522019 Chomsky Lecture w QA

welcome everyone to the fourth and final lecture in Noam Chomsky's lecture series here at UCLA before Noam starts I just wanted to make a couple of brief announcements first I wanted to acknowledge the support of the Dean of humanities and his discretionary fund to help support this visit I also want to acknowledge two graduate students at UCLA whose idea it was to invite norm to come and speak to us and they actually made the initial move to invite him one of them I think is here Richard Stockwell wherever you are and oh there you are right okay and Nico's Angelo Poulos who unfortunately had to go back to Greece to finish his dissertation but we'll be taping we're taping these and he'll be here in spirit finally I thought I would just share a little piece of trivia that I learned today and that is that Nome has a be named after him do you know about this yeah it's called let me get this mega Kiley Chomsky and it was discovered in Texas in 2013 so now you're all in possession of the same trivia without further ado please welcome Nome for his final lecture I never did find out whether it's a honeybee with a waggle dance but I hope somebody will find that out for me well this has been pretty sketchy especially yesterday there's lots of loose ends that ought to be tied up I hope actually I think it would be best maybe to delay that to the discussion period tomorrow so I hope you guys will come in with lots of questions and objections and will be able to see if we can fix up what wasn't done properly of which there's a lot there was actually one question came up in the question period yesterday about an important matter and I didn't actually give the right answer to it so let me go through it here get this to work so the paradigmatic example that I was giving about what distinguishes the legitimate from the illegitimate ones who can illustrate it with parallel merge the thing that people tend to write like this where you have this is the work space and then you would join this to this and that's supposed to give you this thing which in fact gives you and this one becomes illegitimate because you can explode this to something complex as you like and since this is accessible you can move there then you have a chain connecting these two things which can violate all conceivable properties of language so that's legitimate the argument that it is illegitimate is it adds two instead of one accessible terms namely this one here and this one here and you're only allowed to add one in the minimal system of the resource control so the question was well why doesn't this block ordinary I am so suppose you have this guy over here now we have why can't you do the same thing you think about it for a minute which I didn't when the question was asked this turns out to be a case of late merge what you're doing is trying to tack this into something that you've already test and that doesn't work for the reasons I discussed last time late merge first of all adds too many new accessible items and then it also requires an extra operation of substitution so it's the worst case of illegitimate City so that one's out the difference is once you've intuitively once you've passed a certain point in the derivation you can't go back and do something to what you've passed but that a problem doesn't arise in the parallel merge case because these two guys are separate and nothing is protecting the thing that you moved from so that makes the right distinction well there are a lot of other things that need to be clarified but instead of that I'd like to turn to something else at least touch on a few other things briefly just to bring them up and won't be able to go into them in sufficient detail one point is that in addition to the symmetrical operation merge there's quite good in fact I think completely compelling evidence that we also need another operation an isometric operation in addition to merge this is clear intuitively simply from simple adjuncts so for example if you have a phrase like young man it's there's an asymmetry between the two that's clear that element that's formed is a noun phrase not an adjective phrase so the old is an adjunct that's not changing the category notice this is quite different from symmetrical merge in symmetrical merge if you have a if you haven't have a head and an XP then the head will in fact be the label the thing that traditionally projects but that's just a case of minimal search and that doesn't work for i junked a symmetry so if you have old portrait of John portrait of John is not a head it's complex could be arbitrarily complex but it's still the label of the of the whole unit so there's a clear asymmetry beyond that there's a problem that's been lingering for 60 years that's the serious problem it's the problem of unbounded unstructured coordination so if you have something like the guy is young happy eager to go to college you know tired of wasting his time endless number of possible adjuncts there's no structure among them they're strung together they're unbounded and that's a real problem for one thing you can't do it in any kind of phrase structure grammar because you'd need an infinite number of rules that's even true interestingly of of the universal case unrestricted rewriting systems the ones that allow you to do anything rewrite anything you'd still need an infinite number of rules notice that unrestricted rewriting systems as was proven by a great mathematician ml post long ago or Universal those are universal Turing machines in effect which means you can code the right result by some devious means but you don't get it by just rewriting the rules and looking at the structure so that's out totally from phrase structure grammar about 60 years ago when george miller and i were working on mathematical linguistics we thought we had an answer to this in terms of generalized transformations but howard lies and it can pointed out that that doesn't work either so there's no method around no device around that allows this however it is allowed immediately simply by the device of a symmetric pair emerge what we need is a new device and since we're trying to get the simplest possible operations the simplest operation after set for ordinary simple set formation is just pair formation so we need an operation pair emerge which will also apply to the simple adjunct case like young man a young will be a join two will be attached to man but you don't see it in the labeling okay because it's often some other dimension and the unbounded unstructured case show you in effect that there were unbounded ly many dimensions to what's going on up there it's not two-dimensional like a blackboard you can add any number of adjuncts at any point so we want some device that we want to we want to work this out that means intuitively that every member of this coordinated construction every individual member of it is individually predicated of what it links to now we know that the order of the unstructured unbounded elements this sequence we know that the order of it matters it matters because of reasons that were pointed out by Jim McCauley about 40 years ago namely notions like respectively so if you say John and Bill's or common Mary respectively the order in which they appear affects the semantic interpretation so it's not just a set of paired things it's a sequence of pair of things furthermore the more evidence that it's a sequence is that adjuncts can repeat so you can say the guy is young tall happy young eager to go to Harvard so on and so forth you can repeat them as much as you want so what we have a situation where in order to generate these objects you generate a set finite set you pick out of you form from that set a sequence and it could be any sequence of elements and there's in fact infinitely many possible sequences you pick one out of those and that sequence s call it is the thing that you're then going to merge into the construction to proceed with the interpretation this operation of picking a particular element out of the set of sequences is the this formal ways of doing it which are familiar those of you who know some logical recognize that this is David Hilbert's an epsilon operator which picks a single thing out of a said it was part of his work on foundations of mathematics basic operation so it's a straightforward operation but it does have the property that it's indeterminate okay so it's that's part of the nature of production as I discussed last time well what this means is actually there's clearly two kinds of coordination conjunction and disjunction so what we're doing is forming objects that look like this we're forming a sequence which begins with some conjunction and then contains a sequence of elements each of which is predicated of something so we have things of sequence of things that look like this link and in fact we couldn't say ever'one and but in fact all the links have to be identical but one of the aspects of coordination is your your attachment or joining everything to the same point okay so we can so we have an object like that that's the basic object that gives you an unbounded coordination when you get down to just one case when N equals one that's just playing a junction so the old young man's on no notice that each element of these pairs is inaccessible so if you say the phrase old man say you can't extract man and leave old that you can't extract the old and leave man so the elements of the pair's are inaccessible and if that's if that were all there was to it notice that this would yield at both the coordinate structure and the adjunct island that constraint you have the coordinate structure constraint because every term is inaccessible you have the adjunct Island extra constraint because you can't pull the things out however quite it's not quite that simple in fact as the I mentioned earlier paper by Jaco Boscovich which shows that there's which just reviews lots of different kind of complicated cases there are languages in which you can extract the odd you can't extract the anything inside the adjunct there are others where you can extract something that's inside but not the edge until and in a couple of other cases so there's more work to be done and what it in fact shows is that this is that the concept of adjunct is just not sufficiently refined there's a number of different kinds of adjuncts which behave quite differently and this is in fact kind of a an unexplored domain it hasn't been looked at sufficiently but you have to here's a major research project is to ask how can we deal with the class of cases that Boscovich left unexplained in his bringing together the two kinds of mysteries there are they're still there so here's a first step and how to capture it but more has to be said well the next question is what is ill what do you link things to so let's take the simplest case say noun phrase and verb phrase coordination you know John bill Tom the young man etc etc they read the book walk to the store so on and so forth what's the linking in those cases well the assumption that comes to mind right away is that saying incidentally I should say that I'm going to assume here that not that nominal phrases are actually NPS the DP hypothesis which is widely accepted was very fruitful that do a lot of interesting work but I've never really been convinced of it I think these things are fundamentally nominal phrases actually is a very good paper by Maus Oishi who's here who kind of spells out and how this could work this would mean that say Ana nominal phrase things like definite articles are actually features of the nominal phrase they're not elements merged into it very much like semitic where it's just a feature of the nominal phrase and that the definite article appears in every element of the nominal phrase including the determiner that seems to me probably the way it works as far as determiners are concerned like say that i suspect they're at so I'll be assuming that the core system is basically nominal as I say Moss's paper spills a lot of this out that's so one suggestion you might have is that L is just n but that runs into an immediate question if you accept L which I am accepting here the Hoggett Barrera Malik Marantz theory of route categorization which i think is pretty strongly motivated you have the roots in the lexicon are independent of category they become nominal or verbal adjectives Allah by virtue of a categorize ur that attaches to them and incidentally that would surely be pair merged so pair emerges to them and it's probably the it has to be the first operation in forming forming a derivation you search first attach the root to the category the category to the root and you sort of go on from there with a pair of merged element but that element say n for a root that's going to become nominal it can't be identified with the end that is the link here because this was there this one is just much higher up in the in the syntactic object that you're constructing same with verb it means that the small V which is the categorize ur which is linked to the root bye-bye pair emerge can't be the same as what we call small v or small v star high up in the derivation and in fact i think that things like v and v star have been somewhat misinterpreted they're not really the it's not really a verbal categorize err it's basically a phase marker and I think we should think of both in what we've been calling or what I'm now calling small in and small we used to call small V as being basically the two kinds of phase markers the and I think kind of natural way of capturing this is to go back to a classical distinction actually goes back to classical Greece which has been used sometimes in generative grammar and to assume that we really have two fundamental notions classical terms they're the notions which gives you four categories plus s minus P which is minus s plus P which is verbal and in classical grammars that's all there is so if you look at the classical Greek grammar errors up till late into the Latin the Roman period those are the only categories there's no adjectives adjectives we can add but we'll consider it a non primitive category it would be both substantive and predicated and then there's the course the fourth one which is neither and that's the rest of the junk that's lying around preposition as and so on and so forth but the two major categories the ones that really count are probably the ones that are the perfect substantive and the perfect predicate the ones from essentially classical grammar and there is some and these are distinct from the categorize errs N&V those are totally different notions on these things then would mark the phases and it's kind of striking that and they would be the links all of these would be the things that coordination conjunction disjunction link to the notice that there's a well-known phenomenon about extraction that holds of both nominal phrases and verbal phrases there are some nominal and verbal phrases which resist extraction you can only extract to the edge you can't go out of the element so we'll call those strong in the case of verb phrase it's the transitive verb phrases the ones that usually marked these store those constitute strong phases that you can't move out of you've got a first go to the edge then you can go on the weak phases say annika sitive passives you can extract all the way and there's the same distinction in nominal phrases so the complex noun phrase constraint it's been known for a long time it works strongly for definites but not for indefinite it's nonspecific and specific so the extraction out of a definite noun phrases violates the rules extraction out of a indefinite nonspecific it's a very straightforward so it looks like both the nominal and verbal phases have the category break into the categories say strong and weak and what we're calling V and little the two types of flavors of a V and V star probably just strong and weak phase markers perhaps framed in this in this system the there is another that there's been a lot of discussion over the years about whether noun phrases are really phases like verb phrases and the one problem about it was that they have a lot of similarities but there's one notable difference and that is you don't have an escape hatch in noun phrases there's no nothing that allows you to go to the edge and then go on but fortunately as I learned a couple of weeks ago there's a nice African language called bully which has which does have an escape hatch has a marker that says okay if this is here we can add on to the noun phrase and then go on so if we simply assume that the basis for all languages is Semitic proto Semitic and boolean then everything works out so we're off and running just have to find the right languages other languages like say English are defective in this respect they don't have that extra morphological element that's one of the nice things about doing comparative work you can fill in the blanks that oughta be there well if you play with this you'll notice that it can get pretty complicated so for example it could be that one of the con junks say s0 yes you know the third element in the sequence it could if this is a conjunction the third one could be a disjunction okay you can stick a disjunction inside a conjunction and that disjunction could be an unbounded on structure element and if you play with it gets very complex I'll put that aside the the principles for dealing it are straightforward but the actual working it out becomes pretty hairy but it's not a fundamental problem just a technical problem now there is a fundamental problem which was made very clear by recent work of Barry shines I don't know if he's around here somewhere but Barry has a very terrific book a great book that every cement assist or a study carefully which I think is the longest book with the shortest title and the entire literature's the book called and 600 pages of details about the extreme intriguing complexities that you get in the somatic interpretation of coordinate structures it's all done in a kind of an event calculus Neil David's own in event calculus so here's the task the task is to take assuming that this is the right kind of syntactic formalism you might play around with it a little assume that if this is correct how can you go to this from this to those somatic objects okay I'll come back later if there's a little time to the nature of these semantic objects but I think he gives good evidence that that's the kind of target that the mapping from syntax to formal semantics is trying to achieve so here's a huge unsolved problem for those of you are looking for dissertation topics one of many that nice fortunately there's lots of such topics hanging around I should say incidentally that this is a dramatic difference from say the 1950s and even the 1960s at that time it looked like there was almost no topics it looked like this whole field was finished I think I mentioned this earlier and one of the really striking things that's happened during my lifetime is that the field went from one that was terminated to 1 that's entirely endless and open everywhere you look there's just more problems that's a good sign that at least something's on the right track because that's the way things ought to be people who are just coming into the field now probably aren't struck by the dramatic difference but if you look back it really is dramatic well the another case that's an intriguing one is see how to bring this in well that's probably too complicated to go into it I'll skip this well there are lots of other there are plenty of other candidates for para merge some of them are interesting unsolved problems so for example if you take the paradigm things like John Mary saw the man walking down the street saying or Mary saw the man walk down the street you run through that and familiar paradigm you notice that there's a gap a strange gap in it with the bare verbs like Mary saw the man walk down the street you can't passive eyes so you can say the man was seen walking down the street but not the man was seen walk down the street this holds for two categories of verbs perception verbs and quasi causative verbs like make or let you know so I made the guy walk down the street but not the guy was made walking down the street those two categories this has been a kind of a funny sort of gap for a long time no good explanation for it there is a proposal and interesting proposal by Norman Richards in terms of his continuity theory that it has to do with the output of passive ization but that doesn't work because the same problem arises with in situ passives so kind of a little awkward in English normal in most languages but not bad in English if you saw say things like there were seen walking down the street three men let's say that's more or less okay but if you try there were seen walk down the street three men that's hopeless okay so there's a sharp distinction between passives and non passives whether or not you move the object okay so that can't be what's involved so what could be involved well what could be involved this kind of suggested by the fact that this holds for the quasi positives let and make one might imagine may be that these things really are just spelling english-style spilling out of causative morphemes which attach to a to a verb which are all over the place English tends to not use affixes to spell things out us words but these things could be affixes so it could be that let and make are just the causative affix is attached to the verb bear verb then comes an interesting question why should perception verbs have the same property but putting that on the Shelf if they all have this property what you might expect is that all of these are hair emerged to the bare verb and the pair emerged element is just immune to the passive operation say whatever you think passive is maybe dropping the case case let's say so that's that would fill in the gap making use of the pair emerge device there are other interesting proposals one of them's he said key Tijeras suggestion about what has always been a pretty serious problem the head raising head raising has none of the right properties it violates the extension condition it it's always described incorrectly if a verb raises to inflection say to tea it's always described as if it's a tea but it's not it's a V the things somehow that's a joined is really verbal not inflectional and the fur will further move of V to C is really kind of visa the second phenomena not a tea second phenomena so there's all kind of problems another kind of problem with it is it's sort of head raising has properties that are kind of shared the cross syntax and phonology so it's almost entirely like phonological processes in that it doesn't have semantic consequences so with the arrays say take Pollock's analysis of french and english which many of you know if you raise in french you tend to raise the verb and english not to raise it but the semantics doesn't change if you have a v second language it's has the same semantic properties as a non v second language and in fact you have to look very hard way out of the fringes to find some kind of semantic consequences to hedge raising that makes it look phonological but the trouble is its successive cyclic which makes it look syntactic so it's kind of problematic all alone but the real question is how do you get around this how can you actually create it without violating all the rules well he says proposal is this essentially we'll just take the VMP t2c case the same for the others you generate you generate see then that's just ordinary sign and generate see is in the workspace because that's where the lexicon is so you just look at it it's there in the workspace then you merge these two para merge them see we now have a set the workspace now contains this this notice by merging C and T you haven't increased the number of accessible elements because neither of these is accessible okay that was as we've already pointed out they're just like old man so and once you've done this you now simply merge this thing and if you think about it you can do all the head raising that way I thought for a while that there were some problems about this he said I responded about it but I think the problems are easily overcome so that's a possible way of approaching the entire problem of head more of head raising which has been a tricky and serious problem so leave it in your your hands to play around with that and see if you can handle the rest of head raising that way but a simple way to proceed which doesn't violate any of the rules doesn't add accessible elements just follows all that's just available you know and then anything new for it and it's crucially has to be para merge for one thing because that's the way these things end up they end up as adjuncts not as sets and secondly it doesn't add accessible items so it doesn't violate the resource constraint so it has the right properties in that respect well the there's plenty of questions like this there's there's a lot of the a major task that is faced is to take all of the cases that took two tasks an our first one take the cases that have been described usefully by illegitimate operations to take all the cases that have been handled by say late merge which is very widely used or by parallel merge which gives you all the multidimensionality cases and so on take that whole class of cases and show how you can handle them by legitimate operations I gave a couple of examples last time like a TB parasitic gaps but do it for the whole slew of them and of course the broader task is take a look at that massive data out there and see if you can make some sense out of it that's we we now have a criterion for what make sense means make sense means get a genuine explanation where a genuine explanation means one that's that that can be that meets the conditions the crucial conditions of learnability and evolvability and if you can reduce it to merge you've solved that problem because there's no question of learnability and the problem with evolvability as I mentioned it's basically solved you have to look at the details of how it worked but we know it happened okay so if you can get to that you have a genuine explanation if you can reduce things to pair merge you've come pretty close you still have to ask the question of how pair merge could have evolved but at least that's an easier question than lots of other things you can imagine we would then like to try to show that the other devices that seem critical for explaining things like say phase theory can be reduced - third factor properties of minimal computation they do in fact reduce computation so if you could show that that's another third factor condition like resource or striction you'll have a genuine explanation now that's the general project but notice it and make its again should be clearer going back to my first lecture that this project only arises if you accept a particular version of what the whole enterprise of linguistics is if you pick a particular kind of answer the initial question what's what is language what are we doing the answer is the the enterprise is one that attempts to deal with to address that what I called earlier the Galilean challenge the challenge that was posed at the beginning of the Scientific Revolution which translates in our terms into capturing the basic property of language the property of that each of the Faculty of language permits the generation of an infinite number of structured expressions with somatic interpretations and the option of ancillary operation of mapping at the one or another sensory motor condition if you accept that project and then a particular variant of that project which not all people who do generative grammar accept the second variant is to assume that this is part of natural science ok in other words follow what's called the bio linguistic program take this property to be a property of human beings not some object out in the external world which somehow magically human beings connect with but just an actual property of human beings then you meet you have to meet the conditions of evolve ability and learn ability then we have the notion of genuine excellent nation then all this mass of questions arises often with solutions often not notice again that not excepting this enterprise is perfectly legitimate it's a way of getting describing things in ways which might be very useful it's also useful for engineering projects kind of Silicon Valley linguistics where you don't really try to explain it just get something that works lots of interesting things you can do but this is a very specific approach to language saying it's part of natural science it's trying to meet the galilean challenge if that's the enterprise then these are the kind of questions that arise well there's another topic that I oughta at least mention there's not much time for it which is just left out entirely a computational system has rules and atoms things that are the smallest elements of the computation I mentioned earlier that if you have if you have merge and a one element lexicon you know just one thing in the lexicon one atom then in fact internal merge gives you the successor function and I showed some reasons to think that that's a natural outcome but in any event they've gotta be atoms and I haven't said anything about those so what are the atoms of computation and right at this point we're moving to a domain that I really didn't talk about namely the general domain of cymatics so what how does the domain of cymatics fit into this enterprise well here you have to be a little careful I have to ask what you mean by semantics there's a classical notion she goes way back in history but in the modern period it's a frigging charles sanders peirce tarski Quine you know that tradition this concept of semantics has basically to two essential notions the one is one is the notion truth and reference which are essentially the same thing and the other is the notion entailment with many variants so let's take a look at those two notions first of all the notion entailment is not semantics its syntax okay it's called logical syntax entailment as a matter of the formal relations among expressions the world doesn't even have to exist okay semantics is the field that tries to relate the internal mental computations to the world the actual real world out there that semantics that's truth and reference but entailment is logical syntax and if you look at what's called formal semantics which are some of the richest and most exciting work going on in the field in the last couple of decades notice it's all syntax it's all pure syntax and it's an interesting kind of syntax it's sort of analogous to phonology if you think about the externalization process what do we have the syntax generates a set of objects we then want to externalize it and that process of externalization has two steps the first it's called you know traditionally more phonemic sore phonology in the sense that say Mars and Allie Mars Hallie and I use the term but it's the mapping of the syntactic structures to some sort of phonetic form the phonetic form as a syntactic object of course and the mapping is a it's a symbolic manipulation so it's again syntax in the general sense it's a particular part of syntax I should say that this task of mapping with generative phonology mapping syntactic structures the phonetic form has pretty much been abandoned in modern phonology contemporary phonology is mostly optimality theory and that simply doesn't look at this question okay its deals with other questions but this question of generative phonology which was in fact the first thing that was discussed back in the late forties early fifties has been pretty much forgotten but it's it's there you know it's a problem that has to be solved if you are concerned with externalization with how language you know gets out there to be sounds or signs the so one-to-one task which is really part of syntax is generative phonology mapping syntactic objects to say narrow phonetics then there's another task and that is taking the narrow phonetics and telling us how it relates to motions of the articulator Zoar to sound waves or if you haven't be doing sign now your gestures end up you know actually being formed you know how do you move your fingers and so on that's phonetics acoustic and articulatory phonetics if you think go back to semantics formal semantics is like phonology it's syntactic operations setting the stage you hope for eventual relation to the outside world just as phonology generative phonology is syntactic operations setting the stage for the connotation to fill in relates to moving your articulate errs around okay which gets into the outside world something outside language well this then comes the question notice that of all of the domains of the study of language it's almost all syntax it's what we call kind of narrow syntax just forming the syntactic objects what I've been talking about the last couple of days another a part of syntax is generous phonology another part of syntax is formal semantics all of this stuff is going on inside the head right doesn't get to the outside world it's neither phonetics and/or semantics the of all of these areas the area of formal semantics is different from the other two in the way it's been pursued which raises interesting questions it is the only domain that has not been subjected to the effort to find genuine explanations so there's no effort within formal semantics to ask what's the simplest way to do it the problem is let's just find a way to do it you know if you can get a way to do it by sticking in lots of lambdas and so on and so forth okay we'll do it that way but the question of what's the best way to do it the way that really gives explanations that really hasn't been raised now that's not a criticism it's a very hard question and so if it's not been raised well hard questions often tend not to be raised but one should remember that it's a question to be dealt with someday as formal semantics develops this branch of syntax develops it should go the way of the generative syntax and generative phonology in the case of generative phonology is a lot of work on how to get the optimal way of doing it in fact that but things like sound pattern of English are about and many other plenty of other work and narrow syntax kind of stuff we've been talking about formal semantics that part of syntax the question really hasn't arisen sooner or later it should arise another interesting aspect of formal semantics is it's it's kind of tacitly assumed to be invariant so if you there's no proposals that languages have parametric differences and how you interpret say quantifiers and variables there's a way of doing it and if you do it that's universal which is a very reasonable assumption because if you think about the things that are being investigated there's absolutely no empirical evidence for them a child has no evidence about any of them so if you look at the literature you look at the things that are people are studying they're totally without evidence so either they're just not accurate or if they're accurate they're gonna be universal so it does make sense to say that this part of syntax is invariant I've talked a little about whether narrow syntax is invariant that's an open question it doesn't look like it on the surface but a lot of the work that's been done is kind of restricting the domain of variability of narrow syntax to the point where you might imagine that the core of narrower syntax what's dealing with the galilean challenge the yields of a struct the operations that are yielding the semantically interpreted objects maybe that doesn't vary that we don't know it's an open question could if it doesn't vary at all then the variability of language is entirely an externalization we're certainly varies all over the map no not any question there I mean not you know it's not free there's very interesting restrictions but a lot of variation possible so that's the kind of picture that seems to be emerging I think and it leaves open this thing here where does that fit in how do we relate this stuff that's going on inside the head to the outside world turns out not to be such a simple question the in fact if you look at the proposals and formal semantics we they sound as if they're relating to the outside world but when you look at them they're actually not so it takes a model theoretic approaches the elements of the model are sort of one treats them as if they're actually out there but in fact if you look at the elements of the model they're completely mental objects there's nothing in the outside world that is corresponds to those postulated objects of the model so the model theoretic semantics works fine you know you can do all kind of things with it study necessity you know so on and so forth but it's all pure syntax it's not getting to the outside world what about event semantics which is a very rich and productive field the kind of neo David Stone in event semantics well what are events I mean our events things in the world does the world come packaged in events actually it doesn't you know events are our mental constructions imposed on whatever is going on so events are really internal and we know that to try to count the number of events that's going on it's kind of meaningless so for example thanks to Zeno we know that if you walk across the room there's a continuous the number of events is the power of the continuum okay and same with anything else you're looking at so there are about as many events as you decide to impose so event semantics which is very productive and rich is again another form of syntax so how do we get beyond syntax well we have to look at the notion of reference truth will depend on reference and if we want to look at the notion of reference we start naturally by looking at the words that purport to refer so you look at the literature on reference mostly philosophical literature say London is taken to be a prototypical a referential word a lot of problems arise say Kripke's polls of belief when you try to ask questions about London but is London a thing in the outside world well you can ask yourself so for example I can tell you that I visited London before it was burned down and rebuilt 20 miles up the Thames so is there an object in the real world which is physical because it can be burned down but then can be rebuilt somewhere else with different physical objects and looking differently is there such an object in the physical world and the outside in the material world certainly not I could we could decide to rebuild Carthage let's say what's gone a couple thousand years ago we could rebuild it somewhere else look different but it could be carthage so whatever these things are they're not names of things in the outside world what about other words are there in fact we might rest the question whether there are any words and language at all that refer to anything in the outside world I think the answer is there aren't some of the reasons for this were given as far back as classical Greece so for example Aristotle asked the question what is a house and his answer was a house consists of the amalgam of two different kinds of elements in his metaphysics they're matter and form so the matter of the house is the timbers the bricks you know the stuff it was made of the form of the house is the design that what it's for that what the architect had in mind how it's used and so on so something might look physically just like a house but in fact be something totally different it could be a library let's say it could be garage you know could be a paperweight for a giant could be all kinds of things it depends on what the architect had in mind and how it's used what it's for basically but the properties of form are not in the physical world they're in the mental world know when our Aristotle talked about this this is from his point of view metaphysics in fact it's in his book metaphysics but when you go to some 17th century again there was a kind of a cognitive revolution and a lot of these ideas were reinterpreted in terms of modes of cognition which i think is the right way to look at them so reinterpreting it and that those terms a house is something that we constructed our minds which has a material element but of course the crucial part of it is what Aristotle called the form that's something that's part of our mental operations so when we use the word house we're not referring to an object and we are and we're carrying out the act of referring like I'm referring to that object over there but the word house is not referring okay you think about every word in the language it's exactly the same there's a example even earlier than Aristotle the pre-socratic example used by a Heraclitus who argued that you can't cross the same river twice because when you cross it the second time it's completely different physical object and you're a different physical object so it's impossible to cross the same river twice of course we do cross the same river twice which simply tells us that our notion river does not refer to anything in the material world it refers to a construction that we developed and if you start playing around with how do you individuate these objects becomes pretty tricky so for example take River again say the the Charles River which I used to cross on the way to work up and back every day in boss mode suppose you take the Charles River and you reverse the course of the water goes the opposite direction okay you've still the Charles River I suppose you diverted into a different direction because you don't want it to go into the bay but somewhere else still the Charles River in fact you can play around with it and make all sorts of massive changes in the physical object and will still be the Charles River on the other hand there are trivial changes that will prevent it from being a river at all so suppose you put barriers along the side and you start using it for say commercial tankers going up and down now it's between two points then it's a canal it's not a river anymore suppose you even make a more miniscule change you make what and physics is called a phase change from the liquid to the glassy state tiny change almost undetectable instruments but now it has a hard surface and you paint a line down the middle and you start using it to commute to Boston okay now it's a highway you've made an almost undetectable change if you play around with it some more you can see that you can make huge changes in the physical object stays the Charles River you can make virtually undetectable changes it's not a river at all take a look at any other word in the language you finds pretty much the same so it may be that language simply does not have the concept of truth and reference at all which means it doesn't have semantics there is no semantics there's syntax and there's modes of use that using the objects you've constructed roughly what's called pragmatics it could turn out that the whole study of language reduces to syntax and pragmatics with syntax having these many different variants in fact my guess is that's probably what we're gonna find when we when we pursue the topic further that raises many very interesting questions so for example take these terms like house or River London whatever you like where do they come from if we look at animal symbolic systems a very striking feature of them is that the atoms of these systems the elements within them do in fact pick out identifiable physical events okay so you take the take the B that's has my name attached to it if it's a honeybee and does the right source of things then it's you know it does the famous waggle dance it flies out to a flower comes back to the waggles and the things that it's doing are one-to-one identified with particular physical phenomena to the B the distance to the flower the height of the flower the orientation in which you have to fly and quality the flower and so on every animal system we know is like that monkey calls for example if there's something we call a warning call what it means is the leaves are fluttering in a certain way that monkey reflexively emits some noise other monkeys run away maybe an Eagles coming something like that animal systems seem to be like this completely the human systems are not like this at all that raises a very interesting question another one of those mysteries where did this come from and if you look at the array of mysterious things there are plenty of them and they fall into a couple of categories two categories one category is the category of mysteries where we just can't get the right kind of evidence because it's just empirically impossible for us to do it we can't go back say 200 thousand years and hear what people were saying and in principle yo its it's an empirically possible it's a theoretically possible test but you just can't do it that's one kind of mystery but then there's the deeper kinds of mysteries which have to do with things like selecting the sequence given by the Hilbert epsilon operator when you're trying to form a coordination or more generally just picking what sentence you're going to produce and these kinds of mysteries are the kind where you really don't even have bad ideas there's no idea as to how to proceed those we could ask what that category of mysteries is like could mean that some restriction on human cognitive capacities a lot of debate about this most scientists vigorously reject the idea that there could be a limit on cognitive capacities but if you think about it it's just strange position if humans are similar or part of the organic world not say angels we ought to have the same properties as other kinds of organisms some respects every organism we know has scope of cognitive capacity and limits of cognitive capacity and in fact the two are kind of logically related whatever is yielding the scope is also imposing certain limits and you might ask why humans should be exempt from this so I hope the former phonologists and Samantha cysts in the audience who are now sin tactician z-- will have questions may the questions begin yes here in the front down here yep the mic is coming well first of all thank you so much and I would like to ask you considering all this fascinating theory how could you define madness for us define madness in Portuguese locota given what you have said so far could you define madness and and I could go I would like to go a little further and ask you what could you say about these writers I'm thinking people like Samuel Beckett or Antonin Artaud Antonin Artaud and the French one whose texts are always dealing us this being seen thanks with disturbing syntax yeah okay literature would you maybe also like to comment on Samuel Beckett or the second author was otto Anton uh octo whose writings are about disturbing syntax yes no not about but whose writing are always really disturbing syntax oh okay thank you know a good bit of creative literature makes use of the properties of language but purposely distorts them to impose tasks for the reader so if you read say serious literary criticisms a book like William empson's seven types of ambiguity which is a kind of a famous work of the 1950s Empson argues that poetry in particular he said is designed to have actually completely minimal use of language to impose on the reader the task of creating interpretations and using the readers imagined of resources to reconstruct what's not presented in the text so poetry would be different from say a literal description which tells you everything that's happening because it's calling on the reader to create and invent theirs if you go back further of the classical aesthetic theory like Slagel for example Coleridge borrowed from we're familiar with this mostly from Coleridge but it comes from Schlegel he argued that he had an interesting argument that he argued that poetry was the most exalted of the arts and he pointed out that when you talk about the poetic quality and music you're talking about its imaginative character when you talk about the musical quality of poetry you're talking about something less significant you know the way it sounds or something and then he asked the question why should poetry have this special status and he argued that every art form has some medium so sculpture deals with marble painting deals with colors you know each one has its medium poetry also has a medium language but language differs from the others and it's inherently creative it has an inherently creative aspect that's the galilean challenge again recognized it's got this capacity of creativity internally to it so therefore the art form that makes use of this modality will be uniquely creative and in fact will be used in such a way as to draw from the reader the the imaginative resources that are inherent and the in the medium itself now when you get to say Samuel Beckett or literary artists in general they're playing with this all the time for trying to compel the reader to think and create and imagine I wouldn't call it madness Plato did Plato argued that there was recognized there's various levels of creativity there's the kind of creativity that exists and say normal language use when you're kind of like constantly creating but there's a higher form of creativity that has this true aesthetic value and a Plato argued that that's as a form of madness so okay question over here thank you I would like to know what your perspective is about natural semantics metalanguage the approach that provides sixty-four words that are supposed to be semantic Prime's and that provides a semantic foundation for universal grammar oh that my proposed a foundation for that do you have a comment on natural semantics which provides a meta language of 64 developed by an Africa and Qatar in Australia I don't but I can repeat it again it was by Godard and another pika another Sica Australian well I think it's interesting work she's trying to find fundamental elements that appear and somatic systems generally then it's I'm not really very conversant with the work but I've looked at it and I think it's in you know it's a promising virtual direction she has some interesting insights but it's an attempt to find as I understand that the basic elements that can't that are constitutive of the knight of the systems of natural semantics in the world which is a reasonable project over here Gabe in the red shirt hi um so it seems like one of the main philosophical motivations for positing an externalist notion of reference is that you need that in order to get some notion of objective truth so when I make claims about London say I'm not just expressing some perspective I have on the world I'm making a claim which can be independently evaluated as true or false do you think that this sort of internalist notion of reference that you have is there a way of reconstructing objective truth or are we kind of condemned to have a merely subjective I'm it's true if I'm talking about London if I say I take the sentence I I gave as an example suppose I say again something like this I visited London last year before it was burned down by a fire and was rebuilt 20 miles up the Thames that's a statement that can be evaluated but the question is is there a physical entity identifiable in the outside world that I'm picking out and I don't think there is I'm talking about the world undoubtedly there is what Aristotle called the material element so you know it's 20 miles up not 50 miles up that it was this place that people lived in now it's some totally different place it's still London meaning what the criteria of individuation and identification are in fact mental conditions but notice that that shouldn't surprise us too much because that's true of cognition in general we don't have what's called immediate experience is already constructed by the mind you know there is sensory stimulation hitting your retina but that's not what you they were two experienced depends on what were called in the 17th century modes of cognition ways of interpreting the data that's immediate experience so the most immediate experience we have the most we're most confident about and certain about is already constructed whatever is objective is something we postulate we hope that there is a real world out there which is somehow giving us the data that we experience but what that real world is for that we go to the physicists and they have some very strange answers so for example you go to the Institute for Advanced Study and talk to the late John Wheeler great physicist for him the world consisted of answers to questions that we posed that's what he calls the it from bit there's no world at all we just ask questions we get answers that's the world ok that's what he says but whatever it is it's something we postulate it's not something we experience so in fact if the words we used to refer also have this property it shouldn't really surprise us that much that we are different from other organisms in that respect one of the many respects from Gabe Gabe another game they say that a picture is worth a thousand words let me stand up but we might say that a picture entails a thousand sentences and your picture of entailment is as a kind of syntactic relationship but in this kind of case we have a relationship between domains that very different syntaxes and so i'm curious about what is your general view of entailment coreference other kinds of apparent semantic relations that obtained between representational modalities I don't know that I can do justice yes let me give a one-sentence version sorry what is until if entailment is syntactic how do we entail from vision to language if entailment is syntactic how do we entail from vision to language well that's not until meant that's inference which is something quite different so entailment is a formal relation among syntactic objects you know you take a look at quantification theory tells you how this syntactic object relates to that syntactic object inference is an act that we carry out very mysterious act we have we face problems there's some thing out there that's puzzling you know why does across-the-board deletion work let's say and then you invent you somehow come construct a theoretical idea which you then try to ask does this account for the basic properties of the data nobody knows how that works if you take Corson's Jerseyan studying physics at UCLA I'm pretty sure there isn't a course on methodology there might be in the sociology department but it's very unlikely to find one and the physics or the chemistry department of course there is no method it's just something that you manage to do somehow some people do it better than others but nobody knows how it's done this actually relates to a concept that I just was using as if it's straightforward and neutral but it isn't that's descriptive adequacy so one of the things we want to do in studying linguistics is to make sure that our explanations meet the condition of script of adequacy at least we get the facts right but that's a very complex notion what's descriptive adequacy and it's not just all the data you have around for one thing the data that are out there are not pure data they're not data that specifically come from the particular source that you're trying to investigate that's why scientists do experiments instead of just you know cataloging things that go on the world the reason every experiment is a high level abstraction each experiment is a very complex abstraction theory-driven which says I'm gonna throw away lots of stuff that's going on and look at this particular thing because I think somehow this particular thing bears on an explanation of what might be going on in the world that's that's the most elementary part of science it's already trying to abstract away from all the noise out there and seeing what might be useful for explaining something and those things are the properties of inference but you know the way you learn how to do that in the sciences is just by watching other people do it you say here's the way it was done I'll see if I can do that myself but there's nothing you can you can't teach it you know you can't teach how to carry out successful inference you learn it the way kind of the way a carpenter learns his craft you know by watching skilled artisans do it and you sort of pick it up but I think that's about all the one can say about inference time of course there's you know philosophical discussion of it but you have to really cautious about the discussion so for example one of the ideas that people scientists commonly uses the notion of poppers notion of falsifiability so if you have a theory that can be falsified real and if you get a negative result you throw the theory out every scientist knows you never do that if you get data that falsifies your theory what you usually do is say well I'll put it on the shelf maybe somebody will figure out what's wrong with it but if the theory looks good I don't care if it's falsified you know and if you look at the history of science that happens all the time on crucial cases you know so they're kind of rules of thumb but they don't really work that's not the way scientific enquiries done or rational inquiry in any area you know not just Natural Sciences that the sciences of the best place to look because it's most refined you know they're you really have deep results and so on you learn how things are done but it's part of just ordinary life you had asked earlier I just had a clarification question probably I misunderstood so it is for dealing how we dealt with head raising over there up on the board so I just wanted to clarify that just want a question about head raising oh you've got over here on the board yeah so my question was the parent head raising is it the set C T and T P T V P or is it C and the said T V P once can you repeat that please is it the pairs in the head raised in case T - C case is it the set C T and T V P or is it C under satty VP I didn't hear you it precisely but the question was that when you were discussing head raising and you had the set T VP TVP yeah which then further raises to C so the new set formed SCT or is the set form to C along with T V V is the new set that is formed see along with TVP no just T actually if you want to get the right answer the source of this is sitting right here so if I'm not giving the answer correctly you fix it but I think you have TVP yeah you've got see inside the lexicon so you take C and T form the pair C key that's just some new thing out there you haven't had an accessibility cause both of the elements are inaccessible now that you have C T sitting out there and you have the old TVP you can just merge them that gives you the the thing you're trying to so how does it how is it exactly different from external merge them how exactly is it different from external merge from peril well that's an interesting question like why can't you that's one of the problems that he said and I have been corresponding about for a while so you could imagine that after you've constructed CT TVP you start off somewhere else and you construct some other T V P and then you take this C T and you stick it on to that that would be an illegitimate operation but the question and I thought for a while that that was a serious objection to this but on reflection I don't really think it is because what you're getting that way is something you can always produce and it's illegitimate because the you're just getting a case where the T that you have up there is trying to be a copy of something down below but it isn't it's just a repetition of it so it's it's just something that's incoherent and you could always do that you could not not quite apart from this operation after you've constructed TVP you could make up some other things XP and just merge that on and it would be nonsense so you can always construct nonsense and I think this would just end up being a particulars ends notice that there's a subtle difference between the illegitimate we do want the system to generate non grammatical constructions that but we don't want it to generate things that violate linguistic principles that's quite different there's quite a distinct thing like it takes a negative polarity items things like say ever you can't say John was ever here but you can say I doubt that John was ever here but notice that in order to generate I doubt that John was ever here you first have to generate thought John was ever here so we certainly want to allow the system to generate ungrammatical sentences otherwise we can't get dramatical sentences and in fact we're perfectly happy to have the system generate things like metaphors that deviate from normal argument structure phrases like misery loves company for example which violate the selectional restrictions of the verb we want lots of things like that but we don't want the system to generate things like the books is on the table or a violation of island constraints that's a crucial difference not a very in many ways not a very clear one you have to think about it but that distinction just has to be there and this kind of case that you're suggesting would be one in which you're essentially creating something nonsense nonsense achill but not violating linguistic principles as distinct from parallel merge does it's a subtle but interesting distinction Oumar over here well okay thanks I really enjoyed these lectures and I just wanted to ask you what role does language change what role does the study of language study change to play in increasing our understanding of language in general quite alone I mean languages don't change in arbitrary ways they change in very specific ways and by looking at the kinds of ways in which language change we get evidence about what the system is like what are the dimensions in which language can change and which can't change and what what kinds of phenomena could cause that change notice this is very similar to the study of language acquisition we're asking how bits of data that come around lead to modifications of the system that's in the mind the language acquisition does it in one way language change does it another way so it's a probe into the kinds into the structure of the system that you're investigating what kind of things can happen what can't happen what makes them happen what doesn't have an effect and so on whether it's language acquisition or language change or you know dialect things like when you when you're mixing pieces of one one dialect into another or forgot the name for that also tells you things it can happen in certain ways not other ways lots of ways of studying the system you have a complex system you perturb it in all sorts of different ways you learn things over here hi my question is about you saying that truth and reference are essentially the same thing because while we were listening to the lecture and doing the Q&A everyone's taking notes in these really different ways reference I'm actually taking notes in three different ways I have an AI AB transcribing everything using an NLP then I have this kind of like josef novák concept map of my own stuff and I'm also texting my friends like oh my god Noam Chomsky said truth and references basically the same thing so why are they not exactly the same thing why did you say they are essentially the same thing and yeah that I mean that sentence was like she was intrigued by your statement that truth and reference are essentially the same thing and she asks she says she's familiar with three very different kinds of reference but she wonders why you didn't say they're exactly the same thing maybe you want to elaborate on that issue well you know a theory of truth says typically I have a name and a predicate and the name refers to an object out there and the predicate either does or doesn't apply to that object and if it does apply the statement subject predicate is true so if I say you know the Sun is in the sky the idea is the Sun refers to a particular object that's reference in the sky is a certain property my statement the Sun is in the sky is true just in case that property holds of the object for language that that doesn't really work for the sciences it it does work but it does work by stipulation when you're tests which are doing is constructing inventing concepts which do you hope have the property of reference so when a linguist develops the concept phoneme let's say they're saying I hope that there's something in the external world that this thing that this concept refers to if a physicist defines electron say they're saying I hope there's a thing out there that this one the one refers to okay so scientists are essentially trying to construct systems that do meet the referential criterion and therefore will be producing true or false sentences which can be evaluated by you know by experiment or observation but ordinary language just doesn't work like that at all the concepts that we have involve complicated interactions between what Aristotle called form and matter you know just fact about natural language in contrast animal communication systems seem pretty close to that of course they don't make true statements i'm indy you know the bee who's waggling is not trying to make a true statement he's just doing whatever the internal structure of the bee is telling him to do [Music] so can can we go back and maybe get a little bit of expansion on what you went through yesterday on a DB DB why don't we let that go till tomorrow because it's a long story