Syntax random

[evmdbm]

I'm asking the same question here as I do on Bradrn's ergativity thread, at his suggestion. It comes out of his description of syntactic ergativity and while I think I understand how A-bar movement works with relative clauses, I'm less clear as to how it works with wh-questioning and how that works with syntactic ergativity.

So...if A-bar movement involves (well) movement then "John kicked the cat".... "Who kicked the cat?" involves no movement; the subject of both English sentences is in the same place (SVO word order). It is only with "What did John kick?" that there is movement (it's now object first), so I'm guessing that this means subjects aren't extracted in English, or am I nor understanding extraction in this context?

Pretending English is ergative for a second, in a language with AVO word order replacing John-erg with who-erg involves no movement either. It is only "What-abs did John kick?" that involves movement. Are we treating ergative and absolutive demonstrably differently here? Is that absolutive extraction only and therefore an example of syntactic ergativity? I'm leaving be the fact that in a sentence like John-abs weeps... who is weeping? There's no movement either...

Enlightenment anyone?

[zompist]

So...if A-bar movement involves (well) movement then "John kicked the cat".... "Who kicked the cat?" involves no movement; the subject of both English sentences is in the same place (SVO word order).

It does, though, in Chomskyan syntax. "Who" moves from its original location upward to the TP node.

It's hard to explain why in any quick fashion... plus I have to get to the gym... but the superquick idea is that we need a TP node for other reasons, and it produces some useful simplifications to use it for subject extraction. (One being that the movement rule applies to all sentences.)

(That doesn't mean it's a great idea. But I just wanted to point out that in his versions of syntax, you still have movement here.)

[bradrn]

Pretending English is ergative for a second, in a language with AVO word order replacing John-erg with who-erg involves no movement either. It is only "What-abs did John kick?" that involves movement. Are we treating ergative and absolutive demonstrably differently here? Is that absolutive extraction only and therefore an example of syntactic ergativity? I'm leaving be the fact that in a sentence like John-abs weeps... who is weeping? There's no movement either...

Actually, I just remembered one thing I forgot to mention when you first asked this question. In a language with syntactic ergativity, you can’t usually ask questions like *Who-erg kicked the cat-abs? — you can only say What-abs did John-erg kick. Even though it appears that the second sentence here has movement while the first doesn’t, there is still a demonstrable difference between them, since the second is ungrammatical while the first is not.

[akam chinjir]

I'll follow up on zompist a bit.

One thing is that if you've got a language in which ergative subjects cannot be questioned or relativised or whatever, that's by itself reason to think that the constructions in question involve movement, even if the movement isn't evident on the surface. Syntactic ergativity usually, and maybe always, involves constraints on constructions that in many language involve A-bar movement (questions, relative clauses...), and those constraints are naturally thought of as constraints on A-bar movement---so wherever they apply, that's a reason to suspect movement.

In English you don't have anything quite like that. In fact I'm not sure that there's any construction-specific reason to think there's movement in English subject questions. So maybe there's no movement in English subject questions. (That's an option available to Chomskyans, by the way, maybe I disagree with zompist about that.)

That's all put in terms of movement, though that might not be essential. And actually I've been wanting to try to formulate some thoughts about movement, so I'll try that out here. Er, it got pretty long, so I've gone and put it between [more] tags.

More: show

I'll start with a simple relative clause:

the booksᵢ that he wrote ∅ᵢ are interesting

I've included a bit of theory: a ∅, signifying a gap, and subscripts that link that gap to the noun "books." That's meant to be pretty minimal theory, though, something most everyone who thinks about relative clauses can probably agree with.

The idea is that to understand this sentence, you've got to understand "the books" both as the subject of "are interesting" and as the object of "wrote"---as if it has two positions in the sentence, one in the matrix clause and one in the relative clause. Of course it's only pronounced in the matrix clause, and that's why we posit a gap in the relative clause.

Now try this one:

*the booksᵢ that she asked who wrote ∅ᵢ are interesting

As the asterisk indicates, this is supposed to be ungrammatical; anyway it's ungrammatical in my English. And in fact analogous sentences are ungrammatical in many languages.

It turns out that there are some very good generalisations about when you get failures like that: there certain configurations, which get called islands, and a gap that occurs within an island can't get linked to a position outside the island. "who wrote ∅ᵢ" is what's called a wh island, and it's a really good generalisation that a gap inside a wh island can't get linked to a position outside it.

One reason these generalisations are interesting is that they tend to cover not just relative clauses but also questions. So this is also bad (at least in my English):

*Whatᵢ did she ask who wrote ∅ᵢ?

That's to say, relative clauses in many languages and questions in many languages (and in fact focus constructions in many languages) are governed by a fairly unified set of island constraints, and it's reasonable to suppose that when they are, there's some particular thing that's going on that explains why they're subject to island constraints.

Now, it's not the case that all relative clauses and questions in all languages are governed by island constraints. Here's a Mandarin example:

你買了誰寫的書？
nǐ mǎile [shéi xiě  ] de shūᵢ
2s buy    who  write  DE book
"Who is the x such that you bought books that x wrote?"

(I'm taking this and other upcoming Mandarin examples from lecture notes by Michael Yoshitaka Erlewine and Hadas Kotek, available here.)

This contains a relative clause (which I've bracketed), and relative clauses themselves are islands---but the question is fine. An obvious guess is that it's fine because the question word is in situ, so there's nothing in the island that has to get linked with something outside of the island.

It's not so simple, though:

*他為什麼寫的書最有趣？
 [tā wèishénme xiě  ] de shū  zuì  yǒuqù
  3s why       write  DE book most interesting
 Intended: "What is the reason such that the books he wrote for that reason are most interesting?"

Here we have a relative clause island and what looks like an island effect---as if the question word wèishénme "why" has to get linked to some position outside of the island, even though nothing is pronounced in that position.

Here's why I think that's interesting. In a case like "the books that she asked who wrote," it's pretty easy to motivate the idea that there's a gap inside the relative clause corresponding to the noun "books" outside the relative clause, and it's pretty intuitive that the issue is the relation between the gap and that noun. In the Mandarin case (the second, ungrammatical one), there's no gap inside the relative clause, but it's a bit like there's a gap outside the relative clause, and the problem is that the question word inside the relative clause island can't get linked to the gap outside it.

In case you're thinking that the problem with that Mandarin sentence is just semantic or pragmatic, here's another good one:

他為了什麼原因寫的書最有趣？
[tā wèile shénme yuányīn xiě  ] de shū  zuì  yǒuqù
 3s for   what   reason  write  DE book most interesting
"What is the reason such that the books he wrote for that reason are most interesting?"

The semantics and pragmatics haven't really changed. Only the question word is different, wèile shénme yuányīn "for what reason," instead of wèishénme "why." (The Erlewine and Kotek lecture notes talk a bit about why this might be so.) So the problem with the previous example probably isn't semantic or pragmatic in nature.

It's a bit as if some Mandarin question words have to get linked in an island-sensitive way with a gap (or whatever) somewhere else in the sentence, and some do not, even though all the question words are in situ. (And it's also not the case that sentence-initial question words always give rise to island effects, though I don't have examples to illustrate that.)

Anyway, all of that is meant to be fairly neutral between approaches to syntax. Now I'm going to sketch three very general ways you might try to account for what's going on in these cases.

First, there's the unification approach. This doesn't go much beyond what I've already said, in that it proceeds by placing constraints on what positions in a sentence can be linked in the way that's needed for some relative clause and questionn constructions. Er, I know very little about how this approach gets carried out, but my understanding is that this is how LFG and HPSG do things. (And of course once you start spelling out how the constraints are supposed to work, it'll get theory-specific pretty quickly.)

Second, there's movement. The idea here is that a sentence is put together in stages, and that one of its constituents can be in one position at one stage in the derivation, and in another stage in the derivation: it moves. So you might suppose a derivation that goes in part like this:

        are you eating what
→ whatᵢ are you eating tᵢ

(This is how things are described in Government and Binding, and in G&B movement has to leave a trace; that's the tᵢ, and it corresponds to the gap in earlier examples. A real derivation would involve more steps than that, naturally.)

Having set things up this way, you then explain island constraints with the hypothesis that you can't move something out of an island. (And you suppose that relative constructions and question constructions that don't observe island constraints don't involve movement.)

Another point worth mentioning is that nothing I've said requires that it be the last stage in the derivation that gets pronounced. If you have a language where, say, question words are in situ but questions are still governed by island constraints, a possible analysis is that derivation of questions goes through stages just like the English example above, but it's the stage before the movement---when the question word is in situ---rather than the stage after that gets pronounced. The result is what's called covert movement. (This is only one G&B analysis of covert movement, though.)

Third, there's whatever you have in minimalism, which is usually still called movement, but really kind of isn't. You still put sentences in stages, but it looks more like this:

        are you eating what
→ whatᵢ are you eating whatᵢ

This is what Chomsky calls internal merge: something that's already been put into the sentence is inserted again, so that it occupies two positions in the sentence's structure. Some people---not Chomsky---say that it's a copy of the original that's put in the new position, not the original itself, though I don't think I've ever seen someone try to explain what copying amounts to here; and quite a lot of the time it's not obvious how you're supposed to understand what's going on here; but for Chomsky himself it's just the very same thing ending up in two positions in the structure. On no one's view does it really make sense to describe this as movement. Still, your account of island effects is a lot like the G&B one: something that's inside an island can't be re-merged into a postion outside the island.

Of course you don't say "what are you eating what"---you only pronounce "what" in one of its two positions. (Alternatively: you only pronounce one of the two copies.) In English, you pronounce it in the sentence-initial position; in an alternate in situ English, you'd pronounce it in the other position.

(This view makes it conceivable that you could have a construction where some constituent gets pronounced more than once, at more than one position in the structure, and analyses along such lines have been made of constructions in a number of languages, including certain verb-copying constructions in Mandarin, for example.)

All that said, what about syntactic ergativity? Well, it's an intriguing fact that syntactic ergativity usually, and maybe always, constitutes a constraint on exactly the sorts of constructions where you find island constraints. This is why Chomskyans tend to describe it as a constraint on movement (A-bar movement, or extraction, in particular). But, first, I think the issue doesn't require you to believe in syntactic movement. Even evmdbm's question about English subject questions isn't necessarily about movement---you can still ask whether English subject questions require a link to be established between the question word and a gap, just like English object questions seem to. And, second, "movement" is actually a pretty bad description of what contemporary Chomskyans mostly think is going on, the fact that mostly they do and other syntacticians maybe mostly don't may bemore a cultural difference than anything at this point.

[Kuchigakatai]

Regarding my question many pages ago about the innovations of Chomskyan syntax, here's a blog post written by Martin Haspelmath about this topic. It may be of particular interest because, as he himself notes, he is a regular critic (I'd add perhaps even a rather harsh one) of Chomskyan syntax (example 1, example 2).

[circeus]

I've been re-readig my copy of the Syntax construction kit, and I'll be the first to admit that syntactic thinking doesn't come naturally to me, but one thing stuck out like a sore thumb.

In the bestiary, there's a short bit about object deletion (p. 302) and the argument is basically "it's hard to explain ambitransitive verbs that are also unergative because it's an object deletion" with the implication that deleting the object of a transitive verb should automatically make the sentence ungrammatical. I do not dispute that conclusion, but I do dispute its premise...

The problem I see with it is that all examples listed are essentially cases of cognate objects (briefly discussed earlier on p. 228), which are more or less semantically empty. It's not a case of fundamentally transitive verbs losing objects: it's a case of fundamentally intransitive verbs that have been given dummy objects. If we construct parallel sentences where the object is not cognate, they cannot easily be deleted anymore (Such sentences are very difficult to construct for bite, probably because there is no sense for that verb where the object is not somewhat cognate):

I am smoking cigarettes → I am smoking
I am smoking salmon → *I am smoking

I am riding a horse → I am riding
I am riding a motorcycle → *I am riding

The goat ate my report → The goat ate
The photocopier ate my report → *The photocopier ate [maybe no ungrammatical in the strictest sense, still highly infelicitous at best]

[bradrn]

Um, I’m not sure that ‘cognate object’ means what you think it means. Wikipedia defines it as ‘a verb's object that is etymologically related to the verb’, and that’s always been what I thought it means as well. Examples (again from Wikipedia): ‘He slept a troubled sleep’; ‘He laughed a bitter laugh’. They’re interesting because it’s hard to determine their transitivity: in those examples, ‘slept’ and ‘laughed’ are normally intransitive, but here they’re transitive.

So the examples here…

The problem I see with it is that all examples listed are essentially cases of cognate objects (briefly discussed earlier on p. 228), which are more or less semantically empty. It's not a case of fundamentally transitive verbs losing objects: it's a case of fundamentally intransitive verbs that have been given dummy objects. If we construct parallel sentences where the object is not cognate, they cannot easily be deleted anymore (Such sentences are very difficult to construct for bite, probably because there is no sense for that verb where the object is not somewhat cognate):

I am smoking cigarettes → I am smoking
I am smoking salmon → *I am smoking

I am riding a horse → I am riding
I am riding a motorcycle → *I am riding

The goat ate my report → The goat ate
The photocopier ate my report → *The photocopier ate [maybe no ungrammatical in the strictest sense, still highly infelicitous at best]

…are not cognate objects. In particular, they are not ‘more or less semantically empty … dummy objects’ at all, as you assert. For instance, ‘cigarettes’ in ‘I am smoking cigarettes’ is not semantically empty at all: compare ‘I am smoking cigarettes’, ‘I am smoking cigars’, ‘I am smoking a pipe’. (Or, if you happen to be preparing smoked salmon, ‘I am smoking salmon’, though I appreciate that that’s a different sense of the verb.) What you have observed here is that you can only delete the object when it is semantically expected for that verb — which of course is true. (And, given the appropriate context, all of your supposedly ‘ungrammatical’ sentences can become more grammatical — though still dispreferred — if you make the intended argument expected enough: ‘I haven’t handed in my report because the photocopier ate it. The photocopier ate and ate and eventually spat out some mangled shreds.’ Or, ‘I always ride my motorcycle to work. Today I saw a car accident while I was riding.’)

[circeus]

…are not cognate objects. In particular, they are not ‘more or less semantically empty … dummy objects’ at all, as you assert. For instance, ‘cigarettes’ in ‘I am smoking cigarettes’ is not semantically empty at all: compare ‘I am smoking cigarettes’, ‘I am smoking cigars’, ‘I am smoking a pipe’. (Or, if you happen to be preparing smoked salmon, ‘I am smoking salmon’, though I appreciate that that’s a different sense of the verb.) What you have observed here is that you can only delete the object when it is semantically expected for that verb — which of course is true.

Calling to cognate object may have been a mistake. It's easy to dismiss with examples like "dream a dream", it,s harder to dismiss when the examples include things like "heave a sigh" or Zompist's "take a look" and "take a walk" (there's a reason I specifically mention that it's his own examples that draw me to contest the so-called oddity!).

My meaning is really that here the transitives should be read as derived from the intransitive, not the other way around, hence deletion of the object is not anomalous in any fashion: it's really just a return to a base state.

(And, given the appropriate context, all of your supposedly ‘ungrammatical’ sentences can become more grammatical — though still dispreferred — if you make the intended argument expected enough: ‘I haven’t handed in my report because the photocopier ate it. The photocopier ate and ate and eventually spat out some mangled shreds.’ Or, ‘I always ride my motorcycle to work. Today I saw a car accident while I was riding.’)

You can validate a LOT of otherwise ungrammatical statements if you expand beyond the base sentence. I always understood that a major point in Chomskyan syntax (or at least a huge point of disagreement in analysis techniques between it and other approaches) is precisely that you're not supposed to do that.

Worth pointing out also that "motorcycle" is just the more polite example. It's significantly harder to explain away the distinction when you're riding a dildo!

[bradrn]

…are not cognate objects. In particular, they are not ‘more or less semantically empty … dummy objects’ at all, as you assert. For instance, ‘cigarettes’ in ‘I am smoking cigarettes’ is not semantically empty at all: compare ‘I am smoking cigarettes’, ‘I am smoking cigars’, ‘I am smoking a pipe’. (Or, if you happen to be preparing smoked salmon, ‘I am smoking salmon’, though I appreciate that that’s a different sense of the verb.) What you have observed here is that you can only delete the object when it is semantically expected for that verb — which of course is true.

Calling to cognate object may have been a mistake. It's easy to dismiss with examples like "dream a dream", it,s harder to dismiss when the examples include things like "heave a sigh" or Zompist's "take a look" and "take a walk" (there's a reason I specifically mention that it's his own examples that draw me to contest the so-called oddity!).

I would call those light verb constructions, rather than cognate object constructions: almost all of the meaning is concentrated in the nominal argument, leaving the ‘light verb’ nearly semantically empty. (As it happens, I’ve been reading heavily about such constructions for the past week or so; I find them really interesting!)

My meaning is really that here the transitives should be read as derived from the intransitive, not the other way around, hence deletion of the object is not anomalous in any fashion: it's really just a return to a base state.

Interesting interpretation — but is there any difference really between saying that they are originally intransitive verbs which have been made transitive vs originally transitive verbs which are made intransitive?

(And, given the appropriate context, all of your supposedly ‘ungrammatical’ sentences can become more grammatical — though still dispreferred — if you make the intended argument expected enough: ‘I haven’t handed in my report because the photocopier ate it. The photocopier ate and ate and eventually spat out some mangled shreds.’ Or, ‘I always ride my motorcycle to work. Today I saw a car accident while I was riding.’)

You can validate a LOT of otherwise ungrammatical statements if you expand beyond the base sentence. I always understood that a major point in Chomskyan syntax (or at least a huge point of disagreement in analysis techniques between it and other approaches) is precisely that you're not supposed to do that.

I always thought that a sentence is grammatical if it’s grammatical in any context — otherwise you eliminate of a lot of otherwise fine sentences. (One example that comes to mind is sentences such as You I like; such sentences are very common in spoken and written English, and are perfectly fine given the appropriate context, but by themselves they seem decidedly ungrammatical.)

[circeus]

My meaning is really that here the transitives should be read as derived from the intransitive, not the other way around, hence deletion of the object is not anomalous in any fashion: it's really just a return to a base state.

Interesting interpretation — but is there any difference really between saying that they are originally intransitive verbs which have been made transitive vs originally transitive verbs which are made intransitive?

Well, yes. The original argument (as I understand the entry in the SCK), is that object deletion is strange, because there is no transformational explanation for the grammaticalness (grammaticity?) of the resulting sentence. If we take it instead the position that the underlying verb was really intransitive all along, then this so-called "object deletion" is not just entirely unremarkable, it's a backward analysis of what I believe is happening, namely an object insertion.

In diachronic terms, "object deletion" is the logical (inevitable, even) consequence of a previously intransitive verb gaining a transitive construction.

[zompist]

I am smoking cigarettes → I am smoking
I am smoking salmon → *I am smoking

I am riding a horse → I am riding
I am riding a motorcycle → *I am riding

The goat ate my report → The goat ate
The photocopier ate my report → *The photocopier ate [maybe no ungrammatical in the strictest sense, still highly infelicitous at best]

I think you've found something really interesting, though I'm not sure I agree with your interpretation.

I don't think these are cognate objects, simply because they're not at all semantically vacuous. "Take a ride" is vacuous; it just means "ride". "Ride a horse" is not vacuous; there are many things you can ride, a horse is just one of them. What you're pointing out is that there's a kind of automatic fill-in of the most likely object. Maybe pragmatic, but it's interesting that the default object is "horse" when people are far more likely to ride a bicycle or motorcycle or train these days. In support of that, I'd note that you can change the default object with the right context: "I just bought a bicycle. I'm going to go out riding!"

You also suggest that they're ambitransitive. Wikipedia calls these agentive ambitransitives: e.g. "Mary is knitting a sweater" vs. "Mary is knitting". OK, cool, except that its examples (eat, follow, help, knit, read, try, watch, win, know) don't show the discrepancy you've pointed out. That is, there is no default object for "knitting" that can be omitted, while other objects of "knitting" must be specified.

(Thinking more: maybe there is, in the case of metaphorical extensions. "Mary is knitting her brow" cannot be expressed "Mary is knitting." But that leaves your examples unexplained. Plus there are reasons that idioms break many transformations.)

Now, as I said, one syntacticians's Delete is another one's Add. I'm certainly not going to die on the hill of insisting that there is an underlying object somewhere that is deleted. I think the default object thing is quite interesting though. Does French work like this? I think "fumer" does; I don't think "monter" does?

[bradrn]

I don't think these are cognate objects, simply because they're not at all semantically vacuous. "Take a ride" is vacuous; it just means "ride". "Ride a horse" is not vacuous; there are many things you can ride, a horse is just one of them.

Isn’t it the other way around with “take a ride”? There, the argument ‘ride’ is certainly meaningful; it’s the verb ‘take’ that’s vacuous. I think a better example would be something like ‘calculate a number’, where the argument carries pretty much none of the meaning.

What you're pointing out is that there's a kind of automatic fill-in of the most likely object. Maybe pragmatic, but it's interesting that the default object is "horse" when people are far more likely to ride a bicycle or motorcycle or train these days. In support of that, I'd note that you can change the default object with the right context: "I just bought a bicycle. I'm going to go out riding!"

You also suggest that they're ambitransitive. Wikipedia calls these agentive ambitransitives: e.g. "Mary is knitting a sweater" vs. "Mary is knitting". OK, cool, except that its examples (eat, follow, help, knit, read, try, watch, win, know) don't show the discrepancy you've pointed out. That is, there is no default object for "knitting" that can be omitted, while other objects of "knitting" must be specified.

…

Now, as I said, one syntacticians's Delete is another one's Add. I'm certainly not going to die on the hill of insisting that there is an underlying object somewhere that is deleted. I think the default object thing is quite interesting though. Does French work like this? I think "fumer" does; I don't think "monter" does?

Possibly there are two classes of object-deleting verbs, one which has implied objects and another which doesn’t? e.g. ‘Ride’, ‘smoke’ have implied objects, whereas ‘eat’, ‘knit’ don’t. Potentially we could go further and say that the former class have intransitive forms derived from transitives, while the latter have transitive forms derived from intransitives.

[circeus]

You also suggest that they're ambitransitive. Wikipedia calls these agentive ambitransitives: e.g. "Mary is knitting a sweater" vs. "Mary is knitting". OK, cool, except that its examples (eat, follow, help, knit, read, try, watch, win, know) don't show the discrepancy you've pointed out. That is, there is no default object for "knitting" that can be omitted, while other objects of "knitting" must be specified.

(Thinking more: maybe there is, in the case of metaphorical extensions. "Mary is knitting her brow" cannot be expressed "Mary is knitting." But that leaves your examples unexplained. Plus there are reasons that idioms break many transformations.)

I read knitting as having a defined object. It's not a specific one like "ride" (indeed ride is the only one of these with a specific word/concept as the unstated object), but it is well circumscribe the same way that of eat ("something edible") or Smoke ("Tobacco or an equivalent substance in one form or another") is.

Now, as I said, one syntacticians's Delete is another one's Add. I'm certainly not going to die on the hill of insisting that there is an underlying object somewhere that is deleted. I think the default object thing is quite interesting though. Does French work like this? I think "fumer" does; I don't think "monter" does?

They both do. Manger, tricoter, regarder, lire and gagner also have an intransitive form that matches exactly a transitive in the same way the English does. Aider, savoir, suivre, essayer and connaître, on the other hand, require that the implied object be established pragmatically in some way first, and not all meanings allow it (you cannot delete an animate object for connaître). I'm not entirely sure some of the English "agentives" are a perfect fit for the first category. "I read." and "I knit." are entirely innocuous statement in isolation, but we instinctively feel the lack of information with an isolated "?I help" or "?I followed".

I'm starting to think the ability to not mention an object that is well established pragmatically may be a separate characteristic of English. Probably a side effect of how easily English words jump between lexical classes (or subclasses): I don't think this (and especially not shifting from an intransitive to a transitive or vice-versa) is nearly as common in German. Whether it must be account by syntactic transformation, bringing us back where we started, I do not know: as I said, I am not super good at theoretical syntactic thinking XD It's probably one of those cases that construction syntax handles more easily than Chomskyan approaches, though.

[KathTheDragon]

but we instinctively feel the lack of information with an isolated "?I help" or "?I followed".

Do we? "I'm helping" and "I'm reading" are perfectly fine. I think the awkwardness in the simple present is rather to do with aspect.

[bradrn]

A most interesting sentence I thought up: ‘Reading to myself is nice’. Clearly this is a grammatical sentence (at least it is for me) — yet it has a reflexive without any antecedent. What is happening here?

[zompist]

A most interesting sentence I thought up: ‘Reading to myself is nice’. Clearly this is a grammatical sentence (at least it is for me) — yet it has a reflexive without any antecedent. What is happening here?

"Reading" has an obvious understood subject, "me". So I'd say the underlying S is "[me reading to me] is nice". Reflexivization applies first, then the subject is deleted.

As confirmation, note that overt subjects can occur in this construction: "Johnny reading to himself is so cute."

[bradrn]

A most interesting sentence I thought up: ‘Reading to myself is nice’. Clearly this is a grammatical sentence (at least it is for me) — yet it has a reflexive without any antecedent. What is happening here?

"Reading" has an obvious understood subject, "me". So I'd say the underlying S is "[me reading to me] is nice". Reflexivization applies first, then the subject is deleted.

As confirmation, note that overt subjects can occur in this construction: "Johnny reading to himself is so cute."

Thank you! I had assumed that there was no underlying subject due to the ungrammaticality of *‘Me reading to myself is nice’; I didn’t realise that other subjects would be fine, so thanks for pointing that out.

A further question: under what conditions can the subject be deleted?

[Richard W]

I had assumed that there was no underlying subject due to the ungrammaticality of *‘Me reading to myself is nice’

And what's wrong with saying, "My reading to myself is nice."?

[bradrn]

I had assumed that there was no underlying subject due to the ungrammaticality of *‘Me reading to myself is nice’

And what's wrong with saying, "My reading to myself is nice."?

I dunno… it just sounds really weird. It’s the sort of sentence I would prefix with ?? if I were using it as an example.

[zompist]

A further question: under what conditions can the subject be deleted?

I think it can always be deleted:

Reading to yourself is a good habit.
Reading to himself is how Felix gets to sleep.
She's always in the back row reading to herself.