Another day, another paper. This time a rumination on Chomsky’s Syntactic Structures arguments about the autonomy of syntax. I think, despite Fritz Newmeyer’s excellent attempts to clear this issue up over many years, it’s still reflexively misunderstood by many people outside of generative grammar. Chomsky’s claim that syntax is autonomous is really just a claim that there is syntax. Not that there’s not semantics intimately connected to that syntax. Not that syntactic structures aren’t susceptible to frequency or processing effects in use. Just that syntax exists.
Current alternatives to the generative approach to dealing with language still, as far as I can tell, attempt to argue that syntactic phenomena can be reduced to some kind of stochastic effect, or to some kind of extra-linguistic cognitive semantic structures, or to both. This paper attempts to look at the kinds of arguments that Chomsky gave back in the 1950s and to examine whether the last 60 years have given us any evidence that the far more powerful stochastic and/or cognitive semantic systems now available can do the job, and eliminate syntax. I guess most people that know me will be unsurprised by my conclusion: even the jazziest up-to-the-minute neural net processors that Google uses still don’t come close to doing what a 3 year old child does, and even appealing to rich cognitive structures of the sort that there is good evidence for from cognitive psychology misses a trick when trying to explain even the simplest syntactic facts. I look at recent work by Tal Linzen and colleagues that shows that neural net learners may mimic some aspects of syntactic hierarchy, but fail to capture the syntactic dependencies that are sensitive to such structure. I then reprise and extend an argument that Peter Svenonius and I gave a few years back about bound variable pronouns.
One area where I do signal a disagreement with the Chomsky of 60 years ago is in the semantics of grammatical categories. Chomsky argued that these lack semantics, but, since my PhD thesis back in the early 1990s, I’ve been arguing that grammatical categories have interpretations. Here I try to show that the order of Merge of these categories is a side effect not of their interpretations, but of whether the kind of computational task they are put to is more easily handled with one order or the other.
The idea goes like this (excerpted from section 4 of the paper).
“Take an example like the following:
(20) a. Those three green balls
b. *Those green three balls
As is well known, the order of the demonstrative, numeral and descriptive adjective in a noun phrase follow quite specific typological patterns arguing for a hierarchy where the adjective occurs closest to the noun, the numeral occurs further away and the demonstrative is most distant (Greenberg 1963, Cinque 2005). Why should this be? It seems implausible for this phenomenon to appeal to a mereological semantic structure. I’d like to propose a different way of thinking about this that relies on the way that a purely autonomous syntax interfaces with the systems of thought. Imagine we have a bowl which has red and green ping pong balls in it. Assume a task (a non-linguistic task) which is to identify a particular group of three green balls. Two computations will allow success in this task:
(21) a. select all the green balls
b. take all subsets of three of the output of (a)
c. identify one such subset.
(22) a. take all subsets of three balls
b. for each subset, select only those that have green balls in them
c. identify one such subset
Both of these computations achieve the desired result. However, there is clearly a difference in the complexity of each. The second computation requires holding in memory a multidimensional array of all the subsets of three balls, and then computing which of these subsets involve only green balls.
The first simply separates out all the green balls, and then takes a much smaller partitioning of these into subsets involving three. So applying the semantic function of colour before that of counting is a less resource intensive computation. Of course, this kind of computation is not specific to colour—the same argument can be made for many of the kinds of properties of items that are encoded by intersective and subsective adjectives.
If such an approach can be generalized, then there is no need to fix the order of adjectival vs. numeral modifiers in the noun phrase as part of an autonomous system. It is the interface between a computational system that delivers a hierarchy, and the use to which that system is put in an independent computational task of identifying referents, plus a principle that favours systems that minimize computation, that leads to the final organization. The syntax reifies the simpler computation via a hierarchy of categories.
This means that one need not stipulate the order in UG, nor, in fact, derive the order from the input. The content and hierarchical sequence of the elements in the syntax is delivered by the interface between two distinct systems. This can take place over developmental timescales, and is, of course, likely to be reinforced by the linguistic input, though not determined by it.
Orders that are not isomorphic to the easiest computations are allowed by UG, but are pruned away during development because the system ossifies the simpler computation. Such an explanation relies on a generative system that provides the structure which the semantic systems fill with content.
The full ordering of the content of elements in a syntactic hierarchy presumably involves a multiplicity of sub ordering effects, some due to differences in what variable is being elaborated as in Ramchand and Svenonius’s proposal, others, if my sketch of an approach to the noun phrase is correct, due to an overall minimizing of the computation of the use of the structure in referring, describing, presenting etc. In this approach, the job of the core syntactic principles is to create structures which have an unbounded hierarchical depth and which are composed of discrete elements combined in particular ways. But the job of populating these structures with content is delegated to how they interface with other systems.”
The rest of the paper goes on to argue that even though the content of the categories that syntax works with may very well come from language external systems, how they are coopted by the linguistics system, and which content is so coopted, still means that there is strong autonomy of syntax.
The paper, which is to appear in a volume marking the 60th anniversary of the publication of syntactic structures is on Lingbuzz here.
I’ve been railing on for a while about this issue, but have just finished a brief paper which I’ve Lingbuzzed, so thought it deserved a blogette. My fundamental concern is about the relationship between restrictiveness and simplicity in syntactic theory. An easy means of restricting the yield of a generative system is to place extra conditions on its operation with the result that the system as a whole becomes more complex. Simplifying a system typically involves reducing or removing these extra conditions, potentially leading to a loss of restrictiveness.
Chomsky’s introduction of the operation Merge, and the unification of displacement and structure building operations that it accomplishes, was a marked step forward in terms of simplifying the structure building component of generative grammar. But the simplicity of the standard inductive definition of syntactic objects that incorporates Merge has opened up a vast range of novel derivational types. Recent years have seen for example, derivations that involve rollup head movement, head-movement to specifier followed by morphological merger (Matushansky), rollup phrasal movement (Koopman, Sportiche, Cinque, Svenonius and many others), undermerge (Pesetsky. Yuan), countercyclic tucking-in movements (Richards), countercyclic late Merge (Takahashi, Hulsey, and the MIT crowd in general), and, the topic of this brief paper, sidewards movement, or, equivalently, Parallel Merge (Nunes, Hornstein, Citko, Johnson).
An alternative to adding conditions to a generative system as a means of restricting its outputs is to build the architecture of the system in such a way that it allows only a restricted range of derivational types, that is, to aim for an architecture that embodies the constraints rather than representing them explicitly (cf. Pylyshyn’s Razor). This opens up the possibility of both restricting a system and simplifying it. In my Syntax of Substance book for example, I argued for a system that does not project functional categories as heads, following Brody’s Telescoped Trees idea. This immediately removes derivational types involving certain kinds of head movement from the computational system. Apparent head movement effects have to be, rather, a kind of direct morphologization of syntactic units in certain configurations. No heads means no rollup head movement, no head to specifier movement followed by morphological merger, no `undermerge’ and no parallel merge derivations for head movement (a la Bobaljik and Brown). That same system (Adger 2013) also rules out roll-up phrasal movements via an interaction between the structure building and labelling components of the grammar (essentially, roll-up configurations lead to structures with two complements). It follows that the kinds of roll-up remnant derivations argued for by Kayne and Cinque are ungenerable and the empirical effects they handle must be dealt with otherwise. In all of these cases the concern was to reduce the range of derivational types by constructing a system whose architecture simply does not allow them. Adger 2013 makes the argument that the system presented there is at least no more complex than standard Bare Phrase Structure architectures.
In the draft paper I just posted, I’ve tried to tackle the issue of Sidewards Movement/Parallel Merge derivations, by attributing a memory architecture to Merge. The basic idea, which I presented in my Baggett lectures last year, is to split the workspace into two, mimicking a kind of cache/register structure that we see in the architecture of many computers. One workspace contains the resources for the derivation (I call it the Resource Space) and the other is a smaller (indeed binary) space that is where Merge applies, which I call the Operating Space. So a syntactic derivation essentially involves reading and writing things to and from the Operating Space, where the actual combination takes place.
This architecture makes Parallel Merge derivations impossible, as there is just not enough space/memory in the Operating Space to have the three elements that are needed for such a derivation. This is really just a way of formally making good on Chomsky’s observation that Parallel Merge/Sideways Movement derivations are in some sense ternary.
In the paper I define the formal system that has this result, and argue that it makes sense of the fact that the two gaps in a parasitic gap construction do not behave interpretively identical, extending some old observations of Alan Munn’s. But the main point is really to try to reduce the range of derivational types, and hence the restrictiveness of the system, without explicitly constraining the computational operations themselves. The extra complexity, such as it is, is actually a means of simplifying or economising memory in the computational system.
The paper is here.
With all the interesting discussions about alien languages emerging from the excellent film Arrival, I thought I’d reblog this piece I wrote for E-Magazine in 2004 (can you reblog something which has never been blogged?). It was aimed at getting A-level students of English Language to think a little differently about syntax and phonology.
There you are, in the library, studying for your English Language A Level. You’ve done all the interesting bits: language and gender, sociolinguistics, discourse, and now it’s come to phonology and syntax. Why do I have to study this, you think, as your eyes begin to droop, and your head begins to nod and you begin to dream of a galaxy far, far away …
FROM: The Director, Gargoplex Institute, Alpha Centauri TO: Chief Exo-Scientist Jenl Itstre
VIA: Direct Telepathocrystal Network
RE: New Mission
New discovery made on planet 3, system 3.387. Species (designation Yu-Man), highly successful within biosphere, developed intelligence (level 2), culture (level 2.1) and technology (level 2.2), yet apparently no telepathic ability. Raises severe theoretical problems for doctrine 6.8 of Gargoplexic Code: Intelligence Culture Technology (ICT) count above level 2 requires communication; sophisticated communication is possible only via telepathy; therefore level 2 intelligence requires telepathy.
Council is worried about ethical questions relating to experimentation on non-telepathic beings, and about apparent support the existence of Yu-Mans gives for Anti-Vivisectionist Movement. Please investigate immediately.
FROM: C.E.S Jenl Itstre
TO: Director, GI, Alpha Centauri
RE: re: New Mission – Report 1
Have entered orbit. Confirm intelligence, culture and technology levels. Species seems to have highly developed communication abilities. No telepathy observed. Hypothesis: communication system is simple symbolic, associating one external sign with one thought (cf. Report ‘Octihydras of Glarg: 7, 203 tentacular positions for 7, 203 distinct messages’. Octihydras classified below level 2 on general ICT count.). Perhaps Yu-Mans simply have many external signs? Will determine nature of external physical signs relevant to species. No tentacles observed.
FROM: C.E.S Jenl Itstre
TO: Director, GI Alpha Centauri
VIA: DTN 6.8.9
RE: re: New Mission – Report 2
Exciting new discovery. Like many other species on planet 3, species Yu-Man’s physical manifestation of thought is not visual nor olfactory, as is usual for lower species across the galaxy (cf. Report ‘Lower Xenomorph Communication’, subsection 3.3.45). Unbelievably, it involves instead the manipulation of orifices for oxygen intake and food intake to create pressure waves in the air. Such systems have been hypothesised before, but it has always been assumed that creatures would find it too difficult to extract the relevant air pressure modulations from general noise and that such physical signs would thereby be impossible to use as the basis of communication systems (see Itstre and Itstre ‘On the impossibility of sound based communication’. Report GI).
RE: re: New Mission – Report 3
Troubling discovery. As is well known, simple symbolic systems are capable of only having a finite number of possible messages (Istre and Grofr ‘Finiteness and Communication’, GIM Monographs, AC1). Species Yu-Man have no such limitation – they can communicate an apparently infinite number of sophisticated thoughts between each other without telepathy, apparently just by physically changing the air pressure around them. First hypothesis must be rejected. This gets more perplexing by the hour.
RE: re: New Mission – Report 4
Psychological profile completed: Yu-Mans have complex thoughts in usual multi-dimensional, non-linear form. Communication profile: all members of this species (and none others in the biosphere as far as can be determined) have this curious ability to communicate an infinite number of thoughts without telepathy. Suggestion: Yu-Mans somehow map from their thought structures into physical structures directly so that air pressure modulations directly mirror the structure of thought. No, this can’t be correct. Air pressure structures cannot bear such a load of information. Am not thinking clearly. Obviously turning native.
RE: re: New Mission – Report 5
New evidence which is most perplexing: I have established that Yu-Mans can extract patterns from different air structures as hypothesized in previous communications. Some of these patterns clearly relate to thoughts, in a simple symbolic way. For example, a special pattern of sound waves (somewhat abstracted) links the symbol ‘planet-3’ to the right thought and so on. But there is also something we have never seen before. Connecting the symbols in different ways allows Yu-Mans to create complex patterns of symbols that mirror the structure of the relevant thought. It turns out that for some groups of Yu-Mans the order in which the symbols come matters. So even though the pattern ‘the sun orbits planet-3’ has exactly the same symbols as ‘planet-3 orbits the sun’, the fact the parts come in a different order means that it relates to a different thought. Even more peculiar, not all combinations of symbols are possible. For the same group ‘orbits planet-3 the sun’ has no corresponding thought. How on Alpha-Centauri do they manage?
RE: re: New Mission – Report 6
Have new inspiration: Yu-Mans have some sort of an internal symbol manipulation device, part of their psychological makeup, just as telepathic abilities are part of our psychology. Perhaps the usual telepathic development mutated in this species. (I hope it wasn’t as a result of the Interstellar Radiation Dump we set up 20,000 years back in nearby Cygnus 3!) My idea is that this internal symbol manipulation device can be used to build complicated structures, which mirror the structure of thought well enough to be used for communication purposes. Bizarre, I know. Need advice.
FROM: Director, GI Alpha Centauri
TO: CES Jenl Itstre
RE: previous report
Intriguing: you are suggesting that Yu-Mans can manipulate symbols in their minds to mirror the structure of thoughts and then they turn these same symbols into movements of their breathing and food intake orifices which eventually create puffs of air in the surrounding atmosphere! And you are suggesting further that other Yu-Mans can sense these different air-puffs, turn them back into mental symbols and then use that to work out what the original thought was. Amazing abilities, but I can see how telepathic communication could be mimicked in this way. How do you propose to conceptualise these processes?
FROM: CES Jenl Itstre
RE: Hypothesis 2
Following Concept Processing Protocols I propose the following: Yu-Mans’ ability to manipulate symbols to approximate the structure of their thoughts will be syntax, and their capacity to turn these symbols into physical instructions for their mouths and lungs will be phonology.
FROM: Director, GI Alpha Centauri
TO: CES Jenl Istre
Make the study of syntax and phonology your prime concern. It is clearly the key to the communication that makes these Yu-Mans so successful in their non-telepathic environment. It’s also perhaps a good route to understanding the structure of Yu-Mans’ thought processes and the ways in which the turn these thoughts into actions. Repeat: make the study of syntax and phonology your prime …
… ow! As your head hits the table, you wake up. Hmmm, so syntax is about mirroring the structure of thought itself, and phonology involves the amazing ability to glean meanings from how the air moves around you. No wonder they make us study this stuff.
The film turns on the visual language of the heptapods, the name given to the aliens because of their seven tentacular feet. In Chiang’s short story, the spoken language looks pretty familiar to Dr Banks; nouns have special markers, similar to the grammatical cases of Latin or German, that signify meaning; there are words, and they seem to come in particular orders depending on what their function is in the grammar of the sentence. But it is the visual language that is at the heart of the story. This language, as presented in the film, is just beautiful; the aliens squirt some kind of squid-like ink into the air which resolves holistically into a presentation of the thought they want to express. It looks like a circular whorl drawn with complex curlicues twisting off of the main circumference. The form of the language is not linear in any sense. The whorls emerge simultaneously as wholes. The orientation, shape, modulation, and direction of the tendrils that build the whorls serve to convey the meaningful connections of the parts to the whole. Multiple sentences can all be combined into more and more complex forms that, in the film, require GPS style computer analysis. The atemporality and multidimensionality of the heptapods’ written language is a core part of the plot.
I had a brief twitter exchange with @david_colquhoun the other day. Prof Colquhoun tweeted a response to a UCL press release about how learning something about grammar could be good for school children (a point made by Bas Aarts). Colquhoun’s view was that teaching children things about formal grammar was ‘daft’, and I’m sure he’s not alone in this view. When I suggested that learning about a fundamental attribute of human beings was a good thing for children, Colquhoon responded that ‘syntax isn’t an attribute. It’s a custom and it changes’.
I thought this exchange was interesting, though I did find it irksome. A well-respected scientist (a Fellow of the Royal Society), who comes from outside of linguistics, thinks that syntax is trivial enough that it’s legitimate to make categorical (and quite incorrect) pronouncements about it. Would he make such pronouncements about, say, palaeolimnology, or astrophysics? Why are we syntacticians doing such a bad job that academics from other fields think they know enough about syntax to say that it’s a ‘custom’? I don’t know, but I want to just give a few arguments here that syntax does not derive from culture and it is not a custom. The syntax of languages changes as Colquhoun noted, of course, but it does in ways that we have been describing quantitatively for years, and that we have have some theoretical grasp of. None of that work involves thinking of it as a custom.
In fact, one of the important findings of syntactic research over the last 50 years (if not longer) is that structure doesn’t reduce to custom or culture. The idea that the structure of a language is intimately connected to its culture is, I think, quite a common view amongst people who don’t study languages. Though it’s a stronger position than `syntax is custom’, it is, therefore, worthwhile to address first.
This view abjectly fails. Take how languages ask questions about things. In English, if you bought something, and your partner sees the shopping bag, they can ask What did you buy? There’s something funny going on with the syntax here. The part of the sentence that you are asking a question about appears, not where it would go if you weren’t asking a question, but right at the front. I bought a book but Which book did you buy? Linguists call this kind of syntax question-movement (or, more commonly, wh-movement, though that’s not such a good term). There are question-movement languages. English is one, so is Inuktitut, so is Mohawk. But there aren’t question movement cultures. There’s nothing about the culture of the Anglo Saxons and the Mohawks that leads to them having the syntax of question-movement in their languages.
There are also languages that are wh-in-situ. When speakers of these languages ask a question like what did you buy? they say it as you bought what?, leaving the question word in the same place (‘in situ’) as it would be in a declarative statement like you bought a book. This is a property shared by Turkish, Chinese, Malayalam, and many other languages. But again these are not wh-in-situ cultures. There are other languages that adopt a mixed strategy. Indeed, languages with very similar cultures can vary quite wildly in what they do with questions, with some dialects of Italian moving question words and others leaving them in situ.
There is a weaker version of the idea that syntax is custom, which is, I think, what Colquhoun was meaning. If someone comes visiting you, you tell them `the custom in London is to stand on the right hand side of a moving staircase’, so they avoid being mown down by irate Londoners. You can see people obeying customs like this, and once you know the custom, you can tell others what it is. But the syntactic rules of a language are not like this. Take movement of question words in English. Now I’ve told you what that is, perhaps you could think of it like a custom: you know it when you see it, and you can tell me about it. But that won’t work. Though you can say David likes the man you gave the book to, you can’t ask a question about the book, following this `custom’. You can’t say in English Which book did David like the man you gave to? Why not? There are hidden patterns, discovered by syntactic research over the years, that capture when this syntactic pattern works, and when it doesn’t. Unlike standing on the right, English speakers are neither aware of the syntactic patterns they use, nor can they say what they are. In fact, it required a lot of research to find out how this bit of syntax works, and we now have a good understanding of the abstract generalizations that predict whether a speaker of English will react to a sentence positively or not. That speaker has no idea of what these patterns in their use of language are, but syntactic theory can predict them. That’s not how customs work. Syntax is not a custom: it’s a complex, highly abstract system of rules generating patterns in the sentences we use. Different languages have different syntax because, as children acquire a language, there is to-and-fro between the set up of their brains (how human brains process linguistic data they are exposed to and what they do with it) and the linguistic acts of the people around them. Syntax is constrained by the ways our mind works, and it’s within these limits that historical change takes place.
The whole of syntax is like this: there are tiny scraps of it that are similar to customs (don’t end a sentence with a preposition, etc.), and those tend to be what people know about. But these are close to scientifically trivial, though they act as cultural shibboleths. The rest of syntax goes unnoticed; because we find it so effortless, we are unaware of the rich, abstract, and complex flow of syntax in our language.
The study of syntax is a straightforward scientific enterprise. There are many complex facts and phenomena that you need to do a lot of descriptive research to find. There are many possible hypotheses about what is going on, most falsified by new data that comes from observation or experiment. And there are fairly good explanations of many of these in terms of basic theoretical primitives and formalised theories of how they relate and combine. The theories are, without doubt, in a primitive state (maybe around about the level of chemical theory pre-Dalton). Indeed, we are very possibly not even thinking about this stuff in the right way. Nevertheless, research in syntax has been extremely intellectually fertile for years now, revealing many new discoveries about how the grammar of many different languages work, and uncovering broad laws and principles that govern these. There are areas of great controversy, but the basic phenomena, generalisations, and concepts are well researched.
I’ll finish on what the initial tweet was about. Is it a good idea to teach grammar to kids? My own view is that it’s a good idea to teach syntax to kids, looking at many languages, and showing them what some of the basic ideas are. Linguistics, and syntax particularly, is an excellent way to teach the basics of the scientific method. Children can go very quickly from observation to hypothesis to experiment to (dis)confirmation (though less easily to theory, it’s true). All using sentences of their own languages. From this they can learn precision in thinking, the rudiments of science, and, it must be said, some facts about the grammar of the language(s) they speak. None of these is daft.
This brief squib revisits a problem that’s been bothering me for a long time, first noticed by Jim McCloskey in a 1996 paper (presented at the first conference on Celtic I ever went to!). Jim noticed that in Irish certain verbs single argument appeared either as a prepositional phrase or as a nominal phrase. In the former case, the PP seemed to appear low in the structure, while the nominal phrase moved to the standard subject position. This seems like a perfectly sensible pattern, but is actually quite difficult to capture in the kind of approach, mainstream at the time, that said movements had to be triggered. If you need to trigger the movement of the nominal subject, what happens to that trigger when the subject is a PP. This squib shows that some ideas from Chomsky’s recent POP framework make this pattern much less weird. Overall, it suggests that there aren’t EPP features of heads; rather, phrases that can’t stay in situ, end up wherever they can be licensed.
The draft is on lingbuzz at:
I’ve just finished a writeup of my NWAV ‘cornerstone’ talk. It’s an attempt to try to explain why a particular set of changes took place as the Gaelic dialect spoken in East Sutherland, and studied by Nancy Dorian, died in the latter half of the last century. The explanation, unlike Dorian’s, which relied on primarily sociolinguistic factors, appeals to the weakening and loss of syntactic agreement features, and how this leads to the replacement of a variety of null pronoun structures with alternatives. Which particular alternatives are chosen depends on what I call the ‘syntactic ecology’ of the language.
The draft is on Lingbuzz at: