Syntax random

[bradrn]

Possibly it might help to clarify how I define the concepts I’ve been using. Fundamentally, my suggestion is there are three different ways to define ‘word’ (rather than the two which are usually suggested), with their definitions as follows:

• Phonological words have their own stress and minimal word length, and are the domain of (some) phonological rules.
• I have already quoted Julien’s definition of grammatical words.
• A lexeme is a conventionalised expression for a particular lexical meaning; that is, they are expressions which are used to express a particular meaning more commonly than would otherwise be expected. (This is the extra term I want to introduce; as far as I can tell, it has not been rigorously defined before, though I have seen people refer to the need for it, e.g. in Pawley’s writings on Kalam.)
  • Additionally, some lexemes are non-compositional as well as conventionalised; we could distinguish an extra class of non-compositional lexemes, though I’m not yet convinced of its utility. (On the other hand, what is useful is knowing that anything which is non-compositional must be a lexeme.)

Notably, all of these concepts are independent of each other (with the notable exception of non-compositionability, which has conventionality as a pre-requisite). Of course, many things are instances of all three, e.g. ‘dog’. But it is easy to find more involved examples:

• English has no examples of phonological words which are not grammatical words or lexemes, but they’re common in other languages: for instance, East Ambae has many verbal particles (e.g. realis mo, recent bei) which are independent phonological words, but have a rigid order forming a grammatical word with the verb.
• Grammatical words which are not phonological words or lexemes are of course clitics such as =’s.
• I know of no unambiguous examples of lexemes which are not grammatical or phonological words, but one possibility might be the ‘lexical affixes’ of some polysynthetic languages, e.g. Nuu-chah-nulth -y̓iiḥa ‘die from’, -stiił ‘at the collar bone’ (source: Polysynthesis for Novices), or Saanich ∥=əqsən∥ ‘nose, point’, ∥=aŋ̓əʔ∥ ‘berry’ (source). Another possibility might be some incorporated nouns.
• A similar possibility is lexemes which are not grammatical words, but make up more than one phonological word. This is the category we have been talking about: mountain climb, Panama papers, kick the bucket.
• Phonological and grammatical words which are not lexemes are unbound morphemes. English doesn’t have too many, but it’s easy to find examples, e.g. the negative particle in Bagirmi (source).
• Lexemes which make up (more than one) phonological word but are not grammatical words are rare, but attested; this category includes English phrasal verbs such as eat sth up, as well as other types of discontinuous lexeme.
• It is rare to find words which are both lexemes and grammatical words but not phonological words, but the possibility is attested, e.g. the Kalam verb d= ‘get’, which cliticises to the following word in the sentence.

"Panama papers" is a lexeme. That's not a word, unless you want to maintain that all titles ("If on a winter's night a traveler", "The Cambridge Encyclopedia of Linguistics", "General Secretary of the Chinese Communist Party") and idioms ("kick the bucket") are words. They're fixed because that's how lexemes work.

I stand by my comment above: those titles are not words phonologically, but at the same time they are both grammatical words and lexemes: ‘grammatical’ because they resist being split by other words, and ‘lexemes’ because they are conventionalised non-compositional phrases.

(I’m not so sure what you see as so unusual about this… I’m just applying the usual definition of those terms!)

Well, I feel like I am too! To me you're conflating "word" and "NP" and conflating syntax and morphology.

No, I’m not. Most NPs aren’t words: ‘that big bus which I saw yesterday’ is not a word by any of the definitions above. Similarly, most words aren’t NPs, e.g. all English verbs.

On the other hand, I can’t argue with ‘conflating syntax and morphology’. Although they are of course separate in many cases, I would say the boundary between them is fuzzy rather than sharp; many of those more involved cases I listed above are cases where the boundary is ambiguous.

You give two reasons above, so let me address those.

One, that they're "conventionalized". Of course they are, but I deny that conventionalization makes things a word. Otherwise the Lord's Prayer, the Rosary, or Psalm 23, or Hamlet's "to be or not to be" soliloquy, are "words".

This is why it’s useful to have a more precise definition, such as the one I gave above: ‘a conventionalised expression for a particular lexical meaning’. Of course those are all conventionalised, but none of them have ‘a particular lexical meaning’. If ‘to be or not to be’ starts coming into general conversational use to express a particular meaning, then yes, it will have become a lexeme.

Two, that they resist insertions and other operations. I agree with that, but that's because they're lexicalized. Not every term can be a single word, so we have conventions to choose certain NPs and make them the complex term for something. You can't just change the accepted term for "Panama Papers" without confusing everyone. But in formation, "Panama Papers" is a completely ordinary NP following NP rules.

But that doesn’t apply to all lexemes! I already gave the example of eat sth up: a lexeme which allows arbitrary material to be inserted in the middle.

Maybe a better way of looking at this is not to just say "words can't be separated", but to ask, what can('t) you insert in various places, and why. There are rules of various kinds and absoluteness. The reason you can't change "Panama Papers" is not anything structural about the phrase, but because then people wouldn't recognize the reference. It's not the same sort of reason you can't say "my the papers" or "Australareallylovelycontinentia."

Yes, I am very sympathetic to this point of view: of course it’s a continuum, practically everything in linguistics is a continuum! Some expressions allow nothing to intervene (‘Panama papers’, ‘Australia’), and hence are grammatical words; others allow arbitrary things to intervene (‘people like dogs’), and thus are not; still others have restrictions of varying sorts on the intervening material (‘my papers’). But I don’t see how that is relevant to my argument, apart from emphasising that ‘grammatical word’ and similar concepts are fuzzy rather than boolean categories.

[zompist]

[*] A lexeme is a conventionalised expression for a particular lexical meaning; that is, they are expressions which are used to express a particular meaning more commonly than would otherwise be expected. (This is the extra term I want to introduce; as far as I can tell, it has not been rigorously defined before, though I have seen people refer to the need for it, e.g. in Pawley’s writings on Kalam.)

• Additionally, some lexemes are non-compositional as well as conventionalised; we could distinguish an extra class of non-compositional lexemes, though I’m not yet convinced of its utility. (On the other hand, what is useful is knowing that anything which is non-compositional must be a lexeme.)

You do realize that I've been using "lexeme" previously in the discussion? And that the term has been used in linguistics for 80 years? (It dates back at least to a 1940 article by Benjamin Lee Whorf.)

[*] A similar possibility is lexemes which are not grammatical words, but make up more than one phonological word. This is the category we have been talking about: mountain climb, Panama papers, kick the bucket.

Yes, we agree they're lexemes. You seem to agree with me here that they are not what you call "grammatical words", but in other places you claim they are.

Though we may be at an impasse, I'll try one more time. The disagreement is over what it means if you can't freely insert other words. For you this seems to automatically make things a "grammatical word". I think you're failing to recognize different reasons for why you can't insert material.

My view is that both syntax and morphology (and for that matter phonology!) have rules on how you can combine elements, and how you can't. The morphological level is highly restricted. E.g., the Quechua frame ruma__nki is extremely limited. Cf rumanki 'you speak', rumachkanki 'you spoke', rumawanki 'you speak to me'... this list isn't complete but there's not many more possibilities, and often you can list them all in a table. The syntactic level often allows a huge variety of insertions, but restricts their possible kinds. E.g. "the ___ house" can be filled by an indefinite number of things, but not by a (single) S or Adv node.

Now, the frame "to be ___ to be" can be filled by a million things too... but if you're quoting Shakespeare it can only accept "or", because that's the text he wrote. That's not a syntactic reason at all; it's a matter of citation etiquette.

Syntactically, "Panama ___ papers" is fairly open-- you could insert "rolling", "and Nicaragua", "financial", "City", "Canal", "um", etc., but not "if", "or not to be", "very", "she", etc., etc.

Finally, "the Panama ___ Papers", as a recent political term, don't admit much at all ("um" still fits), because that's their name. This is not a syntactic reason at all, it's more of a journalistic convention. It is just not the same sort of restriction as the syntactic restrictions I've given above. And though I haven't read Julien, I'd be rather surprised if she disagreed. (That is, I expect she's talking about syntactic restrictions too.)

Another example: the frame "Edgar __ Poe" is just as limited, if you're referring to the writer. This isn't a syntactic fact: someone else might be named Edgar Tecumseh Poe. But we call the writer Edgar Allen Poe because that's his name.

Because nothing is every cut-and-dried in language, I should also note idioms (one type of lexeme), vary from extremely limited to very free. Often they resist common transformations, to say nothing of substituted words. Others are remarkably free, such as the "way" construction I cover in my book.

[bradrn]

[*] A lexeme is a conventionalised expression for a particular lexical meaning; that is, they are expressions which are used to express a particular meaning more commonly than would otherwise be expected. (This is the extra term I want to introduce; as far as I can tell, it has not been rigorously defined before, though I have seen people refer to the need for it, e.g. in Pawley’s writings on Kalam.)

• Additionally, some lexemes are non-compositional as well as conventionalised; we could distinguish an extra class of non-compositional lexemes, though I’m not yet convinced of its utility. (On the other hand, what is useful is knowing that anything which is non-compositional must be a lexeme.)

You do realize that I've been using "lexeme" previously in the discussion? And that the term has been used in linguistics for 80 years? (It dates back at least to a 1940 article by Benjamin Lee Whorf.)

Yes, I do realise that — I’ve also used ‘lexeme’ previously (in the context of parsing algorithms). I just thought it might be useful to have a more precise definition for this discussion. (Besides, I’m not sure I’m using the term the same way it’s usually used… possibly it might have been better to avoid pre-existing terms entirely and call it a ‘lexical word’ or something novel.)

Though we may be at an impasse, I'll try one more time. The disagreement is over what it means if you can't freely insert other words. For you this seems to automatically make things a "grammatical word". I think you're failing to recognize different reasons for why you can't insert material.

Yes, that is exactly my position. (For now, that is; I am of course open to being convinced otherwise.)

My view is that both syntax and morphology (and for that matter phonology!) have rules on how you can combine elements, and how you can't. The morphological level is highly restricted. E.g., the Quechua frame ruma__nki is extremely limited. Cf rumanki 'you speak', rumachkanki 'you spoke', rumawanki 'you speak to me'... this list isn't complete but there's not many more possibilities, and often you can list them all in a table. The syntactic level often allows a huge variety of insertions, but restricts their possible kinds. E.g. "the ___ house" can be filled by an indefinite number of things, but not by a (single) S or Adv node.

Now, the frame "to be ___ to be" can be filled by a million things too... but if you're quoting Shakespeare it can only accept "or", because that's the text he wrote. That's not a syntactic reason at all; it's a matter of citation etiquette.

Syntactically, "Panama ___ papers" is fairly open-- you could insert "rolling", "and Nicaragua", "financial", "City", "Canal", "um", etc., but not "if", "or not to be", "very", "she", etc., etc.

Finally, "the Panama ___ Papers", as a recent political term, don't admit much at all ("um" still fits), because that's their name. This is not a syntactic reason at all, it's more of a journalistic convention. It is just not the same sort of restriction as the syntactic restrictions I've given above. And though I haven't read Julien, I'd be rather surprised if she disagreed. (That is, I expect she's talking about syntactic restrictions too.)

Another example: the frame "Edgar __ Poe" is just as limited, if you're referring to the writer. This isn't a syntactic fact: someone else might be named Edgar Tecumseh Poe. But we call the writer Edgar Allen Poe because that's his name.

Because nothing is every cut-and-dried in language, I should also note idioms (one type of lexeme), vary from extremely limited to very free. Often they resist common transformations, to say nothing of substituted words. Others are remarkably free, such as the "way" construction I cover in my book.

So, if I’m understanding what you’re saying correctly: you’re saying that words and phrases vary dramatically in what may be inserted in them. Some are totally free, admitting basically anything. Others are more limited, allowing few or no insertions. But those don’t make up a particularly coherent category; in fact, from what I can see, we can distinguish at least two major types. Some allow limited insertions for morphological reasons, allowing only a closed set of bound morphemes to be inserted, if anything (e.g. ‘ruma__nki’, ‘Australia’). Others allow limited insertions for semantic reasons: the phrase is still grammatical when arbitrary material is inserted, but it dramatically changes in meaning (e.g. ‘Panama Papers’, ‘to be ___ to be’).

And this is something I can totally agree with! I agree that the set of ‘grammatical words’, as I’ve been calling them, is rather heterogeneous. But on the other hand, this is something I already took account of in my typology! In particular, if you look back at it, you’ll see I briefly mentioned a category of ‘non-compositional lexemes’:

• Additionally, some lexemes are non-compositional as well as conventionalised; we could distinguish an extra class of non-compositional lexemes, though I’m not yet convinced of its utility. (On the other hand, what is useful is knowing that anything which is non-compositional must be a lexeme.)

What you have demonstrated is that I was wrong in the quoted definition: the class of ‘non-compositional lexemes’ is actually one which is extremely useful! In fact, I’d say that the distinction between those two classes of ‘grammatical words’ is precisely that between compositional and non-compositional grammatical words. In more detail:

• Non-compositional grammatical words are more than the sum of their parts; they resist insertions precisely because that would change their meaning.
• Compositional grammatical words are formed generatively; thus, if they resist insertions, it must be because they are forbidden by syntax.

We might even draw up a little table, with examples to demonstrate:

	+grammatical word	-grammatical word
compositional lexeme	‘mountain climbing’, ‘ruma__nki’ (insertions forbidden due to syntax/morphology)	‘put ___ out’ (insertions allowed due to syntax)
non-compositional lexeme	‘Panama Papers’, ‘kick the bucket’, ‘Edgar Poe’ (insertions forbidden due to semantic shift)	‘put ___ off’ (insertions allowed due to… syntax, I think, but not sure)

So I am definitely in full agreement with you that my ‘grammatical word’ class is fairly diverse — but that doesn’t mean it’s invalid, it just means we have to figure out what’s causing the diversity, which is something my typology seems to do very well.

[zompist]

	+grammatical word	-grammatical word
compositional lexeme	‘mountain climbing’, ‘ruma__nki’ (insertions forbidden due to syntax/morphology)	‘put ___ out’ (insertions allowed due to syntax)
non-compositional lexeme	‘Panama Papers’, ‘kick the bucket’, ‘Edgar Poe’ (insertions forbidden due to semantic shift)	‘put ___ off’ (insertions allowed due to… syntax, I think, but not sure)

I'm not clear on what makes the right-hand cells "not grammatical words". Because they're not constituents, or not contiguous? And I'm not clear on what distinction you're making between "put off" and "put out". ("Put off your vacation" is an idiom, but so is "He's feeling put out.")

Also, I don't think you've addressed the freer type of idioms. "Kick the bucket" is not quite fixed:

He kicked the proverbial bucket.
Johnson is about the kick the old bucket.

Also compare sentences like "You're about to open a very messy can of very crawly worms." Or "Politicians prefer to kick the can, as long as they are able to, down the road."

[bradrn]

	+grammatical word	-grammatical word
compositional lexeme	‘mountain climbing’, ‘ruma__nki’ (insertions forbidden due to syntax/morphology)	‘put ___ out’ (insertions allowed due to syntax)
non-compositional lexeme	‘Panama Papers’, ‘kick the bucket’, ‘Edgar Poe’ (insertions forbidden due to semantic shift)	‘put ___ off’ (insertions allowed due to… syntax, I think, but not sure)

I'm not clear on what makes the right-hand cells "not grammatical words". Because they're not constituents, or not contiguous?

Because they allow arbitrary insertions: ‘put [the big cat which I took home yesterday] out’. (Admittedly, that’s close to ungrammatical, but it’s still just about fine; it depends on the length of the insertion.)

And I'm not clear on what distinction you're making between "put off" and "put out". ("Put off your vacation" is an idiom, but so is "He's feeling put out.")

You put out the cat or the bins; you put off work. The first is directional, the second isn’t. (And yes, put out can be either compositional or non-compositional depending on how you use it, though I believe it becomes resistant to insertions when it’s used non-compositionally.)

Also, I don't think you've addressed the freer type of idioms. "Kick the bucket" is not quite fixed:

He kicked the proverbial bucket.
Johnson is about the kick the old bucket.

Also compare sentences like "You're about to open a very messy can of very crawly worms." Or "Politicians prefer to kick the can, as long as they are able to, down the road."

Yes, so those are intermediate types: somewhere between compositional and non-compositional, somewhere between +grammatical and -grammatical. As usual, these are all continua rather than binary.

[Richard W]

I'm not clear on what makes the right-hand cells "not grammatical words". Because they're not constituents, or not contiguous?

Because they allow arbitrary insertions: ‘put [the big cat which I took home yesterday] out’. (Admittedly, that’s close to ungrammatical, but it’s still just about fine; it depends on the length of the insertion.)

Are you sure you don't have an arbitrary definition of arbitrary? The problem is tmesis - for example, one can insert many expletives in im-possible. Or is impossible not a grammatical word?

And I'm not clear on what distinction you're making between "put off" and "put out". ("Put off your vacation" is an idiom, but so is "He's feeling put out.")

You put out the cat or the bins; you put off work. The first is directional, the second isn’t. (And yes, put out can be either compositional or non-compositional depending on how you use it, though I believe it becomes resistant to insertions when it’s used non-compositionally.)

So how do you categorise Put the bushfire out? And for me, it is far more natural to say You put work off.

[zompist]

Because they allow arbitrary insertions: ‘put [the big cat which I took home yesterday] out’. (Admittedly, that’s close to ungrammatical, but it’s still just about fine; it depends on the length of the insertion.)

Are you sure you don't have an arbitrary definition of arbitrary? The problem is tmesis - for example, one can insert many expletives in im-possible. Or is impossible not a grammatical word?

The Wikipedia article on tmesis is worth it just for this example:

In the work of the poet Ennius, the literal splitting of the word cerebrum creates a vivid image: saxo cere comminuit brum "he shattered his brain with a rock.

Which is way more cool than the usual examples like "abso-fuckin-lutely."

Still, tmesis is surely a kind of self-conscious playing with language and needn't upset anyone's theories (unless those theories don't allow playing with language).

[bradrn]

I'm not clear on what makes the right-hand cells "not grammatical words". Because they're not constituents, or not contiguous?

Because they allow arbitrary insertions: ‘put [the big cat which I took home yesterday] out’. (Admittedly, that’s close to ungrammatical, but it’s still just about fine; it depends on the length of the insertion.)

Are you sure you don't have an arbitrary definition of arbitrary? The problem is tmesis - for example, one can insert many expletives in im-possible. Or is impossible not a grammatical word?

Tmesis poses an interesting problem, since it can apply to so many words. I’d say the best solution to that is to adopt zompist’s approach and realise that ‘insertionability’ is a continuum rather than a boolean, and ‘grammatical word-ness’ must be the same. For instance, ‘achieve’ resists tmesis, thus is obviously a grammatical word; ‘absolute’ less so, since it allows tmesis; ‘Edgar Poe’ and ‘Panama Papers’ even less so, since it allows insertions but only with a dramatic change in meaning (i.e. most insertions change the lexeme involved); and ‘Panama papers’ is not a grammatical word at all, allowing a great many insertions.

(The other definitions of ‘word’ I provided are similar, actually. Lexeme-ness is dependent on exactly how often a particular collocation is used; ‘Panama Papers’ and ‘mountain climbing’ definitely are lexemes, while e.g. ‘puppy dog’ is less obvious. Similarly the problems with defining phonological words are well known.)

And I'm not clear on what distinction you're making between "put off" and "put out". ("Put off your vacation" is an idiom, but so is "He's feeling put out.")

You put out the cat or the bins; you put off work. The first is directional, the second isn’t. (And yes, put out can be either compositional or non-compositional depending on how you use it, though I believe it becomes resistant to insertions when it’s used non-compositionally.)

So how do you categorise Put the bushfire out? And for me, it is far more natural to say You put work off.

I already covered those in my table, in the last column:

	+grammatical word	-grammatical word
compositional lexeme	‘mountain climbing’, ‘ruma__nki’ (insertions forbidden due to syntax/morphology)	‘put ___ out’ (insertions allowed due to syntax)
non-compositional lexeme	‘Panama Papers’, ‘kick the bucket’, ‘Edgar Poe’ (insertions forbidden due to semantic shift)	‘put ___ off’ (insertions allowed due to… syntax, I think, but not sure)

(Also, I find your naturalness judgements interesting… for me, put off work is more natural, though both are acceptable.)

[bradrn]

I just found myself sending an email starting as follows:

In case anyone didn’t get my previous email about this (which it sounds like some people didn’t) …

But when I thought about it for a moment, I realised this is a really weird thing to say, from a syntactic point of view. Specifically: what exactly is the relative clause modifying here? It looks like the deleted phrase should be ‘it sounds like some people didn’t [get my previous email about this]’ — except that’s a VP rather than an NP. Is it even possible for a relative clause to modify a VP? I’d appreciate an analysis of this construction.

[KathTheDragon]

My reading is that it's modifying "my previous email about this", and "get" has been dropped from the relative clause due to redundancy.

[zompist]

In case anyone didn’t get my previous email about this (which it sounds like some people didn’t) …

But when I thought about it for a moment, I realised this is a really weird thing to say, from a syntactic point of view. Specifically: what exactly is the relative clause modifying here? It looks like the deleted phrase should be ‘it sounds like some people didn’t [get my previous email about this]’ — except that’s a VP rather than an NP. Is it even possible for a relative clause to modify a VP? I’d appreciate an analysis of this construction.

This particular example doesn't seem hard-- as Kath says, the relative clause is attached to the NP "my previous email about this", and indeed follows it as a relative clause should. All VP Deletion has to delete is the word "get".

"Which" can be used to refer to an action (a VP or an S, I'd have to check further): "I fed the gremlins after midnight, which you should never do."

[bradrn]

My reading is that it's modifying "my previous email about this", and "get" has been dropped from the relative clause due to redundancy.

This particular example doesn't seem hard-- as Kath says, the relative clause is attached to the NP "my previous email about this", and indeed follows it as a relative clause should. All VP Deletion has to delete is the word "get".

I didn’t know that English has a rule of VP Deletion outside conjunctions and disjunctions¹; do you have any more info? (Of course, there’s a good chance you already covered it in the SCK and I simply forgot.)

(¹Hmm, is there a generic name for this?)

"Which" can be used to refer to an action (a VP or an S, I'd have to check further): "I fed the gremlins after midnight, which you should never do."

Now this is a much more interesting example than mine, I think. In this case, what sort of constituent exactly is “which you should never do”? It looks like a relative clause, but in that case, what’s it modifying? Relative clauses, as far as I’m aware, are nominal modifiers by definition, but this one looks like it’s modifying the whole clause. In this regard it looks almost like an adverbial, but it doesn’t have the same freedom of movement: *“Which you should never do, I fed the gremlins after midnight.”

[Moose-tache]

Don't take freedom of movement as a test of adverbialness in English. There are numerous secret classes of adverb based on where they can and cannot go.

Surprisingly, I don't care about my injury.
I don't care about my injury, surprisingly.
I surprisingly don't care about my injury.
But
I don't even care about my injury.
I don't care about my injury, even. (grammatical for me, ungrammatical for most people over 50)
*Even, I don't care about my injury. (where "even" is not modifying "I" but the whole clause - frequently heard by L2 speakers nonetheless)
But
I don't care about my injury now.
Now I don't care about my injury.
*I don't now care about my injury.
And many, many more...

As for dropping redundant verbs (outside of "adjunction" ), English does this all the time, especially in the sorts of relative clauses we're dealing with here: "I give her what you don't."

[Rounin Ryuuji]

I don't care about my injury now.
Now I don't care about my injury.
*I don't now care about my injury.
And many, many more...

This one isn't registering as ungrammatical for me, though it could come off a bit old-fashioned ("I don't now care about it, but I might later"), though you might be using "now" as an intensifier rather to mean "in this present moment", in which case it does feel ungrammatical...

[zompist]

I didn’t know that English has a rule of VP Deletion outside conjunctions and disjunctions¹; do you have any more info?
(¹Hmm, is there a generic name for this?)

"Conjunctions"! It's both a category and a supercategory...

Moose gave a nice example of VP Deletion with no conjunction.

"Which" can be used to refer to an action (a VP or an S, I'd have to check further): "I fed the gremlins after midnight, which you should never do."

Now this is a much more interesting example than mine, I think. In this case, what sort of constituent exactly is “which you should never do”? It looks like a relative clause, but in that case, what’s it modifying? Relative clauses, as far as I’m aware, are nominal modifiers by definition, but this one looks like it’s modifying the whole clause. In this regard it looks almost like an adverbial, but it doesn’t have the same freedom of movement: *“Which you should never do, I fed the gremlins after midnight.”

The first question is easy— "which you should never do" is an S— but that's because both relative clauses and the alternatives are S's. What alternatives? Well, it could be a conjunction, similar to "...as you should never do". Or even a preposition, as in "...after you went to bed."

I'm leaning against a preposition. Prepositions can take S's but also NPs ("after midnight"), and "which" can't take an NP. Also, as you note, the construction can't be fronted.

Conjoints can often be fronted too: "As you should never do, I fed..." But this isn't infallible; cf. "*But you should never do so, I fed..."

I don't think we'd be tempted to analyze "as you should never do" as a relative clause.

My intuition is that "which S" kind of feels like a relative clause. Maybe that's why it resists fronting, just as you can't say "*Which the system lost, I'm resending my email." Note that I accept fuzzy categories, which Chomsky would shudder at.

[bradrn]

Don't take freedom of movement as a test of adverbialness in English. There are numerous secret classes of adverb based on where they can and cannot go.

Surprisingly, I don't care about my injury.
I don't care about my injury, surprisingly.
I surprisingly don't care about my injury.
But
I don't even care about my injury.
I don't care about my injury, even. (grammatical for me, ungrammatical for most people over 50)
*Even, I don't care about my injury. (where "even" is not modifying "I" but the whole clause - frequently heard by L2 speakers nonetheless)
But
I don't care about my injury now.
Now I don't care about my injury.
*I don't now care about my injury.
And many, many more...

Adverbs are different from adverbials. I agree that adverbs are extremely heterogeneous, and should be analysed as a conflation of several different word classes. But I was talking specifically about adverbials, that is, phrases like:

I was going home when I saw it
I paid because I have to
I’ll stop if you make me

Some of these have single-word equivalents, but I’m not interested in those cases.

"Which" can be used to refer to an action (a VP or an S, I'd have to check further): "I fed the gremlins after midnight, which you should never do."

Now this is a much more interesting example than mine, I think. In this case, what sort of constituent exactly is “which you should never do”? It looks like a relative clause, but in that case, what’s it modifying? Relative clauses, as far as I’m aware, are nominal modifiers by definition, but this one looks like it’s modifying the whole clause. In this regard it looks almost like an adverbial, but it doesn’t have the same freedom of movement: *“Which you should never do, I fed the gremlins after midnight.”

The first question is easy— "which you should never do" is an S— but that's because both relative clauses and the alternatives are S's. What alternatives? Well, it could be a conjunction, similar to "...as you should never do". Or even a preposition, as in "...after you went to bed."

Hold on, I think I’ve missed something — what’s an S? I thought it corresponded to a clause.

I'm leaning against a preposition. Prepositions can take S's but also NPs ("after midnight"), and "which" can't take an NP. Also, as you note, the construction can't be fronted.

Conjoints can often be fronted too: "As you should never do, I fed..." But this isn't infallible; cf. "*But you should never do so, I fed..."

I don't think we'd be tempted to analyze "as you should never do" as a relative clause.

My intuition is that "which S" kind of feels like a relative clause. Maybe that's why it resists fronting, just as you can't say "*Which the system lost, I'm resending my email." Note that I accept fuzzy categories, which Chomsky would shudder at.

I can agree with this — I also accept fuzzy/prototype-defined categories. It just doesn’t feel especially satisfying in this particular case. (I think this is because relative clauses are purely a matter of syntax, whereas other prototype-defined categories like aspect and alignment are more semantically-based.)

[zompist]

Hold on, I think I’ve missed something — what’s an S? I thought it corresponded to a clause.

Nothing mysterious; it stands for "sentence", and can usually be identified by having an NP and VP in it. In GG, "clauses" are sentences in deep structure.

(I'd have to rewrite that carefully to make it acceptable to a Chomskyan, but so far as I'm concerned they just like to re-label things in unhelpful ways.)

[bradrn]

An interesting example of ambiguity I came across recently: ‘the stakes of the meeting cannot be overestimated’. The intended meaning is that cannot should be interpreted as a marker of dynamic modality: the meeting is so important that the reader is literally unable to overestimate its stakes. But several people in the article comments seem to have interpreted it as a marker of deontic modality instead: as saying that the meeting is so unimportant that the reader is forbidden to overestimate its stakes. Of course, this is nearly opposite to the intended meaning.

What I find so interesting about this is that normally modals like can are analysed using terms like ‘dynamic possibility’ and ‘deontic possibility’, which emphasise the similarity of the alternate interpretations. And these categories normally are pretty similar — if I say that ‘you cannot run quickly’, then no matter whether cannot here is dynamic or deontic, the outcome is more or less the same; namely, that you will end up not running quickly. ‘Cannot be overestimated’ in fact seems somewhat odd from this perspective, in having two near-opposite interpretations. Can anyone else find any similar sentences? Or even figure out what the conditioning factor is?

(Another sentence which nearly works is ‘this bag cannot be full enough’, but here the deontic interpretation is entirely incoherent rather than opposite. ‘Enough’, for some reason, seems to be a word which cannot occur in a deontic context at all: e.g. ?‘you are required to work on your assignment enough’. But I won’t worry about this for now; I suspect this is one of those areas which should be studied one rabbit-hole at a time.)

[zompist]

An interesting example of ambiguity I came across recently: ‘the stakes of the meeting cannot be overestimated’. The intended meaning is that cannot should be interpreted as a marker of dynamic modality: the meeting is so important that the reader is literally unable to overestimate its stakes. But several people in the article comments seem to have interpreted it as a marker of deontic modality instead: as saying that the meeting is so unimportant that the reader is forbidden to overestimate its stakes. Of course, this is nearly opposite to the intended meaning.

Interesting. However, I'm not sure that the difference is deontic.

With "cannot", it may be useful to be clearer about where the impossibility lies. Logical necessity? Physical law? Exhaustion? Human law? Political factors? Morality? "Overestimating" cannot be a matter of physical or logical impossibility-- importance is not a physical or logical fact. So it's more like two forms of deontic reasoning.

I'm not sure, but I think the ambiguity here is between "cannot = we are prevented (by something)" and "cannot = should not". A simpler example might be "We cannot kill this snake", as seen by the possible continuations "...it's too powerful" or "...it would be wrong."

Of course, "should not" could be restated as "prevented by morality"! But morality is notoriously a very weak constraint, so saying that X is immoral is also implying that X is actually pretty easy or, humans being what they are, pretty likely.

[bradrn]

Reviving this thread to mention an interesting observation about English quantifiers (discovered in the process of conlanging):

1. We only didn’t see the Louvre ⇒ It was only the Louvre that we didn’t see

Under focalisation (I think this particular construction is clefting?), the quantifier seems to ‘jump’ from modifying the whole VP, to modifying the object NP alone. From a logical perspective, the latter would seem to be more coherent — there was a single thing we didn’t see, and that single thing was the Louvre. But in the first sentence (before the arrow), instead of staying in that position, the quantifier seems to move back to before the whole VP.

The really curious thing here is that it needs to move all the way through the VP. If you only move it past ‘see’, the result is quite different:

2. We didn’t only see the Louvre ⇒ It was seeing the Louvre which we didn’t only do

Another curiosity — when focalising, you need to be careful with where you put the ‘only’:

3. It was only seeing the Louvre which we didn’t do

This no longer means the same as (2); instead, it’s nearly synonymous with (1). Usually, this is the point where I throw up my hands and say ‘I’ve never been able to understand quantifiers’, but there must be some explanation for all this…