Postpositions?

[bradrn]

The immediate point is, you can take a VP and treat it as a noun; and this isn't restricted to English-- cf. French gratte-ciel 'skyscraper'. This is exactly the sort of grammatical flexibility François talks about. The larger point is that languages are really complex, and looking at the most prototypical behaviors of word classes leads to many errors.

Sure, I’m not arguing about this at all. I just don’t see how it invalidates François’s argument in any way.

‘Know’ is not acting as a noun in to know, which is a construction applicable to all verbs without exception — and also no nouns I know of.
[...] ‘Knowing’ fits in its own class, with some properties characteristic of nouns and some characteristic of verbs — but all verbs can occur in this form, meaning that ‘can take -ing’ is a characteristic delineating verbs specifically.

But this is just pure dogma. Why do you call infinitives and participles "verbs"? Because some Latin grammarian did so 1500 years ago?

There is no dogma here whatsoever. I am not at all calling infinitives ‘verbs’ — for one thing, an infinitive is not a single word, so it has no word class. What I am saying is that ‘know’ is not acting as a noun in ‘to know’, even if the construction ‘to know’ can be used in ways similar to how nouns can be used.

This isn't to say that nouns and verbs can't be distinguished-- only that there's a gradation between them. An infinitive is way more nouny than a verb, and it's arbitrary where we draw the line.

I think you’re misunderstanding what I’m talking about here. I’m talking about word classes, which are a property of individual words. I am saying nothing about the properties of phrases made up of multiple words. Certainly the English infinitive shares some properties with nouns — but that says nothing about word classes, since an infinitive is not a word.

As for the word classes themselves… well, there aren’t really any categories ‘between’ nouns and verbs in English. ‘Nouns’ and ‘verbs’ can be very clearly distinguished in English — I can even draw a little table like François does to outline the distinguishing properties of different English word classes:

	Nouns	Adjectives	Verbs	Prepositions	Adverbs (sensu stricto)
Can occur after ‘the’	✔	✔
Can occur after ‘to’	✔		✔		✔
Can take ‘-ing’			✔
Can occur before ‘the’			✔	✔

Note that this analysis is entirely pre-theoretical: I’m not even distinguishing directional ‘to’ from infinitive ‘to’. Also, some words can act as both nouns and verbs; they are then members of both word classes simultaneously. (François calls this ‘lexical flexibility’.)

[Travis B.]

Take into account, though, that there are differences between nouns derived from verbs and normal nouns in English. For instance, infinitives in English are unmarked for number (and not just that they don't inflect for number but also that verbs always agree with them as subjects as singular) or definiteness and cannot be qualified with adjectives or prepositional phrases as NP's, unlike normal nouns; conversely, they can take NP's as arguments, be qualified by adverbial prepositional phrases, and be qualified by adverbs, like verbs.

[zompist]

The immediate point is, you can take a VP and treat it as a noun; and this isn't restricted to English-- cf. French gratte-ciel 'skyscraper'. This is exactly the sort of grammatical flexibility François talks about. The larger point is that languages are really complex, and looking at the most prototypical behaviors of word classes leads to many errors.

Sure, I’m not arguing about this at all. I just don’t see how it invalidates François’s argument in any way.

Let's take a step back. I didn't say his argument (which one?) is invalid, but let me try to be clearer about what I did and didn't agree with.

His point on lexical vs grammatical flexibility is fine.

Really, my disagreement isn't so much with him as with you. You say "‘Syntactic class X’ is merely a shorthand for saying that ‘words belonging to X are distinguishable by sharing such-and-such traits’. In any given language, there should be very few words which do not fit into one or more word classes." These are both dubious propositions and are not really supported by François’s paper.

The idea that syntactic categories have certain traits is not wrong-headed; it's just an overgeneralization, like "birds can fly" or "mammals don't lay eggs". As I said, languages are complex. People's first generalizations about syntax are usually wrong-- you look deeper and you find exceptions. I gave plenty of examples of this in the SCK.

E.g. you can propose some rules for whether something is a "preposition". Again, I have a section on this in my book: a proposed rule may work most of the time, and yet sometimes fail. Prototype theory ("fuzzy categories") is better at handling this situation than hoping for universal, ironclad rules.

François seems a bit careless about these things. E.g. he says ship is both verb and noun, while paper is not. That's not terrible, it just fails (I think) to capture how the noun-to-verb pipeline works in English. We can verb our nouns pretty easily. It's true that your copy editor will let you ship a cargo, or truck it, but won't let you car it or airplane it. Yet I could tell you that Frodo and Sam eagled it out of Mordor, and you'd surely understand. Using a noun as a verb, with the meaning "use the noun in the ordinary way", is not as productive as N + N compounds, but it does work.

His statements about N + N compounds aren't quite accurate either; perhaps most notably, he forgets the very issue you're a stickler about later-- that the modifier is an NP (or N'), not an N.

In any case, my point is not that he is wrong in general, it's that defining syntactic categories by syntactic behavior turns out to be harder than it sounds. There are almost always weird exceptions, and no syntactic test works all the time.

(Oh-- on "there should be very few words which do not fit into one or more word classes"-- it may well be that 90% of words are easily slotted into a syntactic category. Well, in programming we say "The first 90 percent of the code accounts for the first 90 percent of the development time. The remaining 10 percent of the code accounts for the other 90 percent of the development time." It's about the same for syntax. You can't know a priori that all of your lexicon will neatly fit into the six slots of Latin grammarians. Or any six slots. Maybe you're lucky and they do; but probably not.)

I think you’re misunderstanding what I’m talking about here. I’m talking about word classes, which are a property of individual words. I am saying nothing about the properties of phrases made up of multiple words. Certainly the English infinitive shares some properties with nouns — but that says nothing about word classes, since an infinitive is not a word.

What do you call "know" in "I don't know" or "I must know"? They're infinitives to me.

As for the word classes themselves… well, there aren’t really any categories ‘between’ nouns and verbs in English. ‘Nouns’ and ‘verbs’ can be very clearly distinguished in English — I can even draw a little table like François does to outline the distinguishing properties of different English word classes:

This is full of errors, so you're proving my point rather than yours: facile generalizations about syntactic categories are likely to be wrong.

But more importantly, even if you got your table right, you are not even looking for "categories ‘between’ nouns and verbs". You'll never find the gradations if you look only at prototypical behaviors and simple frames.

[bradrn]

Really, my disagreement isn't so much with him as with you. You say "‘Syntactic class X’ is merely a shorthand for saying that ‘words belonging to X are distinguishable by sharing such-and-such traits’. In any given language, there should be very few words which do not fit into one or more word classes." These are both dubious propositions and are not really supported by François’s paper.
…
(Oh-- on "there should be very few words which do not fit into one or more word classes"-- it may well be that 90% of words are easily slotted into a syntactic category. Well, in programming we say "The first 90 percent of the code accounts for the first 90 percent of the development time. The remaining 10 percent of the code accounts for the other 90 percent of the development time." It's about the same for syntax. You can't know a priori that all of your lexicon will neatly fit into the six slots of Latin grammarians. Or any six slots. Maybe you're lucky and they do; but probably not.)

I agree with your parenthetical, which states my own argument much better than I did. ‘Most words can be fitted into word classes’ does not mean that there are other words which cannot!

Also, I said nowhere that ‘all of your lexicon will neatly fit into the six slots of Latin grammarians’ — I also find this attitude thoroughly irritating, and like efforts to find the best word class system for individual languages.

E.g. you can propose some rules for whether something is a "preposition". Again, I have a section on this in my book: a proposed rule may work most of the time, and yet sometimes fail. Prototype theory ("fuzzy categories") is better at handling this situation than hoping for universal, ironclad rules.

In general, I agree of this. I am a great advocate for prototype theory, even in areas where it isn’t usually applied. (Which reminds me, a while ago I started writing a long post on prototype theory in alignment and TAM; I really should finish it one of these days…) However, it seems to be that emic word classes are one of the few areas where prototype theory is more or less irrelevant. You certainly haven’t given me any counterexamples yet — these would be welcomed if you have any!

François seems a bit careless about these things. E.g. he says ship is both verb and noun, while paper is not. That's not terrible, it just fails (I think) to capture how the noun-to-verb pipeline works in English. We can verb our nouns pretty easily. It's true that your copy editor will let you ship a cargo, or truck it, but won't let you car it or airplane it. Yet I could tell you that Frodo and Sam eagled it out of Mordor, and you'd surely understand. Using a noun as a verb, with the meaning "use the noun in the ordinary way", is not as productive as N + N compounds, but it does work.

I think this is one area where a diachronic analysis is more useful than a synchronic one. English is pretty ‘lexically flexible’ with regards to its nouns and verbs — there are a lot of words which are in both classes. What we are seeing now, I believe, is a situation where more and more nouns are becoming verbs also. The natural conclusion should be a situation where almost all nouns can be used in those places where verbs can currently be used — or, putting it another way, the list of places where ‘nouns’ can be used will encompass everywhere where ‘verbs’ can be used today. (Dixon’s Basic Linguistic Theory vol. 2 has a good section on this sort of thing, IIRC.) ‘Lexical flexibility’ will have turned into ‘grammatical flexibility’.

However, note that none of this requires prototype theory — it requires merely a knowledge of the fact that words may move from one class to another, the places where word classes may be used can expand and shrink, and both of these are gradual processes.

I think you’re misunderstanding what I’m talking about here. I’m talking about word classes, which are a property of individual words. I am saying nothing about the properties of phrases made up of multiple words. Certainly the English infinitive shares some properties with nouns — but that says nothing about word classes, since an infinitive is not a word.

What do you call "know" in "I don't know" or "I must know"? They're infinitives to me.

I call it a ‘verb’. ‘Don’t’ and ‘must’ are auxiliaries, and any verb may occur as a bare stem after an auxiliary. I see no reason why these clauses require a separate ‘infinitive’ category.

As for the word classes themselves… well, there aren’t really any categories ‘between’ nouns and verbs in English. ‘Nouns’ and ‘verbs’ can be very clearly distinguished in English — I can even draw a little table like François does to outline the distinguishing properties of different English word classes:

In any case, my point is not that he is wrong in general, it's that defining syntactic categories by syntactic behavior turns out to be harder than it sounds. There are almost always weird exceptions, and no syntactic test works all the time.

…

This is full of errors, so you're proving my point rather than yours: facile generalizations about syntactic categories are likely to be wrong.

Um, well, then, what errors are there? I don’t see any. These tests seem to work fine to me.

But more importantly, even if you got your table right, you are not even looking for "categories ‘between’ nouns and verbs". You'll never find the gradations if you look only at prototypical behaviors and simple frames.

So then tell me, could you give me any examples of these ‘in-between’ categories please?

[zompist]

In general, I agree of this. I am a great advocate for prototype theory, even in areas where it isn’t usually applied.

OK, cool!

However, it seems to be that emic word classes are one of the few areas where prototype theory is more or less irrelevant. You certainly haven’t given me any counterexamples yet — these would be welcomed if you have any!

I have, and so did Moose! They're all over once you start looking. See for instance the examples on p. 79 in the SCK, or the discussion of prototypes on p. 251. In earlier posts you'll find a gradation between prepositions and particles.

Um, well, then, what errors are there? I don’t see any. These tests seem to work fine to me.

Can occur after 'the': the quickly eaten pancakes; the under $5 books; the come to Jesus meeting.
Can occur after 'to': give to friendly people, set the thermometer to over 70°.
Can occur before 'the': give a man the finger, I felt fine the morning after, consider carefully the circumstances

The morphological one ('-ing') is safe, though it commits you to the idea that English modals are not verbs. (That happens to be the Chomskyan position, but e.g. McCawley disagrees.)

[bradrn]

However, it seems to be that emic word classes are one of the few areas where prototype theory is more or less irrelevant. You certainly haven’t given me any counterexamples yet — these would be welcomed if you have any!

I have, and so did Moose! They're all over once you start looking.

As I said, I disagree as to this analysis of your and Moose’s examples.

See for instance the examples on p. 79 in the SCK, or the discussion of prototypes on p. 251. In earlier posts you'll find a gradation between prepositions and particles.

Unfortunately, I have the Kindle version, so without page numbers; could you provide quotes please?

As for prepositions and particles, well… at the moment I’m not even sure if there is a difference between them — all claimed instances of ‘particles’ seem to actually be prepositions. Even just distinguishing prepositions from other word types will have to wait until I find a good syntactic test which works for prepositions but not other word types. In general, I’m a bit confused about this area.

Um, well, then, what errors are there? I don’t see any. These tests seem to work fine to me.

Can occur after 'the': the quickly eaten pancakes; the under $5 books; the come to Jesus meeting.
Can occur after 'to': give to friendly people, set the thermometer to over 70°.
Can occur before 'the': give a man the finger, I felt fine the morning after, consider carefully the circumstances

I’m inclined to analyse each of ‘under $5’, ‘come to Jesus’, ‘over 70°’, ‘the morning after’ as single grammatical words (i.e. compounds), considering that I have trouble thinking of any words which could be interposed grammatically, and that these constructions show limited productivity^†. Your other examples are valid though, so the point stands.

All that being said, I still believe it should be possible to find syntactic tests (or combinations of tests) which can isolate ‘noun’ and ‘verb’ as distinct word classes, since (in English at least) they’re fundamentally syntactic concepts.

________
^† The construction in ‘under $5’ and ‘over 70°’ can really only be used nominally with those two prepositions; ‘come to Jesus’ is clearly idiomatic, given that I don’t even really understand it; and ‘the morning after’ is limited to ’after’, and practically limited to ‘the morning’ as well, given that in natural speech, that’s the only noun commonly used in this construction. If there are any counterexamples to these, I can’t think of them.

The morphological one ('-ing') is safe, though it commits you to the idea that English modals are not verbs. (That happens to be the Chomskyan position, but e.g. McCawley disagrees.)

Well, I’ve yet to see any good arguments that auxiliaries are verbs in any way; the best you can do is to say that some auxiliaries can also be used like verbs, and inflect for third person like verbs. And even that argument says nothing about auxiliaries like ‘must’ and ‘can’, which synchronically have completely distinct behaviour to verbs.

---

I suppose the point I’m getting at here is: well, if we don’t define word classes via their morphosyntactic behaviour, how exactly do we define them? Semantics doesn’t really work, because ‘noun’ and ’verb’ are semantically heterogeneous in almost all languages, so you always end up imposing a badly-fitting classification from above. And etics has its place, but, well… for such a primitive concept as ‘word class’, emic approaches are almost always best, especially when there is such variation between languages.

So then how, exactly, do I think a ‘word class’ should be defined? I consider a word class to be, quite simply, a set of words sharing the same morphosyntactic behaviour. This definition is not merely satisfactory as a definition: because it can be assessed on a purely formal level, it is really the only definition that can be applied with minimal analysis of the language in question. This is important, to avoid circular reasoning (e.g. ‘the noun is the word which heads a noun phrase’ / ‘the noun phrase is the constituent which contains a noun’)

Due to this, prototype theory doesn’t really work for word classes. Prototypes work best when applied to semantics, or to morphosyntactic issues where semantics plays a heavy role (like TAM, or alignment). But I still find it hard to see how prototype theory could even be applied to a purely formal concept like word classes. I just don’t understand how a single word can be halfway between a noun and a verb, as you imply is possible — either it behaves formally like a noun and hence is a noun, or it behaves formally like a verb and hence is a verb, or it behaves in some ways similarly to nouns and in some ways similarly to verbs, in which case it is neither a noun nor a verb and is in a third word class altogether.

(Apparent exception: the literature on adjectives makes heavy reference to ‘nouny’ and ‘verby’ adjectives. But this isn’t really an exception at all, since they actually refer to formal and functional properties of the adjective class in question: ‘nouny’ adjectives require a copula for equation, while ‘verby’ adjectives don’t.)

Of this whole argument, probably the weakest part is the second paragraph: that word classes should be defined in such a way that they require minimal analysis or theory. I note that even François feels like he shouldn’t go that far: he is happy to define e.g. ‘weak nouns’ in Hiw as those nouns which can act as ‘modifier[s] in argument phrase’, and as ‘head[s] of TAM-inflected predicate[s]’. But as I said, this makes me feel uncomfortable, due to the potential for circular reasoning. (How does one define a ‘modifier in argument phrase’ independent of the concept of nouns and verbs?) I am still unsure as to exactly what is the best way to define word classes; even so, I think we can all agree that word classes should be defined without reference to semantics, so my argument against prototype theory holds.

[zompist]

Even just distinguishing prepositions from other word types will have to wait until I find a good syntactic test which works for prepositions but not other word types. In general, I’m a bit confused about this area.

No problem there-- I haven't seen a great definition either. (Again, naive definitions exist and can get us started, but often don't help with difficult examples.)

All that being said, I still believe it should be possible to find syntactic tests (or combinations of tests) which can isolate ‘noun’ and ‘verb’ as distinct word classes, since (in English at least) they’re fundamentally syntactic concepts.

You can do pretty well with morphology: English verbs can be inflected for past tense (and 3s, and progressive); nouns can be inflected by number. Though this test won't work in all cases-- e.g. in N+N compounds the first noun usually can't be inflected; verbs don't inflect in non-finite clauses, etc.

Other categories are worse-- not all adjectives inflect; adverbs / prepositions / conjunctions have no morphology.

That doesn't mean it's hopeless, but it's tricky, and syntacticians can come up with different analyses.

The construction in ‘under $5’ and ‘over 70°’ can really only be used nominally with those two prepositions;

When I can come up with counterexamples in a minute of thinking, that should make you look widely at more examples and refine your analysis, not double down on your initial idea. More examples of 'the' + preposition: the below zero weather, the on edge actors, the after action report, the around the world trip, the by the book cop. Try to think of examples yourself rather than trying to shoot down counterevidence!

I had a boss who liked to have "come to Jesus meetings"-- this refers to revival meetings, and he thought they were inspirational. (They weren't.) Again, you could take a minute and think of a dozen more examples. The dance all night crowd, the go all out maneuver, the find all the things playthrough.

Well, I’ve yet to see any good arguments that auxiliaries are verbs in any way; the best you can do is to say that some auxiliaries can also be used like verbs, and inflect for third person like verbs. And even that argument says nothing about auxiliaries like ‘must’ and ‘can’, which synchronically have completely distinct behaviour to verbs.

You could track down McCawley's syntax book, which is excellent anyway. But really, "auxiliaries can behave a lot like verbs" isn't exactly a strong argument that auxiliaries aren't verbs. "Be" and "have" inflect like verbs, head a VP like verbs, can be negated like verbs, and don't act like nouns or adjectives. Modals are less like verbs but have at least some of these features. You can separate out most verbs based on other behavior-- mostly, that they require Do-support. But that doesn't tell you if the best analysis is that verbs are a narrow or broad category.

I suppose the point I’m getting at here is: well, if we don’t define word classes via their morphosyntactic behaviour, how exactly do we define them?

Who says we can't? I haven't at any point said we can't define categories using morphosyntax; I've said it was complicated, that naive tests and ideas break down, and that categories can be fuzzy.

But I still find it hard to see how prototype theory could even be applied to a purely formal concept like word classes. I just don’t understand how a single word can be halfway between a noun and a verb, as you imply is possible — either it behaves formally like a noun and hence is a noun, or it behaves formally like a verb and hence is a verb, or it behaves in some ways similarly to nouns and in some ways similarly to verbs, in which case it is neither a noun nor a verb and is in a third word class altogether.

OK, take "reading." It can be used as a noun: "reading is fun." It can be used as a verb: "I am reading Tolstoy."

And that's without bringing in other languages, where participles have a bad habit of becoming verb forms. You can perhaps claim that "reading" is not a verb in English, but is читал not a verb? It's how you form the Russian past tense.

And sure, confronted with a continuum of behaviors, you can split them all up and give them new names. Sometimes that's the right move! But you may also end up with several dozen categories with arbitrary and confusing names. There's no guarantee that the splitting process will end up with a small, easily understood set of categories.

[bradrn]

All that being said, I still believe it should be possible to find syntactic tests (or combinations of tests) which can isolate ‘noun’ and ‘verb’ as distinct word classes, since (in English at least) they’re fundamentally syntactic concepts.

You can do pretty well with morphology: English verbs can be inflected for past tense (and 3s, and progressive); nouns can be inflected by number. Though this test won't work in all cases-- e.g. in N+N compounds the first noun usually can't be inflected; verbs don't inflect in non-finite clauses, etc.

I actually did think of this noun test; unfortunately it only works as a test for count nouns, since mass nouns have no plural. And then there’s words like fish or sheep, where the number inflection is not overt. (Hmm, is that the right word? Probably not.) Morphological tests seem to work better for verbs than for nouns in English — I assume this is related to the cross-linguistic generalisation that verbs tend to have more morphology.

Other categories are worse-- not all adjectives inflect; adverbs / prepositions / conjunctions have no morphology.

Adjectives are an interesting, since depending on the language they can act either like nouns or like verbs. In English, they act exactly like nouns, except they can’t head an indefinite noun phrase. This would be fine for François, since he accepts this sort of thing as a criterion, but I feel uncomfortable for the already mentioned reasons. I still haven’t worked out good tests for the other word classes.

The construction in ‘under $5’ and ‘over 70°’ can really only be used nominally with those two prepositions;

When I can come up with counterexamples in a minute of thinking, that should make you look widely at more examples and refine your analysis, not double down on your initial idea. More examples of 'the' + preposition: the below zero weather, the on edge actors, the after action report, the around the world trip, the by the book cop. Try to think of examples yourself rather than trying to shoot down counterevidence!

I never ‘doubled down’ — for reasons of accuracy I pointed out that some ‘counterexamples’ weren’t, but I also noted that many others were valid. I agree with you that my analysis was wrong. That being said, even if that one attempt turned out to be wrong, I still believe that with the right syntactic tests, this approach will work: there is surely some sort of difference in syntax which distinguishes English nouns from verbs and adjectives.

(You might argue that this hypothesis is unfalsifiable, in the Popperian sense. And this would be true! But truly falsifiable hypotheses are hard to come by in syntax.)

I had a boss who liked to have "come to Jesus meetings"-- this refers to revival meetings, and he thought they were inspirational. (They weren't.) Again, you could take a minute and think of a dozen more examples. The dance all night crowd, the go all out maneuver, the find all the things playthrough.

After thinking about these for a bit, I’m not quite sure how to analyse them. Phonologically, they seem to be single words: I pronounce ‘find all the things’ faster and with more phonological reduction (e.g. /d/-flapping) as a modifier than as a VP, though I have yet to properly confirm this on a spectrogram. Then again, phonological wordhood is irrelevant to my argument. Syntactically, at first I thought they might be just adjectival compounds, but they don’t behave like proper adjectives: *the playthrough was find all the things vs the playthrough was quick.

Well, I’ve yet to see any good arguments that auxiliaries are verbs in any way; the best you can do is to say that some auxiliaries can also be used like verbs, and inflect for third person like verbs. And even that argument says nothing about auxiliaries like ‘must’ and ‘can’, which synchronically have completely distinct behaviour to verbs.

You could track down McCawley's syntax book, which is excellent anyway. But really, "auxiliaries can behave a lot like verbs" isn't exactly a strong argument that auxiliaries aren't verbs. "Be" and "have" inflect like verbs, head a VP like verbs, can be negated like verbs, and don't act like nouns or adjectives. Modals are less like verbs but have at least some of these features. You can separate out most verbs based on other behavior-- mostly, that they require Do-support. But that doesn't tell you if the best analysis is that verbs are a narrow or broad category.

Sorry; I phrased my argument badly there. Emphasis should be on the some in ‘some auxiliaries can also be used like verbs, and inflect for third person like verbs: the set is pretty much limited to ‘be’, ‘have’ and ‘do’. But as for the other auxiliaries—‘must’, ‘can’, ‘may’, ‘shall’ and all the rest—I’m still waiting to see what verby behaviour they have.

(Also, would McCawley’s book be The Syntactic Phenomena of English? You’ve recommended that before, so perhaps I’ll try to find a copy. Hmm, I’m going on campus this week; I wonder if the uni library will have a copy…)

But I still find it hard to see how prototype theory could even be applied to a purely formal concept like word classes. I just don’t understand how a single word can be halfway between a noun and a verb, as you imply is possible — either it behaves formally like a noun and hence is a noun, or it behaves formally like a verb and hence is a verb, or it behaves in some ways similarly to nouns and in some ways similarly to verbs, in which case it is neither a noun nor a verb and is in a third word class altogether.

OK, take "reading." It can be used as a noun: "reading is fun." It can be used as a verb: "I am reading Tolstoy."

And that's without bringing in other languages, where participles have a bad habit of becoming verb forms. You can perhaps claim that "reading" is not a verb in English, but is читал not a verb? It's how you form the Russian past tense.

Gerunds such as ‘reading’ are an interesting case. I’d analyse them as an entirely separate word class, which shares all the properties of nouns, most (but not all!) the properties of verbs, and has some new properties of its own — the latter fact being entirely ignored by any analysis which merely calls them ‘something between a noun and a verb’. (My words, not yours, but you seem to agree with that assessment.) As for the Russian example, I know too little to comment there.

And sure, confronted with a continuum of behaviors, you can split them all up and give them new names. Sometimes that's the right move! But you may also end up with several dozen categories with arbitrary and confusing names. There's no guarantee that the splitting process will end up with a small, easily understood set of categories.

Of course! Given a continuum, this would be for the most part senseless. But as I have argued, word classes do not form a continuum. At least in English, there is no set of words which gradually fades from prototypical nouns to half-noun-half-verbs to prototypical verbs in the same way as, say, affixes fade from inflectional to derivational, or aspectual categories fade from perfective to stative to imperfective, or alignments fade from nominative-accusative to split ergative to ergative-absolutive. And therefore, it is entirely reasonable to split words up into as many classes as necessary so that we aren’t conflating separate concepts.

I suppose I should elaborate on exactly why I think word classes differ from these other continua. It is because the defining properties of those continua are themselves continuous. Consider Dahl’s definition of the prototypical perfective:

A PFV verb will typically denote a single event, seen as an unanalysed whole, with a well-defined result or end-state, located in the past. More often than not, the event will be punctual, or at least, it will be seen as a single transition from one state to its opposite, the duration of which can be disregarded.

Practically all of these defining categories are continua: single/double/triple/multiple/infinite/neverending events, seen as a whole/without endpoints/in small sections/a single moment, with a well-defined/poorly-defined/no result, and so on. ‘Located in the past’ is the only part of that which might in any way be binary. It is the same with all the other continua in linguistics: for alignment, for instance, it goes from all A unmarked+no P unmarked, to inanimate A marked+non-inanimate P marked, to non-human A unmarked+human P unmarked, and so on and so forth until we get to no A unmarked+all P unmarked. (Note that this is a more complex continuum: there is more than one way to get from the start to the end!)

With word classes, however, our only, or at least main, tool is morphosyntactic behaviour — you seem to agree with me on this. And morphosyntactic behaviour is binary: either a word fits in a construction/takes a specific affix, or it does not. There is no continuum here. When we start combining syntactic tests (‘Hiw numerals are words which can head modifier phrases and modify argument phrases and head TAM-inflected predicates and head direct predicates’), then it gets a bit easier to define continuum-like things, since we can imagine all possible combinations of those binary conditions — but it’s very rare, probably unattested, for a language to have words which fill in all possible binary combinations, and in any case that would still a lot more discrete than the average linguistic continuum.

(I have a horrible feeling the underlying problem here is differing mental models: when I think of word class definitions, I literally think of little switches in a table toggling on and off. I’m not sure exactly how you think of word classes, but it’s probably different. My posts here have just been arguing that my mental model is a correct way to think about things. I’m not sure if there’s any way to resolve these differences, at least via text over the Internet.)

[Richard W]

I phrased my argument badly there. Emphasis should be on the some in ‘some auxiliaries can also be used like verbs, and inflect for third person like verbs: the set is pretty much limited to ‘be’, ‘have’ and ‘do’. But as for the other auxiliaries—‘must’, ‘can’, ‘may’, ‘shall’ and all the rest—I’m still waiting to see what verby behaviour they have.

Apart from must, they change to past forms in accordance with the sequence of tenses. As to must, it patterns with other auxiliaries, rather than visibly with prototypical vowels. (The pattern is that of subject-verb inversion.) The change to a past form is a null operation, as with several more prototypical verbs, such as hit and put.

[zompist]

After thinking about these for a bit, I’m not quite sure how to analyse them. Phonologically, they seem to be single words: I pronounce ‘find all the things’ faster and with more phonological reduction (e.g. /d/-flapping) as a modifier than as a VP, though I have yet to properly confirm this on a spectrogram. Then again, phonological wordhood is irrelevant to my argument. Syntactically, at first I thought they might be just adjectival compounds, but they don’t behave like proper adjectives: *the playthrough was find all the things vs the playthrough was quick.

Now you're doing syntax. I agree with your asterisk here, but on the other hand I don't have a problem with "My playstyle is find all the things" or "His method is dive in and take no prisoners." I haven't seen an analysis of this construction, so it's not too clear how to handle it. It looks like a VP, but who knows how it fits into the parent NP.

Sorry; I phrased my argument badly there. Emphasis should be on the some in ‘some auxiliaries can also be used like verbs, and inflect for third person like verbs: the set is pretty much limited to ‘be’, ‘have’ and ‘do’. But as for the other auxiliaries—‘must’, ‘can’, ‘may’, ‘shall’ and all the rest—I’m still waiting to see what verby behaviour they have.

Those are the modals, BTW.
1. They inflect for tense, like verbs (as Richard W pointed out).
2. They head a VP, like verbs.
3. They take an NP argument ('subject'), like verbs.
4. They can be negated with __ not or __n't, like verbs.
5. They pattern with auxiliary be/have in not requiring Do-support. (Historically of course this is how regular verbs behaved, and still can in some cases— "I think not.")

Of course they have differences from ordinary verbs also, e.g.
1. They can't take Do-support.
2. They don't have a 3s inflection.
3. They have no infinitive form, and thus can only head the VP.

(Also, would McCawley’s book be The Syntactic Phenomena of English? You’ve recommended that before, so perhaps I’ll try to find a copy. Hmm, I’m going on campus this week; I wonder if the uni library will have a copy…)

Yes, that's it. It's long, but that's one of its virtues— it doesn't have to leave things out as much as most intros do.

Gerunds such as ‘reading’ are an interesting case. I’d analyse them as an entirely separate word class, which shares all the properties of nouns, most (but not all!) the properties of verbs, and has some new properties of its own — the latter fact being entirely ignored by any analysis which merely calls them ‘something between a noun and a verb’. (My words, not yours, but you seem to agree with that assessment.)

Something can be like X and like Y, and also have characteristics of its own. (Off the top of my head: horse / mule / donkey, ice / water / steam, PIE / Old English / Modern English, poem / song / melody.)

I have a horrible feeling the underlying problem here is differing mental models: when I think of word class definitions, I literally think of little switches in a table toggling on and off. I’m not sure exactly how you think of word classes, but it’s probably different.

Fair enough. Your picture is very Chomskyan— that boy dreams in binary features. As should have been clear from my book, I admire him but question a lot of his presuppositions, including this one.

I find prototype theory more compelling for basic organization. As I said, once you start thinking that way, you see it all over. I'm kind of surprised that you see it in the TAM system but not with nouns & verbs!

This may help. Do your feature-based analysis on English, working out purely syntactic definitions for whatever categories you can find. Now do it for Mandarin.

Now... how do you identify which of the categories you found in English are "verbs", and why do you think it's the "same thing" in some sense as verbs in Mandarin?

The thing is, your tests will be abstract things like "English verbs can be inflected for past tense" and "Mandarin verbs can take the 了 le aspect particle." Those things are not comparable: Mandarin verbs have no past tense, English verbs can't take 了. Strictly speaking, you can't say that both languages have "verbs", since you have not provided a definition that covers both.

Now, Chomskyans think they can still do it. They believe in universal grammar, and so a "verb" is something at the heart of a VP, which can be identified purely by the tree structure. That's... interesting. But in practice it's more or less manhandling every language to look like English, and going to a whole lot of trouble that is avoided in other theories which can directly deal with ideas like arguments and valence. And verbs.

So, I guess I'd modify my earlier stance a little. While you're analyzing a particular language's syntax, you should define categories by syntactic methods alone. But as soon as you start hooking up semantics, you can start using semantics; and when you compare languages, our best guide to using the same words the same way is semantic.

[Richard W]

Those are the modals, BTW.
1. They inflect for tense, like verbs (as Richard W pointed out).
2. They head a VP, like verbs.
3. They take an NP argument ('subject'), like verbs.
4. They can be negated with __ not or __n't, like verbs.
5. They pattern with auxiliary be/have in not requiring Do-support. (Historically of course this is how regular verbs behaved, and still can in some cases— "I think not.")

Some of these are very weak. Properties 2 and 3 would also work for an adverb like rarely. Property 4 is misstated - very few verbs can be negated with __n't, and I think not patterns with I think so, though I care not for your opinions is the sort of example you were looking for. For Property 5, I would rather say that all these auxiliaries prohibit do-support.

[KathTheDragon]

I’m inclined to analyse each of ‘under $5’, ‘come to Jesus’, ‘over 70°’, ‘the morning after’ as single grammatical words (i.e. compounds), considering that I have trouble thinking of any words which could be interposed grammatically, and that these constructions show limited productivity^†. Your other examples are valid though, so the point stands.

All that being said, I still believe it should be possible to find syntactic tests (or combinations of tests) which can isolate ‘noun’ and ‘verb’ as distinct word classes, since (in English at least) they’re fundamentally syntactic concepts.

________
^† The construction in ‘under $5’ and ‘over 70°’ can really only be used nominally with those two prepositions; ‘come to Jesus’ is clearly idiomatic, given that I don’t even really understand it; and ‘the morning after’ is limited to ’after’, and practically limited to ‘the morning’ as well, given that in natural speech, that’s the only noun commonly used in this construction. If there are any counterexamples to these, I can’t think of them.

"to over about 70°", "to over what I think says 70°", etc.
"the morning before", "the day after", "the next town over", etc.

These really aren't that hard to come up with. I'm dying to know how you dismiss these.

[zompist]

Those are the modals, BTW.
1. They inflect for tense, like verbs (as Richard W pointed out).
2. They head a VP, like verbs.
3. They take an NP argument ('subject'), like verbs.
4. They can be negated with __ not or __n't, like verbs.
5. They pattern with auxiliary be/have in not requiring Do-support. (Historically of course this is how regular verbs behaved, and still can in some cases— "I think not.")

Some of these are very weak. Properties 2 and 3 would also work for an adverb like rarely. Property 4 is misstated - very few verbs can be negated with __n't, and I think not patterns with I think so, though I care not for your opinions is the sort of example you were looking for. For Property 5, I would rather say that all these auxiliaries prohibit do-support.

The parenthetical on (5) was supposed to go with (4).

You can call modals auxiliaries if you want, but there are clear differences and it's useful to have a word to identify them.

How does "rarely" head a VP?

[bradrn]

Sorry; I phrased my argument badly there. Emphasis should be on the some in ‘some auxiliaries can also be used like verbs, and inflect for third person like verbs: the set is pretty much limited to ‘be’, ‘have’ and ‘do’. But as for the other auxiliaries—‘must’, ‘can’, ‘may’, ‘shall’ and all the rest—I’m still waiting to see what verby behaviour they have.

Those are the modals, BTW.
1. They inflect for tense, like verbs (as Richard W pointed out).
2. They head a VP, like verbs.
3. They take an NP argument ('subject'), like verbs.
4. They can be negated with __ not or __n't, like verbs.
5. They pattern with auxiliary be/have in not requiring Do-support. (Historically of course this is how regular verbs behaved, and still can in some cases— "I think not.")

Of course they have differences from ordinary verbs also, e.g.
1. They can't take Do-support.
2. They don't have a 3s inflection.
3. They have no infinitive form, and thus can only head the VP.

OK, this makes sense (modulo Richard W’s criticisms). I think I’m defining ‘verbs’ via morphology, whereas you’re doing it by syntax — if a verb is that thing which can head a VP, then auxiliaries are definitely verbs. But then again, is that the best definition? It’s hard to know.

Gerunds such as ‘reading’ are an interesting case. I’d analyse them as an entirely separate word class, which shares all the properties of nouns, most (but not all!) the properties of verbs, and has some new properties of its own — the latter fact being entirely ignored by any analysis which merely calls them ‘something between a noun and a verb’. (My words, not yours, but you seem to agree with that assessment.)

Something can be like X and like Y, and also have characteristics of its own.

Agreed! I totally accept that gerunds are like nouns and like verbs. But that doesn’t necessarily imply a continuum, or that ‘noun’ and ‘verb’ are merely prototypes, as was your original claim — for those, you need something which is actually between nouns and verbs, which is an entirely different thing to being merely similar to both.

(For an example of what I mean, look at, say, morphosyntactic alignment. Split ergative alignments are between accusative and ergative alignments, in that their properties are a perfect blend of those of the endpoints. Every single property of a split ergative/tripartite/accusative alignment is present in either accusative or ergative alignment; this is what makes prototype theory such a natural way to analyse them. By contrast, gerunds, as I said, have extra properties, making them ‘more’ than merely the combination of nouns and verbs: they are like both, but not between them. An analogous alignment would be, say, semantic ergative alignment as found in Basque and Tibetan, which have properties found in neither accusative nor ergative alignments — though they do exist on a different continuum, something which cannot be said for gerunds.)

I have a horrible feeling the underlying problem here is differing mental models: when I think of word class definitions, I literally think of little switches in a table toggling on and off. I’m not sure exactly how you think of word classes, but it’s probably different.

Fair enough. Your picture is very Chomskyan— that boy dreams in binary features. As should have been clear from my book, I admire him but question a lot of his presuppositions, including this one.

I find prototype theory more compelling for basic organization. As I said, once you start thinking that way, you see it all over. I'm kind of surprised that you see it in the TAM system but not with nouns & verbs!

Oh, I’m not at all Chomskyan. I have a strong dislike of binary features. However, I have been arguing that word classes seem to be a case where they genuinely reflect the linguistic reality, at least within languages. In fact, I struggle to think of another place where they’re a good analysis… which I suppose, does make me suspicious that I’m wrong here. But there are arguments that word classes are unique in this way: most notably, I have been arguing that word classes are best deduced in the first place from binary syntactic tests.

Also, I suppose I should consider the differences between intra-language and inter-language analyses. I note that, in my analysis of alignment systems above, the ‘continuum’ of alignments only pops into existence when you start doing typology. Within a single language, however, alignment starts to look very ‘binary’: certain arguments are either marked or not marked, certain parts of the grammar are either accusative or ergative, and so on. (Of course this is a massive oversimplification, but it covers most of the most common alignment systems) I will freely admit that the situation is very similar when looking at word classes across languages — they form an obvious continuum. This is especially blatant with the ‘adjective’ class, which ranges from practically-nouns (in English, Imbabura Quechua and Nkore-Kiga) to practically-verbs (in Mandarin, NE Ambae and Wolof). When comparing languages, prototype theory certainly looks like a good approach to defining word classes. But what puzzles me is that you seem to claim that prototype theory works for word classes within a single language, when this seems obviously false to me: English has nouns, it has verbs, it has simultaneous nouns and verbs, and it even has some minor word classes with similarities to both nouns and verbs, but I see no obvious way to analyse this profitably in terms of prototypes, from a purely emic perspective (which I consider the best perspective to adopt for word classes).

This may help. Do your feature-based analysis on English, working out purely syntactic definitions for whatever categories you can find. Now do it for Mandarin.

Now... how do you identify which of the categories you found in English are "verbs", and why do you think it's the "same thing" in some sense as verbs in Mandarin?

I don’t know Mandarin well enough to do this properly. However, I think it fair to say that figuring out why two word classes can be considered ‘the same’ is the million-dollar question in this area. I tend to agree with you on this:

Now, Chomskyans think they can still do it. They believe in universal grammar, and so a "verb" is something at the heart of a VP, which can be identified purely by the tree structure. That's... interesting. But in practice it's more or less manhandling every language to look like English, and going to a whole lot of trouble that is avoided in other theories which can directly deal with ideas like arguments and valence. And verbs.

So, I guess I'd modify my earlier stance a little. While you're analyzing a particular language's syntax, you should define categories by syntactic methods alone. But as soon as you start hooking up semantics, you can start using semantics; and when you compare languages, our best guide to using the same words the same way is semantic.

Semantics does certainly seem to be our best bet here. But, frustratingly, word classes have very vague semantics, and languages can be almost random in their assignments: I doubt you could find a single word which is in ‘the same’ word class across every single known language. (And it does seem a universal that all languages have at least nouns and verbs.) As I briefly mentioned above, prototype theory seems a better approach for defining word classes across language, especially when combined with syntax: after defining word classes across multiple languages, one can look at the syntactic behaviour of each of those word classes, identify commonalities, and from that define the syntax of the prototypical noun and verb. (I strongly suspect it would be impossible to define a ‘prototypical adjective’: those word classes are just too heterogeneous.) So yeah, I suppose I’ve ended up coming down on the Chomskyan position of identifying commonalities purely from the tree structure, although I accept that trees are going to be different across languages, hence the need for prototype theory.

I’m inclined to analyse each of ‘under $5’, ‘come to Jesus’, ‘over 70°’, ‘the morning after’ as single grammatical words (i.e. compounds), considering that I have trouble thinking of any words which could be interposed grammatically, and that these constructions show limited productivity^†. Your other examples are valid though, so the point stands.

All that being said, I still believe it should be possible to find syntactic tests (or combinations of tests) which can isolate ‘noun’ and ‘verb’ as distinct word classes, since (in English at least) they’re fundamentally syntactic concepts.

________
^† The construction in ‘under $5’ and ‘over 70°’ can really only be used nominally with those two prepositions; ‘come to Jesus’ is clearly idiomatic, given that I don’t even really understand it; and ‘the morning after’ is limited to ’after’, and practically limited to ‘the morning’ as well, given that in natural speech, that’s the only noun commonly used in this construction. If there are any counterexamples to these, I can’t think of them.

"to over about 70°", "to over what I think says 70°", etc.
"the morning before", "the day after", "the next town over", etc.

These really aren't that hard to come up with. I'm dying to know how you dismiss these.

I dismissed them because I couldn’t think of counterexamples. Now that you’ve supplied some, I can no longer dismiss those. But as I said, there were other counterexamples I never did dismiss, so this doesn’t actually affect the incorrectness of my argument.

[keenir]

But what puzzles me is that you seem to claim that prototype theory works for word classes within a single language, when this seems obviously false to me: English has nouns, it has verbs, it has simultaneous nouns and verbs, and it even has some minor word classes with similarities to both nouns and verbs, but I see no obvious way to analyse this profitably in terms of prototypes, from a purely emic perspective (which I consider the best perspective to adopt for word classes).

Wouldn't prototype theory (so long as i'm understanding it) just put those "minor word classes" and "simultaneous nouns and verbs" a bit further from the main body of exemplar nouns and verbs?...just as a prototype of a bird (defined as "feathers, wings, flies") would place crows and grouse in the main body, then penguins a bit further away, and kiwi even further still.

[Richard W]

How does "rarely" head a VP?

Compare We could argue all week and We rarely argue all week. What am I missing?

This comparison occurred to me because I'd been considering the question of whether Thai has tenses. The idiom widely consider to be the equivalent of the English perfect can be formed by suffixing the verb มา /maː˧/ 'to come' or the adjective/adverb /แล้ว˦/ 'already' (parts of speech taken from the authoritative dictionary). There is usually very little difference between the meaning of the two idioms. Thai verbs of motion often function like English directives; many so-called Thai prepositions are (also) verbs.

[zompist]

Compare We could argue all week and We rarely argue all week. What am I missing?

Interesting question! But I'd say that headship doesn't merely mean that the word is the first one in the phrase. E.g. "big brains" is an NP whose head is "brains", not "big". (I'd use the example of "the brain" or "all week", but some syntacticians go ahead and call determiners the head!)

What is the head depends on your favorite syntax theory. One consideration, though it's not very rigorous, is that an XP can be replaced by an X: "the big brains" > "brains", not "big" or "the". "John carefully argued his case" > "John argued", not "John carefully."

There's also, FWIW, theoretical efficiency. The English verbal complex is a neat little syntactic puzzle whose solution isn't intuitively obvious. Chomsky came up with a clever explanation 60 years ago, and of course it's been revised since. But the whole mess of modals, auxiliaries, and verb fits together very nicely, acts as a unit, and helped us understand other transformations. "Rarely" doesn't fit in, and I don't see what it gains us to throw it in. It seems a lot easier to treat it as an adjunct to the VP.

To reinforce that, I'd note the equivalent frame disappears when you start looking at variants:

Could we argue all week? / *Rarely we argue all week? [a possible sentence, but not a question]
We couldn't argue all week. / *We rarelyn't argue all week.
We could argue all week, and so could the CBB. / *We rarely argue all week, and so rarely the CBB.
We could be arguing all week. / *We rarely be arguing all week.

We argue all week, though rarely. / *We argue all week, though could.
Rarely is it we argue all week. / *Could is it we argue all week.
We very rarely argue all week. / *We very could argue all week.

So in short, the surface structure looks the same in your sentence, but the words don't act the same and the structure must be different.

This comparison occurred to me because I'd been considering the question of whether Thai has tenses. The idiom widely consider to be the equivalent of the English perfect can be formed by suffixing the verb มา /maː˧/ 'to come' or the adjective/adverb /แล้ว˦/ 'already' (parts of speech taken from the authoritative dictionary).

I don't know Thai, but you could make similar observations about Mandarin. You can work out parts of speech for Mandarin, but things are a lot more fluid than in English, and if you bring in the etymology it looks more confused than ever. (Most particles were originally verbs, and often can still be used in their verbal sense.)

[zompist]

We're mostly agreeing now, so I will just focus on the remaining area of disagreement.

I think I’m defining ‘verbs’ via morphology, whereas you’re doing it by syntax

Where we can do that, we should. But there are two big caveats.

First, there will be gaps. I think you pointed out one earlier, "sheep". Or "data", which has been divorced from "datum". On the verb side, verbs lose their morphology in non-finite clauses. The fact that modals lost all but their past inflection makes it harder to decide what they are. In languages like Mandarin, of course, there is no inflection to help us out. For both reasons we'd better have non-morphological tests also.

Second, in some languages morphology provides pathways, but the roots have no word class and can take any pathway the speaker likes. Quechua often works like this— e.g. kawsay means 'to live on' or 'a thing you live on'. If you inflect it as kawsanku it's definitely verbal ('they live on'), if as kawsaykuna it's definitely nominal ('things you live on'). And lots of English works this way too! (François wants to just double-mark the word in the lexicon— "V, N". Is that the only possible analysis?)

English has nouns, it has verbs, it has simultaneous nouns and verbs, and it even has some minor word classes with similarities to both nouns and verbs, but I see no obvious way to analyse this profitably in terms of prototypes, from a purely emic perspective (which I consider the best perspective to adopt for word classes).

This sounds to me like "I see exactly the variation that prototype theory would predict, but I refuse to consider a prototype explanation." I mean... how does a grid of binary features provide a better explanation? All a list of features provides is, well, a list of features. If you get a gradation in the features, it provides no explanation for that.

I'll reproduce the gradation from my book:

1) I am a dessert chef.
2) I am a baker of cookies.
3) The baking of cookies is a fine art.
4) Baking is a fine art.
5) I love to bake.
6) His baking these cookies was a rare treat.
7) Him baking these cookies was a rare treat.
8) I am baking from four to seven.
9) I baked all day.

My contention is that things get more verby as you go down the list. I'm sure you could create feature lists for each stage, but what such lists will not capture is precisely the gradation:

* 'baker' is a person nominalization, so it's referring to an activity in a way 'chef' doesn't
* noun inflection (pluralization) is lost from 3 on
* the determiner drops off at 4
* 4 allows an object (baking cookies) without the "of" that pure nouns require
* from 5 on modifications (e.g. "quickly") must be adverbs rather than adjectives. Though 4 may allow both
* possessive adjectives can be used up to 6
* 6 is the first to allow an explicit subject for the baking (in 1/2 the subject belongs to 'am')
* 7 uses 'him' for the subject, which we also see of verbs in non-finite clauses
* 8 is in the verb complex, but a) it's the same word as 6/7; b) this is a progressive, and arguably treating an event as a process is more nouny; and c) the word itself still doesn't inflect
* only 9 allows direct marking of number and tense

[Richard W]

Compare We could argue all week and We rarely argue all week. What am I missing?

Interesting question! But I'd say that headship doesn't merely mean that the word is the first one in the phrase. E.g. "big brains" is an NP whose head is "brains", not "big". (I'd use the example of "the brain" or "all week", but some syntacticians go ahead and call determiners the head!)

That makes me want to claim that argue is the head, so we completely lose the claim that the modal auxiliaries are the heads of VP phrases. This is reflected in the fact that we talk of the modals being auxiliary verbs, rather than main verbs as they once were. Admittedly, there are many remnants of their original status.

One common feature between prototypical verbs and auxiliaries is subject-verb inversion - I don't think we've noted that.

To reinforce that, I'd note the equivalent frame disappears when you start looking at variants:

Could we argue all week? / *Rarely we argue all week? [a possible sentence, but not a question]
We couldn't argue all week. / *We rarelyn't argue all week.
We could argue all week, and so could the CBB. / *We rarely argue all week, and so rarely the CBB.
We could be arguing all week. / *We rarely be arguing all week.

We argue all week, though rarely. / *We argue all week, though could.

This last comparison is invalid. The analogy goes wrong because 'rarely' has a negative sense. Your argument doesn't work well when you compare We argued all week. The transformations are different yet again, and again, we have a different structure.

Rarely is it we argue all week. / *Could is it we argue all week.
We very rarely argue all week. / *We very could argue all week.

So in short, the surface structure looks the same in your sentence, but the words don't act the same and the structure must be different.

There are also discordances between We could argue all week and We used to argue all week, which doesn't prove that used to doesn't head a verb phrase. There are also discordances with dare.

[zompist]

Interesting question! But I'd say that headship doesn't merely mean that the word is the first one in the phrase. E.g. "big brains" is an NP whose head is "brains", not "big".

That makes me want to claim that argue is the head, so we completely lose the claim that the modal auxiliaries are the heads of VP phrases.

Good point— I've been a bit loose in terminology here, but that's mainly because I don't want to assume a lot of syntax knowledge.

To be more precise, yes, "argue" is the head of the VP. But "could" is the head of a TP. Most English sentences have an NP and a TP as their immediate children.

For arguments why this is so, see any intro to syntax book, including mine. They're way too long to include here! And of course the details may vary in your favorite theory. (E.g. the TP can also be analyzed as another level of VP.)

(I've already hinted at the reason— the English verbal complex is tricky but analyzable. That means the analysis is too long for an idle linguistics posting, but it's no secret— it's been extensively discussed and reanalyzed for sixty years. Upending all that by making adverbs the head would not be a quick fix.)

We argue all week, though rarely. / *We argue all week, though could.

This last comparison is invalid. The analogy goes wrong because 'rarely' has a negative sense.

OK, fair point— I was concentrating on "rarely" being an adverb. Compare:

We heatedly argue all week. / We could argue all week.
We argue all week, heatedly. / *We argue all week, could.

Not all adverbs are the same— you can't back "rarely" like that— but that only reinforces my point that looking at a surface resemblance in just one frame doesn't prove anything.

There are also discordances between We could argue all week and We used to argue all week, which doesn't prove that used to doesn't head a verb phrase. There are also discordances with dare.

Um... yeah, syntax is complex, but so what? If you care to, do provide some reason why "rarely" should be analyzed as head of a VP (or TP). I don't see the gain, when we already know that a theory of syntax will need to explain where exactly adverbs can go (and there are multiple types and slots).