Syntax random

[akam chinjir]

I wrote a reply to bradrn's question (on Linguistic Miscellany) about Ā movement, but it got too long for a comment, so I'm putting it here. And I'm putting it in a comment rather than the main post because I suppose it wouldn't be terrible to have a sandbox for reasonably involved syntax stuff.

[akam chinjir]

So this is my attempt to explain what Ā movement is. I hope it makes sense. And that there aren't too many mistakes

First pass
First thing: "Ā" is meant to be read "A-bar," and that's how I'll write it. (It's sometimes written "A'" instead.) And A-bar movement contrasts with simple A movement, the bar (or line above, or prime mark) indicating negation.

Second thing: the A/A-bar distinction is one of the worst naming conventions in all linguistics. Especially when you realise it has nothing to do with X-bar theory. It can also be maddening to try googling it.

The A stands for Argument, and the issue here is a contrast between movement that's somehow specific to a verb's arguments (A movement), and certain other sorts of movement (A-bar movement). The idea is that in many languages, there are syntactic requirements that one or more of a verb's arguments have to move to a particular position in the clause's structure, and that sort of movement is called A movement.

English has a requirement of this sort: a finite clause must have a syntactic subject, that is, an NP that occupies a particular position in the clause's structure, and most often this requirement is satisfied by moving one of a verb's arguments into that position. (There are also expletive subjects, which I won't talk about here.)

English also has a requirement that question words in non-echo non-quiz-show questions get moved to the top of the clause. This is not A movement: it doesn't have to target arguments, it doesn't even have to target NPs, and its target is distinguished not by its position in the structure but by some additional characteristic or feature. That's to say, this is A-bar movement rather than A movement.

Second pass
I'm assuming a picture where a clause is assembled bit-by-bit, in such a way that the result is a tree structure (a binary tree structure, to be a bit more precise).

In this picture, a verb generally gets combined with its arguments in a fixed order. There's some room for variation (both within and between languages), but, for example, if a verb's core arguments include an agent and a patient, the verb will combine with the patient before the agent. As a result, the agent will start out higher in the tree structure than the patient does. (I won't try to include a diagram, but I hope it's easy to imagine the sort of tree structure I'm imagining. I'm also assuming that in a passive, even if there's a by-phrase or some equivalent, the overt agent doesn't count as a core argument of the verb.)

A verb's arguments need not stay in their initial ("merge") position, it's common for them to have to move higher in the clause as various other things get added. Some of these movements, the A movements, seem to have only a syntactic motivation. Like, English requires the subject to move, but Italian doesn't, but this by itself doesn't result in any semantic or pragmatic differences between English and Italian sentences, it seems to be pure syntax. And similarly in other languages that require object movement.

When I say that it seems to be pure syntax, in part I mean that no one has any really good idea why it happens. Like, the English requirement that subjects occupy a particular position is sometimes called the extended projection principle (EPP), but that's just a label (and a terrible one), not an explanation. And the same is true for other terminology that's used to describe this sort of phenomenon---like abstract case, or strong features, or nominal licensing.

Further, it seems to be purely a syntactic question which NP gets moved by A movement. For example, when English moves an argument into subject position, it'll always move the agent if there is one, but if there's only a patient, then it'll move the patient. That is, it's sensitive to the position of the argument in the sentence's structure (agents are merged higher than patients), not to semantic role. (But in a language where objects also A move, it can be tricky to explain how this happens the way it does; and that's not the only sort of puzzle that can arise.)

In any case, A movement is the movement of one of a verb's core arguments from its original merge position to some higher position in the clause, and it happens for purely syntactic reasons, whatever exactly that means. Whereas A-bar movement need not involve a verb's core arguments, and generally happens because the thing that move has some particular, often pragmatic, significance.

Kinds of A-bar movement
There are various common sorts of A-bar movement, which I'll run through now.

I've already mentioned question words, and in languages that allow fronted question words that can be one of the plainest sorts of A-bar movement. Here's an example:

(1) Whatᵢ did I eat tᵢ?

(I'm using the common convention of using t to show the earlier position of a moved element, and a subscript to make it clear which overt element it corresponds to.)

The idea is that in this sentence, "what" has moved from after (lower than) the verb to a position high in the clause (the front). This obviously doesn't satisfy a general syntactic requirement on objects. Instead, it happens because "what" is a question word, and in a non-echo, non-quiz-show question, English requires question words to move to the top.

In many languages, including English, you also get A-bar movement in relative clauses. That's to say, there's a gap in the relative clause from which an operator of some sort moves to the front of the clause. When there's an overt relative pronoun, this is usually taken to be the element that's moved:

(2) the person [who₁ I saw t₁]

But it's also possible for the moved element not to be pronounced:

(3) the person [∅ᵢ I saw t₁]

(Why think that something has moved in that case? One reason is that you get island effects regardless of whether an overt relative pronoun is present: "the person who₁|∅₁ the claim that I saw t₁ surprised her" is bad either with or without "who." Another sign that there's movement here is that you can get parasitic gaps: "the person ∅́∅₁ I saw t₁ without greeting t₁." ---I'll come back to parasitic gaps, but not to islands.)

You also get A-bar movement in some forms of topicalisation. My English allows this, at least for contrastive topics:

(4) The apple I ate (but not the banana)

However, many languages allow sorts of topicalisation that probably don't involve movement at all. English "as for" topics are of this sort, for example:

(5) As for the apple, I ate it

(Note that you need the resumptive pronoun here, and there'll often be a pause where I've put a comma.)

Finally, I'll mention focus movement. I don't think we do this in English, but it's pretty common for there to be a position in a clause to which focused elements will move. That can be the front of the clause, but in SOV languages it's also common to have a focus position immediately before the verb.

Diagnostics
I'm going to mention a couple of diagnostics that can be used to distinguish A movement from A-bar movement. I find this stuff pretty hard to get my head around, and am a bit nervous that I'll screw up, but here goes.

First, weak crossover.

Take this example:

(6) ?Her₁ friends admire Mary₁

This is supposed to be a bad, maybe unacceptable, on the assumption it's saying that Mary's friends admire Mary. (It's totally fine if it's supposed to be talking about someone else's friends.)

(7) ?Her₁ friends admire who₁?

And this is also supposed to be bad, on the assumption that it's asking for someone₁ whose friends admire her₁, the very same person. (You also have to interpret it as an echo question.)

(8) ?Who₁ do her₁ friends admire?

Again this is supposed to be bad, on the assumption that it's asking about someone who's admired by her own friends.

But this is fine:

(9) Who₁ is admired by her₁ friends?

(8) puts "who" before "her" by wh-movement, a sort of A-bar movement, and this isn't enough to let "who" bind (be coreferential with) "her." (9) also puts "who" before "her," but does so by passivisation: even before wh-movement (and even if this is an echo question with no overt wh-movement), "who" is the subject and will end up before "her" by plain A movement.

The fact that (8) is bad even though "who" has moved before "her" is an example of what's called weak crossover. Basically, this happens only with A-bar movement.

I'll also mention parasitic gaps.

Consider these two sentences:

(:) *I ate the apple without tasting.
(11) What did I eat without tasting?

In (:), but not in (11), you need "tasting" to have an overt object ("I ate it without tasting it" would be fine). In (11), the object is gapped, and this appears to be possible because the question word has left its own gap:

(12) What₁ did I eat t₁ without tasting t₁

The second gap here is called a parasitic gap, because it's supposed to be parasitic on the gap left by the movement of the question word.

And you can do this with other forms of A-bar movement:

(13) the apple that I ate without tasting
(14) The apple I ate without tasting

((13) illustrates a relative clause, (14) topicalisation.)

So there seems to be a general pattern, that A-bar movement does and A movement does not license parasitic gaps.

Final thoughts
There are a lot of loose ends there, but also this is already really long, so I won't try to address them in this post. But I do want to make one connection with ergativity, since bradrn was wondering about all this after reading about ergativity.

Many languages with ergative morphosyntax have restrictions of the following sort: with a transitive verb, the object can be A-bar moved, but the subject cannot. For example, you can't directly question the subject, or you can't focus it or relativise it. In fact this is the form that syntatic ergativity takes: usually it's specifically a restriction on A-bar movement. (You'll also see the movements in question called "extraction.") Syntactic ergativity doesn't always target all forms of A-bar movement, but it's most often one or more forms of A-bar movement that's at issue---so it's helpful to understand that in some ways these types of movement behave the same.

Incidentally, it's a big puzzle why so many morphosyntactically ergative languages have restrictions on A-bar movement. There are some analyses of some nonergative languages where there are analogous restrictions on the A-bar movement of objects, but this seems to be a rare phenomenon. (There's the Austronesian pattern that allows only one privileged argument to be extracted, and that argument is often identified as a subject, but this doesn't seem to be an accusative counterpart of syntactic ergativity.)

I'll mention another issue. It's not always clear which if any of a verb's arguments is the syntactic subject, in the sense that it obligatorily A-moves to a position high in the clause. Dyirbal, for example, often has patient-agent-verb order in clauses with two overt NP arguments, and you might think that the reason for its syntactic ergativity is that for some reason it's the patient argument that A moves into the subject position; and this sort of analysis has been given for a number of syntactically ergative languages.

In principle, one way to test this would be to try to apply diagnostics of the sort I've mentioned to see whether different word orders pattern differently with respect to scope or parasitic gaps or whatever. That's not likely to be easy, because grammars rarely provide the data that you'd need to settle questions like this, and in many languages it can be very hard to apply the tests. Like, one sort of test I haven't mentioned is the behaviour of words meaning every; but most langugaes don't have a word meaning every (they have all instead). Or some languages let you drop topical objects in any context, making it impossible to test for parasitic gaps.

In fact there are other issues involving subjects and topics and A movement and A bar movement that have nothing to do with ergativity. But that's enough for now.

[bradrn]

This is an amazingly thorough explanation — thank you akam chinjir!!

So, if I’m understanding correctly: Ā-movement is a type of movement which occurs at least partly on semantic or pragmatic grounds, while A-movement is motivated on purely syntactic grounds. Is that correct?

Also, a couple of (very minor) questions and comments:

English also has a requirement that question words in non-echo non-quiz-show questions get moved to the top of the clause.

I like the ‘non-quiz-show’ requirement. Another one I can think of along the same lines is that the question word must not be focalised (e.g. ‘You did what‽’ has no Ā-movement).

I'm assuming a picture where a clause is assembled bit-by-bit, in such a way that the result is a tree structure (a binary tree structure, to be a bit more precise).

I’d be curious to know: why a binary tree specifically?

… you get island effects regardless of whether an overt relative pronoun is present: "the person who₁|∅₁ the claim that I saw t₁ surprised her" is bad either with or without "who."

I know you said you weren’t covering this, but what’s an ‘island effect’?

Diagnostics

I’d be interested to know how reliable those diagnostics are for (non-ergative) non-Indo-European languages. Are parasitic gaps and weak crossover English-specific (or Indo-European-specific), or are they more general properties of languages worldwide?

(There's the Austronesian pattern that allows only one privileged argument to be extracted, and that argument is often identified as a subject, but this doesn't seem to be an accusative counterpart of syntactic ergativity.)

Well, there is one article (which I believe was originally recommended by you) which appears to analyse Austronesian languages as being syntactically ergative. (I’m not sure I agree with that particular conclusion, but I do think that this could well be an instance of the same general pattern.)

[akam chinjir]

I’d be curious to know: why a binary tree specifically?

It's actually a tricky question, and one I'm not sure I can answer well, though it's fair to say that most linguists these days working in a broadly Chomskyan framework are convinced that's how it works.

One reason for this is just that taking two bits and putting them together (and thus forming a node with two branches) is about the simplest structure-building operation you can imagine.

Another is that it can be easier to go from tree structure to linear order if you restrict yourself to binary branching. Like, it's widely assumed that under any node, one of the branches will be privileged---it'll be the head of the phrase, or at least contain the head of the phrase. If you've always just got that privileged element and one other branch, it's easy to state rules about which branch should end up getting pronounced first. But if you've got three or more branches, and only one of them is privileged, rules like that are trickier.

And then there are particular cases. Like, if you look at how languages treat clauses with both an overt agent and an overt patient, in many and perhaps all languages you can find reason to suppose that the verb combines with the patient before it combines with the agent, even in languages where there's plainly no VO constituent in surface order (like VSO languages). Like, if the verb is "take," the sort of eventuality that's being described can be greatly affected by the object---"take a shower" vs "take a walk" vs "take a knee" vs "take a photograph" and so on---but you don't get the same effect with agents. So that maybe is a reason to think that you've got a node governing the verb and the object to the exclusion of the subject.

I know you said you weren’t covering this, but what’s an ‘island effect’?

There are certain contexts such that it's ungrammatical or bad for an element to be A-bar moved out of that context. In my example, you can't move X out of "the claim that I saw X surprised Mary"; the 'complex subject' "the claim that I saw X" is an island. That's why these are ungrammatical:

*the person who₁ the claim that I saw t₁ surprised Mary
*Who₁ did the claim that I saw t₁ surprise Mary?

Sentences like these can be sufficiently ungrammatical that it's hard to tell even what they're supposed to mean---that's actually why I've included the traces, to try to keep that clear. But at the same time, people, even native speakers of the same language, disagree about particular cases fairly often. And there's plenty of cross-linguistic difference here too.

It's often possible to improve things by leaving a resumptive pronoun instead of a gap:

?he person who₁ the claim that I saw him₁ surprised Mary
?Who₁ did the claim that I saw him₁ surprise Mary?

I’d be interested to know how reliable those diagnostics are for (non-ergative) non-Indo-European languages. Are parasitic gaps and weak crossover English-specific (or Indo-European-specific), or are they more general properties of languages worldwide?

I don't know much about this, tbh. Phenomena such as weak crossover and parasitic gaps (and island effects) are certainly discussed in the contexts of a wide variety of languages, but for reasons I mentioned in the main post it can be hard to get data about this sort of thing without doing primary research on a language. (Like, it's no coincidecne that the WALS chapters on relative clauses draw the superficial distinctions they do rather than asking whether relative clauses in a language require movement or are subject to island constraints; most grammars don't cover that sort of thing.) So it's a bit hit or miss what you can easily find out.

Well, there is [url=https://nlp.stanford.edu/~manning/disse ... y-1994.pdf]one article[/url] (which I believe was originally recommended by you) which appears to analyse Austronesian languages as being syntactically ergative. (I’m not sure I agree with that particular conclusion, but I do think that this could well be an instance of the same general pattern.)

Yeah, that's definitely an analysis that's out there. I don't know much about it, though I think the last time I looked into it I ended up warming to the idea more than I had previously. In particular I ended up thinking I should look into what's been done on possible similarities between restrections on extraction in Austronesian and Mayan languages. (But I haven't actually gone on to do that yet.)

[bradrn]

I’d be curious to know: why a binary tree specifically?

It's actually a tricky question, and one I'm not sure I can answer well, though it's fair to say that most linguists these days working in a broadly Chomskyan framework are convinced that's how it works.

One reason for this is just that taking two bits and putting them together (and thus forming a node with two branches) is about the simplest structure-building operation you can imagine.

That makes sense. (Although I’m not convinced that languages would always use ‘the simplest structure-building operation you can imagine’ — I wouldn’t be surprised if languages actually work far less rationally than that!)

Another is that it can be easier to go from tree structure to linear order if you restrict yourself to binary branching. Like, it's widely assumed that under any node, one of the branches will be privileged---it'll be the head of the phrase, or at least contain the head of the phrase. If you've always just got that privileged element and one other branch, it's easy to state rules about which branch should end up getting pronounced first. But if you've got three or more branches, and only one of them is privileged, rules like that are trickier.

So then what about sentences like ‘I saw you in the park at noon’/‘I saw you at noon in the park’, where there seem to be two branches (‘at noon’, ‘in the park’) which can occur in either order? It seems to me that a node with three branches would work well in that case.

Like, if the verb is "take," the sort of eventuality that's being described can be greatly affected by the object---"take a shower" vs "take a walk" vs "take a knee" vs "take a photograph" and so on---but you don't get the same effect with agents. So that maybe is a reason to think that you've got a node governing the verb and the object to the exclusion of the subject.

I remember Dixon described this as a phenomenon which treats S and O the same in all languages, and I was trying to figure out an example for use in my ergativity thread, but I didn’t find an example as nice as this one! May I stea

I know you said you weren’t covering this, but what’s an ‘island effect’?

There are certain contexts such that it's ungrammatical or bad for an element to be A-bar moved out of that context. In my example, you can't move X out of "the claim that I saw X surprised Mary"; the 'complex subject' "the claim that I saw X" is an island. That's why these are ungrammatical:

*the person who₁ the claim that I saw t₁ surprised Mary
*Who₁ did the claim that I saw t₁ surprise Mary?

Sentences like these can be sufficiently ungrammatical that it's hard to tell even what they're supposed to mean---that's actually why I've included the traces, to try to keep that clear. But at the same time, people, even native speakers of the same language, disagree about particular cases fairly often. And there's plenty of cross-linguistic difference here too.

It's often possible to improve things by leaving a resumptive pronoun instead of a gap:

?he person who₁ the claim that I saw him₁ surprised Mary
?Who₁ did the claim that I saw him₁ surprise Mary?

For me, even the examples with the resumptive pronoun are mostly ungrammatical. I think I would prefer something like ‘Who was the person for which the claim that I saw him surprised Mary?’.

[akam chinjir]

(Although I’m not convinced that languages would always use ‘the simplest structure-building operation you can imagine’ — I wouldn’t be surprised if languages actually work far less rationally than that!)

The argument breaks down for me there, too. It sounds good if you're thinking about a creature with two hands manipulating physical objects, but if you're talking about set theory---and in the contexts in which Chomsky gives this sort of argument, you could be forgiven for thinking that he's talking about set theory---the restriction to two elements seems to me to be an unnecessary extra.

So then what about sentences like ‘I saw you in the park at noon’/‘I saw you at noon in the park’, where there seem to be two branches (‘at noon’, ‘in the park’) which can occur in either order? It seems to me that a node with three branches would work well in that case.

Yeah, the linearisation argument loses some power if you're talking about languages or constructions with free-ish word order. In fact it's fairly common to think that adjuncts (like in your examples) are subject to somewhat different rules, partly because they can have freer ordering and because they seem less integrated into the clause's structure. (For example, adjuncts are islands )

Though you can get scope effects that suggest a particular binary analysis. Like, "I saw you in the park at noon" allows the interpretion that you were in the park when I saw you, but I wasn't (maybe I was perched on top of a nearby building); maybe that reading is harder to get with "I saw you at noon in the park"; it's certainly harder to get with "At noon, in the park, I saw you." And you can get contrasts like this, suggestion that the structure isn't 'flat' (assuming you share my judgments!):

I saw you with Dolly₁ listening to her₁ news

*I saw you listening to her₁ news with Dolly₁

[KathTheDragon]

And you can get contrasts like this, suggestion that the structure isn't 'flat' (assuming you share my judgments!):

I saw you with Dolly₁ listening to her₁ news

*I saw you listening to her₁ news with Dolly₁

Isn't that rather a constraint on the use of "her" as an anaphor/cataphor? Compare:

I saw you listening to Dolly₁'s news with her₁

[akam chinjir]

Isn't that rather a constraint on the use of "her" as an anaphor/cataphor? Compare:

I saw you listening to Dolly₁'s news with her₁

Maybe? I think the main issue though is what the constraint is. I was thinking of it as implying that the antecedent has to be higher in the structure than the pronoun, though as far as those particular examples are concerned, it could just be that it has to precede the pronoun in linear order (so they weren't very good examples after all).

[KathTheDragon]

I'm pretty sure in general it is linear order. There are cases where the pronoun comes before its referent ("When I saw her₁, Dolly₁ was in the park") but I'm sure there's a general rule to be made there that doesn't need a Chomskian grammar.

[Richard W]

I’d be curious to know: why a binary tree specifically?

It's actually a tricky question, and one I'm not sure I can answer well, though it's fair to say that most linguists these days working in a broadly Chomskyan framework are convinced that's how it works.

It gives you a sharper theory. Unfortunately, sometimes life just ain't like that.

It's a bit like phylogenists looking for binary branching trees. It's nice when it works, but it doesn't always. When the phylogenists get a 3-way split, it's a bad enough that they have a special cuss-word for it, a 'hard trichotomy'.

[akam chinjir]

A friend caught a mistake in the weak crossover bit (no surprise there). Apparently you're expected to think that "Who₁ do her₁ friends admire?" is substantially worse than "Her₁ friends admire Mary₁," not just that they're both bad. So for some reason the (A-bar) movement makes things worse.

As for structure dependence, here's one common sort of contrast:

Mary₁'s friends admire her₁

*Mary₁ admires her₁

The first one is okay, the second one isn't, and it's reasonable to suppose that the reason for this is that "Mary" is embedded in a larger NP in the first one but not in the second one.

(There's actually a fair bit about this stuff in zompist's Syntax Construction Kit.)

[zompist]

I'm pretty sure in general it is linear order. There are cases where the pronoun comes before its referent ("When I saw her₁, Dolly₁ was in the park") but I'm sure there's a general rule to be made there that doesn't need a Chomskian grammar.

Yes: the pronoun can appear first if it's in a subclause.

[KathTheDragon]

Is that the entire general rule? Are there other cases, or exceptions?

[zompist]

Is that the entire general rule? Are there other cases, or exceptions?

This is language; everything has exceptions.

But it's a pretty good rule! One possible exception (from my book):

Knowing that he_i was a failure didn't bother Wilmer_i.


If you accept the idea of the cycle, this can be explained. It's a little involved to reproduce in a post, but doesn't require X-bar syntax. But the basic idea is that the antecedent "Wilmer" appeared before knowing but got deleted.

[KathTheDragon]

Right, I should've said "major classes of exceptions".

[zompist]

I might not have been clear that I was answering your question. If there are exceptions to the rule I gave, I'm not aware of them. The quasi-exception I'm aware of, I talked about.

But syntax often proceeds when people find a new example that breaks the previous rules.

[Kuchigakatai]

Hmm, a random thread on syntax. I have so many questions about syntax.

(I guess I should've fed them one by one, but meh, I have time to write them all down right now...)

1) Why do syntacticians often talk about working with a framework, instead of a number of individual tools? For example, the concept of phrase structures has been clearly found to be pretty useful, and lots of people, whether they're syntacticians or people as far away as professors of literature writing a practical reference grammar, are using them now. Why not treat binary trees or multi-dependent trees that way, and so on? In the realm of phonology there aren't any pre-packaged frameworks... Your phonetic notation or the way you structure syllables is not tied to whether you derive phones from phonemes with rules on one hand, or on the other hand with a mysterious generator + ranked constraints (Optimality Theory).


2) I think I don't need to argue with anyone here that it is impossible to avoid using some model of syntax when talking about any language, natlang or conlang, whether you're talking about syntax naïvely or not. Now, a lot of books these days (say, in the last 40 years or more) that teach the syntax of a given language in a practical way, such as a lot of reference grammars for language learners, and I mean a loooooot of such books, use a syntax model that never uses trees but assumes some kind of dependency grammar tree with nodes groupable into phrase structures, in which traces are also never used when talking about movement (words are simply said to move around). Why do these common conventions don't have a name?


3) And relatedly, why do English-speaking syntacticians in particular like to pretend it doesn't exist, when huge swaths of syntax are in fact mostly talked about in this form?

I've noticed that over at the French and Spanish-speaking world this model is explicitly taught in university courses that have the word "syntax" in the title (or perhaps "grammar" or "linguistics"). For example, the above model underlies Samuel Gaya's textbook Curso superior de sintaxis española (Advanced Course in Spanish Syntax), as well as the reference grammars of the Royal Spanish Academy (RAE). I once took a course on French linguistics, in French, where we went over a range of ideas to analyze French grammar that were made by French linguists over time, and although we briefly covered Lucien Tesnière's Dependency Grammar in its purer form, for the most part we used the practical hybrid model above plus trees, drawing trees with phrase structures with plenty of commentary on the dependency of individual words to others, all while using the professor's own handouts. And again, in all of these works, the model is indeed never presented with a fancy name such as "Lexical-Functional Grammar" or "Croft's Radical Construction Grammar", but simply as a set of various tools.


4) Is anyone here brave enough to improve English Wikipedia's article on syntax? I noticed a month ago that it mostly consists of tiny section on S-V-O order and then a few packaged frameworks. It is doing nothing to dispel the all-too-common notion that syntax is about these two things, as opposed to any phenomenon that has to do with multiple-word constructions, the selection of word-forms through agreement or the lexical properties of the verb or construction, transformations like questions or passives, etc.


5) Actually, now that I mentioned "transformations"... Correct me if I'm wrong, but assuming I correctly remember hearing various times that Chomsky came up with the concept of transformations in the mid-60s, why do people say this?

About 30 years before Chomsky started to say "mama" and "papa" for the first time, a team of five Latinists (Joseph Henry Allen, James Bradstreet Greenough, George Lyman Kittredge, Albert A. Howard, and Benjamin D'Ooge) produced an influential reference grammar, Allen and Greenough's New Latin Grammar for Schools and Colleges (1903), in which they show immense concern to document the transformation of sentences into questions, indirect speech, and indirect questions. And they do talk about them in terms of basic, direct, declarative sentences being transformed into those other types, e.g. describing the choice of a specific subjunctive in an indirect question as matching an indicative tense it'd have in the declarative equivalent. And I really very much doubt they were the first of thinking of constructions that way. I don't think Chomsky's concept was all that innovative?

I guess there was innovation in treating it as a general phenomenon found in languages at large, since the very concept of linguistics and syntax as discipline and subdiscipline was just getting started, but if we talk about about the concept itself of transforming basic declaratives into other types of clauses...

[zompist]

1) Why do syntacticians often talk about working with a framework, instead of a number of individual tools? For example, the concept of phrase structures has been clearly found to be pretty useful, and lots of people, whether they're syntacticians or people as far away as professors of literature writing a practical reference grammar, are using them now. Why not treat binary trees or multi-dependent trees that way, and so on? In the realm of phonology there aren't any pre-packaged frameworks... Your phonetic notation or the way you structure syllables is not tied to whether you derive phones from phonemes with rules on one hand, or on the other hand with a mysterious generator + ranked constraints (Optimality Theory).

By "syntacticians" I think you mean "Chomsky and his followers." Chomsky is the best and the worst thing to happen to syntax. He created a system that was simple and attractive, but he didn't like where some people took it in the late 1960s, and started a process of changing things while at every point insisting that his was the only and correct theory.

Other syntacticians, notably James McCawley, have done exactly what you recommend: look at all the theories and pick what seems most useful. The cognitive or relational linguistics people, in reaction to Chomsky, resist having either a single syntactic framework or a leader.

All that said, we use frameworks because they help us discover the data. I mentioned exceptions above; many of the important sentences that forced changes in theory would not have been discovered if people weren't working within a framework. The value of syntax is not the observation that questions resemble statements; it's a mass of strange facts about pronouns, quantifiers, conjunctions, and so on.

5) Actually, now that I mentioned "transformations"... Correct me if I'm wrong, but assuming I correctly remember hearing various times that Chomsky came up with the concept of transformations in the mid-60s, why do people say this?

Chomsky's work on transformations is mid-50s. He certainly didn't invent the concept— I haven't seen the book you mention, but Otto Jespersen's English grammar is often cited as a precursor.

As a counter-example, though, I'd point to Whitney's Sanskrit grammar of 1879: hundreds of pages, and no syntax section. Though the 1909 update of Gesenius's Hebrew grammar does have a pretty good syntax section. (Perhaps it's not coincidence that Chomsky's master's thesis is on Hebrew.)

[bradrn]

2) I think I don't need to argue with anyone here that it is impossible to avoid using some model of syntax when talking about any language, natlang or conlang, whether you're talking about syntax naïvely or not. Now, a lot of books these days (say, in the last 40 years or more) that teach the syntax of a given language in a practical way, such as a lot of reference grammars for language learners, and I mean a loooooot of such books, use a syntax model that never uses trees but assumes some kind of dependency grammar tree with nodes groupable into phrase structures, in which traces are also never used when talking about movement (words are simply said to move around). Why do these common conventions don't have a name?

This might fall under what Dixon calls ‘Basic Linguistic Theory’, but I’m not sure.

3) And relatedly, why do English-speaking syntacticians in particular like to pretend it doesn't exist, when huge swaths of syntax are in fact mostly talked about in this form?

I wish I knew the answer to this!

[akam chinjir]

What's supposed to be in a framework that's not in a bunch of tools? I suppose it might also include various results that you take to have been established, a set of priorities, some shared points of reference, things like that. Hard to see how those could be bad things, imo.

Oh, on the first point, here's an interesting list of significant mid-level results of generative linguistics that a group of linguists came up with a few years ago.

Actually, if you read that, you might notice that the structure-dependence of binding is on the list. Does analysing things in terms of structure count as a tool? C-command in particular? Cyclicity? Those are pretty fundamental to how most syntacticians go about things, at least. (Whereas linear order isn't part of the usual toolkit.)

I guess transformations were also a tool, though Ser is right that they're not really used in Chomskyan approaches anymore. (Though I think they lasted past the '60s, till the earlyish '80s and Government and Binding, if I remember right.) And actually the reasons why people dropped them suggests they're more than just a tool. The two big changes, as I understand it, were abandoning construction-specific mechanisms, and (later, in minimalism) the rejection of the idea of deep structure. (So in G&B it might make sense to say that corresponding active and passive sentences are both derived from the same deep structure, just the derivation wouldn't use construction-specific mechanisms; whereas nowadays it's very unlikely you'd find someone saying they share a deep structure in the first place.) So giving up the idea of transformations seems like more than giving up a tool, you're also giving up fairly significant bits of theory.

As for the approaches that Ser mentioned that I guess don't count as frameworks, I don't know that not offering university courses about a thing counts as pretending a thing doesn't exist. Like, say, if you looked at Kayne's book on French syntax, are there relevant phenomena he misses or mistakes he makes because he doesn't pay attention to that work? (That's assuming he doesn't, I don't know either way.)