts Lecture 6: The computational theory of mind 1. Understanding the mind: a brief overview 1.1 Ancient Greek philosophy The earliest explicit and coherent Western account of the mind comes from the Greek philosopher Plato (c. 428/7-348/7 BC), who proposed two separate worlds: the physical world of human experience, and a world of non-physical forms or ideas. The world of ideas contains the eternal and unchanging forms of things, and the world of experience is an imperfect and constantly-changing realization of those forms. Human souls (or, in modern terms, minds) originally dwelt in this ideal world, and can regain knowledge of it. This cannot be done solely on the basis of physics, the study of the world of experience, since the objects of study are imperfect realizations of the forms, though physics is useful in that it moves the mind to remembrance of the ideas it once knew.

4. SEL2211: Contexts Lecture 6: The computational theory of mind 1. Understanding the mind: a brief overview 1.2 Rome and the Middle Ages Plato's thinking on the soul was hugely influential, both during the period of the Roman Empire (c.50 BC - c.500 AD), and in the Middle Ages (c.500 - c.1500 AD). Plato's affirmation of a spiritual order of reality fit well with the Christian view that the spiritual aspect of humanity iss of primary importance in life. The great majority of Christian philosophers until the thirteenth century were Platonists, recognizing in Platonism a powerful ally in the fight against materialism. Thereafter the teachings of Plato's pupil Aristotle (384-322 BC) became dominant. Aristotle challenged the Platonic doctrine of ideals, and paved the way for post- medieval thinking about the mind, but the idea that the soul/mind was a non- physical substance, a piece of divinity which enters the body at conception and leaves it on death, has remained fundamental to Christian thought.

5. SEL2211: Contexts Lecture 6: The computational theory of mind 1. Understanding the mind: a brief overview 1.3 The West since c.1500 AD As a consequence both of Aristotelian thought and of the rediscovery of Ancient Greek and Roman texts, Europe after about 1500 AD entered a period of intellectual revolution now known as the Renaissance. The main characteristics of the Renaissance were, on the one hand, a tendency to denigrate religious ideology, and on the other to promote scientific investigation, which was understood as the empirical investigation of nature. This led to rapid advances in such areas as mathematics, astronomy, and physics, together with the development of technology based on the new sciences.

6. SEL2211: Contexts Lecture 6: The computational theory of mind 1. Understanding the mind: a brief overview 1.3 The West since c.1500 AD The mind now came to be seen not as more or less equivalent to a divine soul, but as something rooted in the physical operation of the brain, and thus amenable to study just like any other aspect of the physical world. As scientific thought and technology progressed, the mind was seen successively as a hugely complex clockwork, as a hydraulic engine, and, by the end of the nineteenth century, as a something like a telephone exchange. In the second half of the 20th century, it came to be seen as a computer.

7. SEL2211: Contexts Lecture 6: The computational theory of mind 2. The computational mind 2.1 The sciences of the mind Turing himself proposed that the human mind was a computer. This suggestion was enthusiastically taken up by a variety of scientific disciplines in the 1950s: -- Cognitive science, which attempts to explain the operation of the human mind. -- Generative linguistics. Chomsky has always seen the linguistics he pioneered as an aspect of psychology whose remit was to understand a specific human cognitive function: language. -- Artificial Intelligence, an engineering discipline which attempts to construct computational systems which display aspects of human cognitive capacity, and whose ultimate aim is creation of machines with general human-level intelligence. Since then, Turing’s hypothesis that the human mind is a computer has been dominant in these disciplines.

8. SEL2211: Contexts Lecture 6: The computational theory of mind 2. The computational mind 2.2 The architecture of the computational mind How does the computational mind work? Like a very complex Turing machine:

9. SEL2211: Contexts Lecture 6: The computational theory of mind 2. The computational mind 2.2 The architecture of the computational mind On this model, the senses and muscles are the input and output tapes of the Turing Machine, and there are numerous algorithms to control every aspect of cognitive functioning: logic, memory, perception, movement, and language. In each case, of course, the algorithm must be extremely complex to account for human behaviour, but the principle remains the same as for the very simple Turing Machine we looked at earlier.

10. SEL2211: Contexts Lecture 6: The computational theory of mind 2. The computational mind 2.3 Mental algorithms What does a cognitive algorithm look like? The truth is that, in the present state of cognitive science, no one knows for sure, but we do have some reasonable theories. One such theory is based on the Language of Thought (henceforth LOT) hypothesis, which proposes that thinking is done in an internal language which has come to be known as 'mentalese‘. Assuming this to be the case, algorithms for the various cognitive functions would specify how the sentences of mentalese are constructed and interpreted. This section considers what a language is, then what a language of thought might look like, and finally what form an algorithm for the language of thought might take.

11. SEL2211: Contexts Lecture 6: The computational theory of mind 2. The computational mind 2.3 Mental algorithms 2.3.1 Language What is a language? From a computational point of view, a language is just a set of strings, that is, a set of symbol sequences where the symbols are taken from a symbol system. Some examples (the brackets {...} denote a set): L1 = {heajk, hf, hju, mh, hfrqiqjhi} L2 = {y=ax+b, y=ax 2 +bx+c, y=ax 3 +bx 2 +cx+d} L3 = {the cat sat on the mat, the man ran}

12. SEL2211: Contexts Lecture 6: The computational theory of mind 2. The computational mind 2.3 Mental algorithms 2.3.1 Language L1 = {heajk, hf, hju, mh, hfrqiqjhi} L2 = {y=ax+b, y=ax 2 +bx+c, y=ax 3 +bx 2 +cx+d} L3 = {the cat sat on the mat, the man ran} L1 consists of strings of symbols taken from the Western alphabet. The strings are gobbledegook, but that doesn't matter: a language is defined as a set of strings, as this is, without reference to what the strings might mean. L2 is a set of strings of symbols taken from the mathematical symbol system. L3 is a set of strings of symbols taken from the symbol set of English words. These and other languages can contain as many strings as one likes --natural languages, for example, are unbounded, as is the set of possible mathematical expressions-- but the principle is always the same: a language is a set of strings.

13. SEL2211: Contexts Lecture 6: The computational theory of mind 2. The computational mind 2.3 Mental algorithms 2.3.1 Language A language can consist of all possible combinations of all the symbols in the associated symbol system. In English, for example, one could have strings like the the the the the or rest bad walks the already, and so on. In general, though, there are constraints on which strings belong to any given language and which do not: the English word sequences just cited don't belong to the English language, but Bill ran home and I won the lottery do.

14. SEL2211: Contexts Lecture 6: The computational theory of mind 2. The computational mind 2.3 Mental algorithms 2.3.1 Language How can one specify which strings belong to any given language and which does not? By establishing a set of rules by which the strings that belong to the language may be formed; all strings not constructed according to the rules are excluded from the language. Such sets of rules are known as grammars. There are various ways of stating grammars, but one familiar to linguists is rewrite rules.

15. SEL2211: Contexts Lecture 6: The computational theory of mind 2. The computational mind 2.3 Mental algorithms 2.3.1 Language The following grammar generates (among others) the strings that constitute L3: {the cat sat on the mat, the man ran} S > NP VP NP > DET N VP > V VP > V PP PP > P NP DET > {the set of English determiners} N > {the set of English nouns} V > {the set of English verbs} P > {the set of English prepositions}

16. SEL2211: Contexts Lecture 6: The computational theory of mind 2. The computational mind 2.3 Mental algorithms 2.3.1 Language In this grammar there is one rewrite rule per line. Rewrite rules are so called because the symbol on the left of the '>' can be rewritten as the symbol of symbols on the right. A string in L3 is constructed by starting with S and then rewriting it according to the grammar. For example:

17. SEL2211: Contexts Lecture 6: The computational theory of mind 2. The computational mind 2.3 Mental algorithms 2.3.1 Language S NP VP DET N VP the N VP the cat VP the cat V PP the cat sat PP the cats sat P NP the cat sat on NP the cat sat on DET N the cat sat on the N the cat sat on the mat S > NP VP NP > DET N VP > V VP > V PP PP > P NP DET > {the set of English determiners} N > {the set of English nouns} V > {the set of English verbs} P > {the set of English prepositions}

18. SEL2211: Contexts Lecture 6: The computational theory of mind 2. The computational mind 2.3 Mental algorithms 2.3.1 Language S NP VP DET N VP the N VP the cat VP the cat V PP the cat sat PP the cats sat P NP the cat sat on NP the cat sat on DET N the cat sat on the N the cat sat on the mat S > NP VP NP > DET N VP > V VP > V PP PP > P NP DET > {the set of English determiners} N > {the set of English nouns} V > {the set of English verbs} P > {the set of English prepositions} Successive application of the grammar rules has generated a string of the language; the man ran in L3 would be similarly generated, as would many other English strings.

19. SEL2211: Contexts Lecture 6: The computational theory of mind 2. The computational mind 2.3 Mental algorithms 2.3.1 Language Generating strings in accordance with a grammar of this sort endows them with constituency structure, which can be represented graphically as a tree: A string, therefore, is a structured object, that is, and object consisting of parts which may themselves consist of parts which may themselves consist of parts, and so on.

20. SEL2211: Contexts Lecture 6: The computational theory of mind 2. The computational mind 2.3 Mental algorithms 2.3.1 Language The structuredness of strings is of fundamental importance in how language algorithms work. To see why, consider the string the saw the man with a telescope. This string is ambiguous. There's no way to tell which meaning is intended from the string alone, but the combination of string and structure allows it to be disambiguated.

21. SEL2211: Contexts Lecture 6: The computational theory of mind 2. The computational mind 2.3 Mental algorithms 2.3.1 Language A language algorithm both endows a string which structure in the course of generating it, and uses the structure associated with a string in the course of interpretation.

22. SEL2211: Contexts Lecture 6: The computational theory of mind 2. The computational mind 2.3 Mental algorithms 2.3.2 The Language of Thought The language of thought proposes that the various aspects of cognition are generated by algorithms operating on structured symbol strings in a way that is sensitive to their structure. What do these strings look like? Again, no one knows for sure at the present time, but given that symbols are physical, one plausible suggestion is that mental symbols in the LOT are neural ensembles, that is, assemblies of neurons whose connections cause them to behave as a group and thereby to constitute a symbol.

23. SEL2211: Contexts Lecture 6: The computational theory of mind 2. The computational mind 2.3 Mental algorithms 2.3.2 The Language of Thought In the following graphic, for example, each ensemble has a different pattern of connectivity and therefore a different firing pattern; each can be taken to be a symbol in the LOT, and a string would be a sequence of different patterns of activation over time.

24. SEL2211: Contexts Lecture 6: The computational theory of mind 2. The computational mind 2.3 Mental algorithms 2.3.3 Algorithms of the Language of Thought The algorithms of the LOT are almost totally mysterious. We know how neurons and neural ensembles work, and we can observe them working by means of the scanning techniques mentioned in the previous lecture, but we are not yet in a position to know what the pattern of activity we observe is doing, that is, we don't yet know the brain's algorithms. Some animations: Youtube. MRI Brain activity: http://www.youtube.com/watch?v=H-De5nOC8K0http://www.youtube.com/watch?v=H-De5nOC8K0 Youtube. Spontaneous brain activity: http://www.youtube.com/watch?v=3nCBLw9Z-xU&feature=relatedhttp://www.youtube.com/watch?v=3nCBLw9Z-xU&feature=related Youtube. Music processing: http://www.youtube.com/watch?v=FVIiYyD2_YU&feature=related http://www.youtube.com/watch?v=FVIiYyD2_YU&feature=related

25. SEL2211: Contexts Lecture 6: The computational theory of mind 2. The computational mind 2.4 The Language of Thought and Natural Language Proponents of the cognitive view have included such luminaries as Wilhelm von Humboldt, Vygotsky, Wittgenstein, and Daniel Dennett. The idea that thought and language are causally interrelated is, however, primarily associated with the American linguists Edward Sapir and his pupil Benjamin Whorf, whose stance on this idea is best summed up by a much-cited quotation from Whorf: “We dissect nature along lines laid down by our native languages. The categories and types that we isolate from the world of phenomena we do not find there because they stare every observer in the face; on the contrary, the world is presented in a kaleidoscopic flux of impressions which has to be organized by our minds - and this means largely by the linguistic systems in our minds. We cut nature up, organize it into concepts, and ascribe significances as we do, largely because we are parties to an agreement to organize it in this way - an agreement that holds throughout our speech community and is codified in the patterns of our language. The agreement is, of course, an implicit and unstated one, but its terms are absolutely obligatory; we cannot talk at all except by subscribing to the organization and classification of data which the agreement decrees.”

26. SEL2211: Contexts Lecture 6: The computational theory of mind 2. The computational mind 2.4 The Language of Thought and Natural Language On the basis of this and other passages in the various writings of Sapir and Whorf, the linguistics research community has constructed the Sapir-Whorf hypothesis, which proposes two associated principles: -- Linguistic determinism, whereby thinking is determined by language – the language that a person speaks determines the way he or she interprets the world. -- Linguistic relativity, whereby people who speak different languages perceive and think about the world differently. The communicative view – that language is purely a mechanism for the communication of independently-existing thought – is based on the computational model of the mind. It was held by, among others, John Locke and Bertrand Russell, and has been standard in cognitive science in the second half of the twentieth century.

27. SEL2211: Contexts Lecture 6: The computational theory of mind 2. The computational mind 2.4 The Language of Thought and Natural Language The standard communicative view of the relationship between thought and language is now being challenged, most recently by Dennet and Carruthers. Both argue that the cognitive view is no less inherently plausible than the communicative one, and that, like the latter, it has substantial empirical support. Carruthers goes on to propose that language is only involved in a specific type of thought – conscious thought. This, of course, begs the question of what a conscious thought might be, but a rough approximation is that it includes the traditionally ‘higher’ functions like reasoning, planning, and memory, and excludes ‘lower’ ones like perceptual and motor functions.

28. SEL2211: Contexts Lecture 6: The computational theory of mind 3. Objections to the computational theory of mind and language In the late 1980s, the consensus that human minds, and language more specifically, are computational systems began to be questioned on a variety of grounds, both theoretical and practical. We will look at the main theoretical grounds here..

29. SEL2211: Contexts Lecture 6: The computational theory of mind 3. Objections to the computational theory of mind and language 3.1 The essential problem: symbols, meaning, and the mind At the root of the theoretical objections to the computational mind is the nature of symbols. Symbols are arbitrary relative to that which they represent: the meaning of a symbol –what is represents- is in the mind of the human who interprets it. Why is this crucial to the hypothesis that the mind is a computer? The argument is as follows:.

30. SEL2211: Contexts Lecture 6: The computational theory of mind 3. Objections to the computational theory of mind and language 3.1 The essential problem: symbols, meaning, and the mind i. Computers manipulate symbols, but they do not themselves interpret them –they do not know what the symbols they manipulate mean. Interpretation of a computer’s response to input is in the mind of the human observing it. A computer does not know the meaning of an essay typed in from a keyboard, nor does it secretly read emails, nor does it take pleasure in space invaders games. The meanings of these things arise in the mind of a human in response to observation of the monitor display.

31. SEL2211: Contexts Lecture 6: The computational theory of mind 3. Objections to the computational theory of mind and language 3.1 The essential problem: symbols, meaning, and the mind ii. There is no meaning in a computer, but there IS meaning in a human mind. A computational explanation of cognition on its own is therefore insufficient: it cannot explain how meaning arises in the mind.

32. SEL2211: Contexts Lecture 6: The computational theory of mind 3. Objections to the computational theory of mind and language 3.1 The essential problem: symbols, meaning, and the mind The lack of meaning in a computational system is analyzed as the 'symbol grounding problem': a computational system consists of symbols and symbol manipulation, and only these things, and cannot by definition incorporate that which the symbols represent.

33. SEL2211: Contexts Lecture 6: The computational theory of mind 3. Objections to the computational theory of mind and language 3.2 Proposed solutions Researchers have taken different approaches to the solution of the symbol grounding problem. -- Some, wishing to retain the traditional computational model of mind, have argued that meaning somehow arises from the interaction of the various computational modules of mind, though it is difficult to see how this could be.

34. SEL2211: Contexts Lecture 6: The computational theory of mind 3. Objections to the computational theory of mind and language 3.2 Proposed solutions -- Others have suggested that not all of cognition is computational -that only certain modules and perhaps only language are. In particular, the brain's processing of sensory and motor input/output is not symbolic, and thus not computational. By associating linguistic symbols with such non-symbolic brain processes, language can be grounded in bodily experience of the world, and this gives rise to meaning in the cognitive system as a whole.

35. SEL2211: Contexts Lecture 6: The computational theory of mind 3. Objections to the computational theory of mind and language 3.2 Proposed solutions Thus, for example, the word ‘cat’ means our experience of cats in the world in terms of how they look, sound, behave, and so on; once word meanings are grounded in experience in this way, sentence meanings can be built up compositionally by symbol manipulation. This development of computational cognitive science is called ‘embodied cognition’, and is currently being intensively developed by psychologists and cognitive science researchers, though it has so far failed to make any real impact in generative linguistics

36. SEL2211: Contexts Lecture 6: The computational theory of mind 3. Objections to the computational theory of mind and language 3.2 Proposed solutions -- Still other researchers have concluded that minds, including the language faculty, are not computational, and have proposed alternative theoretical frameworks for understanding cognition and language. These alternative approaches stress the study of brain mechanisms using MRI scanning technology, and the mathematical theory of complex systems, which treats brains and their cognitive functions in the same terms as other highly complex natural systems like the cosmos, the weather, and biological ecologies. Some psychologists and artificial researchers are taking this approach, but once again generative linguists show no signs of doing so.