Postpositions?

[bradrn]

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.

As I see it, prototype theory is only a useful way of looking at things when there is a gradation between prototypes. If there is no gradation, and categories have rigid boundaries, then erecting discrete classes of things is much more useful for analysis. (This is hard to demonstrate with semantics, though, because practically everything there is a gradation in one way or another.)

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.

Note that I was only talking about verbs in English. Of course, morphological definitions don’t necessarily work for English non-verbs, or for non-English verbs. But for English verbs it works well enough.

Also note that I’m not using criteria like ‘verbs can always take -ing’ — as you note, that breaks down in non-finite clauses. I’m merely saying that ‘verbs can take -ing’, as in, that combination of morphemes is valid: that is, taking, being, saying are fine, but *I-ing, *mereing, *maying, *prójecting [distinguished from projécting], *foruming are not.

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?)

If ‘nouns’ and ‘verbs’ can be independently identified as distinct word classes in Quechua, and kawsay can be used in all the same ways as both nouns and verbs, and that fully describes its behaviour… then, yes, double-marking it as ‘V, N’ is the best analysis. Otherwise, we need to analyse it as a separate word class to both nouns and verbs.

As for the problem posed by affixation, I quite like Evans’s analysis in his grammar of Kayardild, a non–Pama-Nyungan language with an extraordinarily complex word class system. He separates out the notions of ‘morphological word class’ and ‘syntactic word class’, where the former is defined via morphology, whereas the latter is defined via syntax. Thus, for instance, Kayardild has a set of ‘verbal case’ affixes, which apply to nominal stems: the word they form is morphologically verbal, in that it takes verbal agreement and so on, but syntactically nominal, in that it occurs in an NP, may be modified my demonstratives and adjectives and so on:

ngada

1sgNOM

warra-jarra

go-PST

dathin-kiiwa-tharra

that-VALL-PST

ngilirr-iiwa-tharr

cave-VALL-PST

‘I went to that cave.’ [VALL = verbal case, allative]

Of course Quechua and English don’t have that level of complexity, but the basic principle — that affixation can change word class — applies to those languages also.

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.

Yes, this was exactly my point. I had yet to see any sort of continuous gradation between word classes — all you would give me was single examples of individual word classes which were a bit like other word classes.

However, I can’t really say this anymore, because…

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.

…you have now given me the example I have been asking you for.

Now, I accept this is a continuous gradation. You have even given detailed reasoning as to exactly which features are grading in/out at each point. However, I do argue with one, key, part of your reasoning: that this is a gradation in word class.

Let’s look at the actual words used here. There are only five different words used here: chef, baker, baking, bake, baked. I think we all accept that these are each of different word classes. But, if you look at your examples, these don’t seem to follow any sort of simple gradation. To take one example, baking is used in (3), (4), (6), (7), (8) — but not in (5). Bake is in (5), but you can also use it in (9) — I bake all day. If this is a continuum in word classes, it is a very discontinuous continuum!

In fact, the gradation here is not in word class, but in syntactic constructions. Having seen this collection of examples, I have no difficulty accepting that prototype theory is a good way to analyse syntax, which is above the word level. This does not imply, however, that prototype theory is a good way to analyse word classes, because one word class may be required in several entirely different constructions.

[Richard W]

Apropos one of the earlier examples, I am not convinced that monotremes really are mammals. While mammals are their closest living relatives, there are so many differences from therians that I'm not sure it is as useful as one would wish to consider them as mammals.

[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.

As I see it, prototype theory is only a useful way of looking at things when there is a gradation between prototypes. If there is no gradation, and categories have rigid boundaries, then erecting discrete classes of things is much more useful for analysis. (This is hard to demonstrate with semantics, though, because practically everything there is a gradation in one way or another.)

Hmm...so that would - ah, I goofed in the above. I put nouns and verbs as two prototypes within a shared circle, like they were two peaks of a Hawaiian island.

When actually they would each be a prototype with their own circle, and the "minor word classes" and "simultaneous nouns and verbs" would be in the grey area between them...more like a prototype pigeon and a prototype crocodile, with T.rex and Archaeopteryx in the grey between them.

Apropos one of the earlier examples, I am not convinced that monotremes really are mammals. While mammals are their closest living relatives, there are so many differences from therians that I'm not sure it is as useful as one would wish to consider them as mammals.

True, true...but sometimes, like with the five living monotremes, the one or two species of tuataras, and the various caecelians(sp), sometimes most people will gladly swallow some discomfort in how unlike they are from the main group (whether the prototype group the tuatara are compared are turtles, crocs, or even the snakes)...rather than call them their own thing. (i mean, we could say "okay, this zoo has lizards, crocs, turtles, rhynocephalians(sp), eutherians, marsupials, monotremes, and flightless birds"...but thats another matter)

[Moose-tache]

Apropos one of the earlier examples, I am not convinced that monotremes really are mammals. While mammals are their closest living relatives, there are so many differences from therians that I'm not sure it is as useful as one would wish to consider them as mammals.

There are many differences between a sparrow and a stegosaurus, but they are both dinosaurs. Luckily, linguistics doesn't have to limit itself to cladistics.

[zompist]

In fact, the gradation here is not in word class, but in syntactic constructions. Having seen this collection of examples, I have no difficulty accepting that prototype theory is a good way to analyse syntax, which is above the word level. This does not imply, however, that prototype theory is a good way to analyse word classes, because one word class may be required in several entirely different constructions.

OK, close enough. That is, I do agree that we can pretty cleanly distinguish morphology from syntax in English, and that the morphological word classes are separate. (After all, in a conlang grammar, I almost always find it useful to have a separate morphology section!) And if you can see the prototype effects in syntax, I'm happy.

One caveat though-- you have to be careful to keep your morphological assignments from infecting your syntax. E.g. as we've both noted, verbs in non-finite clauses aren't inflected. That means morphology really tells you nothing about them ("don't inflect" isn't really information). So really, you can't just assume that "bake" or even "baking" is always a verb, especially in contexts where the word cannot be inflected. That's a syntactic question.

[bradrn]

Hmm...so that would - ah, I goofed in the above. I put nouns and verbs as two prototypes within a shared circle, like they were two peaks of a Hawaiian island.

When actually they would each be a prototype with their own circle, and the "minor word classes" and "simultaneous nouns and verbs" would be in the grey area between them...more like a prototype pigeon and a prototype crocodile, with T.rex and Archaeopteryx in the grey between them.

I think about prototypes in neither of these ways, actually. Anything involving discrete ‘circles‘ is not a good visualisation of any sort of prototype model.

Instead, my mental model looks more like this:

prototypes-image.png (8.38 KiB) Viewed 6819 times

That is, a prototype model is ‘fuzzy’: it’s hard to know where each category begins and ends.

In fact, the gradation here is not in word class, but in syntactic constructions. Having seen this collection of examples, I have no difficulty accepting that prototype theory is a good way to analyse syntax, which is above the word level. This does not imply, however, that prototype theory is a good way to analyse word classes, because one word class may be required in several entirely different constructions.

OK, close enough. That is, I do agree that we can pretty cleanly distinguish morphology from syntax in English, and that the morphological word classes are separate. (After all, in a conlang grammar, I almost always find it useful to have a separate morphology section!) And if you can see the prototype effects in syntax, I'm happy.

So are you saying you agree with my reasoning behind saying that prototype theory is unhelpful for word classes? I’m finding it hard to understand your viewpoint on this issue now.

One caveat though-- you have to be careful to keep your morphological assignments from infecting your syntax. E.g. as we've both noted, verbs in non-finite clauses aren't inflected. That means morphology really tells you nothing about them ("don't inflect" isn't really information). So really, you can't just assume that "bake" or even "baking" is always a verb, especially in contexts where the word cannot be inflected. That's a syntactic question.

This is conflating levels of analysis. My word class tests are pure grammaticality judgements: ‘word X is in class Y if it can validly occur in frame Z’, where that frame may be syntactic (e.g. ‘can appear after ‘the’) or morphological (e.g. ‘can take -ing’). Such a test can have two outcomes: either the word is attested in that frame, or it is not. (Of course, one test is usually not sufficient to uniquely specify a word class; combinations of tests are needed.)

Once word classes are defined, we can now make statements such as: ‘construction C requires a word of class Y’. For instance, ‘English non-finite clauses require a verb’. The fact that this verb cannot be inflected here is completely irrelevant — all that matters is that you can take that same root, and put it through all the tests necessary for it to qualify as a verb.

In other words, I am treating word class and syntactic constructions as two entirely separate levels of grammar. Word class is a property of individual words — for each lexeme in the dictionary, there is an associated word class. Certain constructions may demand certain word types, but they cannot alter lexical properties.

Discussing this can get quite painful, actually, because we tend to use the same terminology for both these different levels. (In this way it reminds me of the discussion I had with you earlier about aspect. As I see it, using the same terminology for different things is a sure sign of a poorly-analysed prototype system!) When we say, for instance, ‘the word bake is a verb’ — what does that mean? It means that we can put the word bake through a bunch of syntactic and morphological tests, and find that it is a verb rather than any other word class. For instance, we might see if it can appear with ‘-ing’: it does, in that baking is a valid word. (R.M.W. Dixon notes that this is in fact a sufficient criterion to identify an English verb.) Thus, we feel entitled to write in the dictionary: ‘bake, v.’.

[Incidentally, a note on these binary tests, and exactly how many are necessary. Of course, each individual word class can appear in a lot of places: for instance, English nouns can occur as the subject of a verb, as the object of a verb, as the argument of a preposition, as the argument of an article, as something modified by a verb, as the modifier of another noun, and so on. Similarly, English verbs can occur as the head of a VP, as something which may take ‘-ing’, as something which can occur with a subject, as something which may take ‘-ed’, and so on. The trick is to find the minimal set of tests which will uniquely distinguish every single word class in the relevant language. Once we have done that, we can now say: these are the tests which must be assessed in order to identify the class of a given lexeme. The other tests, then, turn out to be merely derivatives of the word class identification: syntactic constructions which merely require a given word class, instead of defining it.]

But now, notice that we also talk about word classes in a completely different way: where we insist that baking is ‘less verby’ in the baking of the cake than in I was baking the cake, or that napkin is ‘more nouny‘ in I took a napkin than in napkin-making is fun. By the definition of ‘word class’ given above, this obviously makes no sense. After all, the word class of baking is a property of the word itself — it can’t change from clause to clause! But really, what we’re now doing is a totally different thing: we are taking the properties of the constructions in which a certain word class may appear — ‘can occur after the’ and so on — flipping them around, and judging other constructions based on how similar they are.

In this way of thinking, instead of using ‘word class’ as an attribute of, well, words, we are using it as an attribute of constructions: based on the syntax of this phrase, what word class would we expect it to require? Does it have more similarities to those constructions in which appear nouns—or adjectives—or verbs? This, I think, is to some extent what you have been meaning when you talk about ‘word classes’. And of course, since here we’re looking at similarities rather than exact equivalences, the idea of prototype theories came naturally.

For a practical example of what I mean, I’ll have another look at your gradation:

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.

When I first saw this, I noted that you were describing constructions rather than words. I was a bit perplexed in understanding how the same word could reappear several times, yet you assign a different ‘word class’ to each appearance — (3) is apparently ‘less verby’ than (8), despite having the same word in both.

With the above-described mode of analysis, the principles involved here become clear. (Not that they weren’t all along! I understood what was happening; I just didn‘t have the terminology to describe it coherently.) In (3), we are not focussed so much on the word baking as on the construction, The ____ of cookies is a fine art. Here, the underscored part is in a construction with numerous similarities to other constructions in which nouns are found: it occurs after the, and it occurs before of. For these reasons, we can call it ‘very nouny’ — even though it actually requires a gerund! By constrast, (8) has the structure I am ____ from four to seven. This one has more similarities to constructions where verbs occur: the underscore occurs after an auxiliary, and has a subject. And yet it takes a gerund, not a verb! Note that the class of the underscore never varies — it’s always a gerund, and nothing can change that, because synchronically, the class of a single lexeme is invariant. But the constructions in which that invariant lexeme is used can vary, from constructions prototypically ‘associated with nouns’, to constructions prototypically ‘associated with verbs’.

[zompist]

In fact, the gradation here is not in word class, but in syntactic constructions. Having seen this collection of examples, I have no difficulty accepting that prototype theory is a good way to analyse syntax, which is above the word level. This does not imply, however, that prototype theory is a good way to analyse word classes, because one word class may be required in several entirely different constructions.

OK, close enough. That is, I do agree that we can pretty cleanly distinguish morphology from syntax in English, and that the morphological word classes are separate. (After all, in a conlang grammar, I almost always find it useful to have a separate morphology section!) And if you can see the prototype effects in syntax, I'm happy.

So are you saying you agree with my reasoning behind saying that prototype theory is unhelpful for word classes? I’m finding it hard to understand your viewpoint on this issue now.

No, I was saying that I was happy that you seemed to accept my application of prototypes to syntax! From the rest of your post it seems we don't agree as closely as I thought.

In English morphology, we can use the traditional word classes pretty well, without much fuzziness. This neat division isn't guaranteed for other languages! As discussed above, adjectives can morphologically be more nouny or verby, likewise infinitives. Participles have a way of turning into something else. And in some languages it's more helpful to talk about verbal vs. nominal pathways rather than verbs and nouns. So I don't think you can keep prototypes out of morphology, even if that mostly works for English. And I think you've already admitted that you can't keep prototypes out of typology.

And you shouldn't want to! Explaining patterns is more interesting than just listing them.

My word class tests are pure grammaticality judgements: ‘word X is in class Y if it can validly occur in frame Z’, where that frame may be syntactic (e.g. ‘can appear after ‘the’) or morphological (e.g. ‘can take -ing’).

This is where you lose me again. You say below that you are "treating word class and syntactic constructions as two entirely separate levels of grammar", but you want to define word class with both morphological and syntactic tests. At the same time you seem to admit that prototype theory works with syntax. Maybe you should take some time to think this out...

For instance, we might see if it can appear with ‘-ing’: it does, in that baking is a valid word. (R.M.W. Dixon notes that this is in fact a sufficient criterion to identify an English verb.) Thus, we feel entitled to write in the dictionary: ‘bake, v.’.

Lexicography does not trump linguistics. But if we're looking at dictionaries, mine has "baking" as a noun, and also lists "baking" as a form of the verb "bake".

Let's take an simpler example. Is "love" a noun or a verb? The lexicographers say it's both. How about an inflected form, "loves"? Also both. I can't believe that this really confuses you, and yet you want to call "baking" the same word class wherever it occurs.

When I think about it, I don't agree that you can label words in the lexicon, and then treat syntax as an entirely separate realm. In generative grammar, people often treat morphology as a dull, solved problem— the real action is in syntax and the "phonological component" just turns the syntactic tree into sounds by a simple algorithm. I'm parodying a bit— morphology is more interesting than that. But I guess I'm not willing to turn the definition of "noun" and "verb" over to the morphologists, or the lexicographers. The gradation I presented is part of English morphosyntax and is closer to a description of how nouns and verbs work in English than a listing of paradigms.

Edit: Trying to be clearer... if we were just starting to analyze English, then the morphology is a good place to start, and form preliminary categories. Those paradigms are part of English grammar. But English can't be treated like Whitney treated Sanskrit: explain the phonology, then sandhi, then the morphology, then send the thing to the printer. You haven't explained English nouns and verbs until you've explained English morphosyntax, including the graded behavior I described. You can't even create a full paradigm of English verbs without including syntax. (Really this is true of Sanskrit and Latin too; you just sneak in the participial constructions at the end of the paradigm and hope no one notices that they're syntax.)

As it happens, morphology is concentrated at the two ends of that gradation. There's probably some lesson in that— that is, it's not coincidence that this nounishness-verbishness gradation happens to be correlated with morphological complexity. But we'd have to look at a bunch of languages to investigate that.

[keenir]

Hmm...so that would - ah, I goofed in the above. I put nouns and verbs as two prototypes within a shared circle, like they were two peaks of a Hawaiian island.

When actually they would each be a prototype with their own circle, and the "minor word classes" and "simultaneous nouns and verbs" would be in the grey area between them...more like a prototype pigeon and a prototype crocodile, with T.rex and Archaeopteryx in the grey between them.

I think about prototypes in neither of these ways, actually. Anything involving discrete ‘circles‘ is not a good visualisation of any sort of prototype model.

not discrete - concentric, like a topography map.

Instead, my mental model looks more like this:
prototypes-image.png
That is, a prototype model is ‘fuzzy’: it’s hard to know where each category begins and ends.

Okay, now I'm confused. If everything is an exception or "only kinda-sorta" (quoting from somewhere else, months ago), then do we really have a prototype...or even have a solid definition?

Yes, there are exceptions to "birds have wings and feathers" and "houses have floors and walls" -- but don't we need a central definition to agree upon, before we can have exceptions?
{ie, should Medieval Europeans have said the Birds of Paradise aren't birds because they didn't have feet? I certainly don't want to buy a house that has floors, but all the walls are windows}

[zompist]

That is, a prototype model is ‘fuzzy’: it’s hard to know where each category begins and ends.

Okay, now I'm confused. If everything is an exception or "only kinda-sorta" (quoting from somewhere else, months ago), then do we really have a prototype...or even have a solid definition?

Yes, there are exceptions to "birds have wings and feathers" and "houses have floors and walls" -- but don't we need a central definition to agree upon, before we can have exceptions?

There is some confusion here, which I probably contributed to by talking about fuzzy categories. You're both wrong.

Prototypes don't mean partial category membership. The idea is not that robins are 100% bird, but penguins are only 50% bird. They're both birds. Prototypes mean that robins are a better example of what birds are.

To be precise, there are a cluster of consequences. People think more easily of the prototypes. They can grade category members on closeness to the prototype. They respond faster to questions about prototypes. They reason about the category using info about the prototype. Prototypes are likely to be learned first. Prototypes act, well, as we expect the class to behave, without weird outlier behavior. If you think of a visual image of the class, you're likely to focus on the prototype.

Boundaries may be precisely defined... or they may not. Take "hot" vs. "cold". Is there a precise temperature cutoff? Nah, and we don't really need one. "Chair" however has (I would say) clear boundaries— if you can't sit on it, it's not a chair.

Prototype effects definitely apply to syntactic categories— if I ask you for some nouns, you're likely to come up with "dog" or "tree" before you suggest "fluorescence" or "anger" or "NATO". Whether syntactic categories have firm boundaries is another question...

In our culture we're brought up to think that words are defined by dictionaries. And technical terms are! You don't know the word "radius" unless you can pretty much reproduce the dictionary definition. But languages can get by just fine without dictionaries, and common words are learned by association with prototypes. (That is, you show your toddler a dog, you don't read them the dictionary definition.)

[KathTheDragon]

"Chair" however has (I would say) clear boundaries— if you can't sit on it, it's not a chair.

There are many things you can sit on that aren't chairs, though.

[Raphael]

And "chairperson" seems to be shortened to "chair" sometimes.

"The chair objects to the gentleman from Illinois's insinuation that a person can sit down on the chair!"

[zompist]

"Chair" however has (I would say) clear boundaries— if you can't sit on it, it's not a chair.

There are many things you can sit on that aren't chairs, though.

Sure, but is someone confused by whether they're chairs or not? You can sit on a car, but no one calls it a chair.

Of course, words can carve out a space for themselves-- we probably won't call a "sofa" a chair, we'll call it a sofa.

I don't think "chair" is a particularly difficult word, but maybe I'm wrong. Compare "sandwich": it's a thing to have jocular arguments over whether something is a sandwich or not.

[zompist]

And "chairperson" seems to be shortened to "chair" sometimes.

"The chair objects to the gentleman from Illinois's insinuation that a person can sit down on the chair!"

I think those are two separate senses, in fact. "The chair objects..." is an old metonomy like "the throne" or "the crown." "Chair" from "chairperson" is much newer.

But yeah, words have multiple senses. (This is in fact extensively discussed in prototype theory.)

[bradrn]

In English morphology, we can use the traditional word classes pretty well, without much fuzziness. This neat division isn't guaranteed for other languages!

I suppose I can’t argue with this. All I can say is that I’ve yet to see a language where the best analysis uses fuzzy word classes, along the lines of ‘this is a list of words which grade between such-and-such prototypical word classes, and their behaviour can’t be discretised sensibly’. Word classes with concrete boundaries seem like a pretty good description for most languages.

As discussed above, adjectives can morphologically be more nouny or verby, likewise infinitives. Participles have a way of turning into something else. … I think you've already admitted that you can't keep prototypes out of typology.

Sure. But these are inter-language comparisons, and as I’ve already intimated, prototype theory tends to excel at those. Applying it to the morphosyntax of a single language is much harder.

And in some languages it's more helpful to talk about verbal vs. nominal pathways rather than verbs and nouns.

I don’t understand what you mean by this. Could you elaborate please?

My word class tests are pure grammaticality judgements: ‘word X is in class Y if it can validly occur in frame Z’, where that frame may be syntactic (e.g. ‘can appear after ‘the’) or morphological (e.g. ‘can take -ing’).

This is where you lose me again. You say below that you are "treating word class and syntactic constructions as two entirely separate levels of grammar", but you want to define word class with both morphological and syntactic tests. At the same time you seem to admit that prototype theory works with syntax. Maybe you should take some time to think this out...

This is why I insisted earlier on strictly pre-theoretic tests, along the lines of ‘this word can appear after the’ or ‘this word can take -ing’. This allows us to define word classes without reference to any sort of syntactic analysis. Then, those word classes may be used as the primitives on which we base further analysis of syntax.

Let's take an simpler example. Is "love" a noun or a verb? The lexicographers say it's both. How about an inflected form, "loves"? Also both. I can't believe that this really confuses you, and yet you want to call "baking" the same word class wherever it occurs.

I accept that ‘love(s)’ is both a noun and a verb. However, there is a vital difference between ‘love(s)’ and ‘baking’. For the former, we already have defined word classes which we call ‘nouns’ and ‘verbs’. If we now compare the distribution of ‘nouns’ and ‘verbs’ to that of ‘love(s)’, we notice some interesting things. ‘Love(s)’ can be used anywhere a ‘noun’ can; it can be used anywhere a ‘verb’ can; and it is disallowed wherever both ‘nouns’ and ‘verbs’ are. That is, the set of allowable positions of ‘love(s)’ is a perfect union of the set of allowable positions of ‘nouns’ and those of ‘verbs’ — so we say that ‘love(s)’ has a word class of noun+verb.

Note that I do not assume a priori that any particular word, such as ‘love(s)’, is a member of two different word classes. I look at its distribution, note that it is exactly described by the union of two other word classes, and from that draw the obvious conclusion. My reasoning with regards to ‘baking’ is exactly parallel: I assume it is its own word class, unless there is a better description. In this case, there isn’t any — its distribution can’t easily be described as a union of that of any other set of English word classes. So, for lack of evidence to the contrary, I assume that it makes up its own, separate word class.

When I think about it, I don't agree that you can label words in the lexicon, and then treat syntax as an entirely separate realm. In generative grammar, people often treat morphology as a dull, solved problem— the real action is in syntax and the "phonological component" just turns the syntactic tree into sounds by a simple algorithm. I'm parodying a bit— morphology is more interesting than that. But I guess I'm not willing to turn the definition of "noun" and "verb" over to the morphologists, or the lexicographers. The gradation I presented is part of English morphosyntax and is closer to a description of how nouns and verbs work in English than a listing of paradigms.

I really don’t understand your point here. Wherever did I say that syntacticists are disqualified from figuring out word classes?

Edit: Trying to be clearer... if we were just starting to analyze English, then the morphology is a good place to start, and form preliminary categories. Those paradigms are part of English grammar. But English can't be treated like Whitney treated Sanskrit: explain the phonology, then sandhi, then the morphology, then send the thing to the printer. You haven't explained English nouns and verbs until you've explained English morphosyntax, including the graded behavior I described. You can't even create a full paradigm of English verbs without including syntax. (Really this is true of Sanskrit and Latin too; you just sneak in the participial constructions at the end of the paradigm and hope no one notices that they're syntax.)

As it happens, morphology is concentrated at the two ends of that gradation. There's probably some lesson in that— that is, it's not coincidence that this nounishness-verbishness gradation happens to be correlated with morphological complexity. But we'd have to look at a bunch of languages to investigate that.

I think you were misled by the fact that, for an example of a word class test, I happened to choose a morphological property. That will not always be the case. The sort of definitions I’m looking for are exemplified by Dixon’s definitions of English word classes (Basic Linguistic Theory vol. 1):

[verbs], takes suffix -ing
[nouns], may be immediately preceded by an article and need not be followed by another word
[adjectives], may be immediately preceded by an article and is then followed by another word (either one from class [nouns] or another word from class [adjectives])

One might quibble with the exact definitions, but that’s the style of thing I’m looking for — simple tests to separate words into different categories, each of which is homogeneous in its syntactic and morphological behaviour.

Prototypes don't mean partial category membership. The idea is not that robins are 100% bird, but penguins are only 50% bird. They're both birds. Prototypes mean that robins are a better example of what birds are.
…
Boundaries may be precisely defined... or they may not. Take "hot" vs. "cold". Is there a precise temperature cutoff? Nah, and we don't really need one. "Chair" however has (I would say) clear boundaries— if you can't sit on it, it's not a chair.

I’d say that there are very few, if any, words with 100% clear boundaries. (Even in technical fields.) But I accept the point.

Prototype effects definitely apply to syntactic categories— if I ask you for some nouns, you're likely to come up with "dog" or "tree" before you suggest "fluorescence" or "anger" or "NATO". Whether syntactic categories have firm boundaries is another question...

Sure, but this is a totally different thing to what you were arguing earlier. Here, you say that the individual examples of word classes may be more or less prototypical — and I agree with this! But, as I understand it, earlier your argument was that the word classes themselves can act as prototypes for other word classes — fuzzy prototypes, no less! — which I disagree with.

[Raphael]

(Going a bit OT)

But yeah, words have multiple senses. (This is in fact extensively discussed in prototype theory.)

Isn't that more a matter of familial resemblance theory than prototype theory? (Don't get me wrong, I generally prefer both of those to dictionary definitions.)

[zompist]

But yeah, words have multiple senses. (This is in fact extensively discussed in prototype theory.)

Isn't that more a matter of familial resemblance theory than prototype theory? (Don't get me wrong, I generally prefer both of those to dictionary definitions.)

Well, these aren't contradictory. Lakoff's Women, Fire, and Dangerous Things in fact starts with Wittgenstein; he considers him a precursor to cognitive linguistics. But Lakoff would probably want to insist that cognitive linguistics is an advance on Wittgenstein.

[Raphael]

Ah, thank you; I mainly know about either theory from your books, and I didn't know how closely related they are.

[zompist]

All I can say is that I’ve yet to see a language where the best analysis uses fuzzy word classes, along the lines of ‘this is a list of words which grade between such-and-such prototypical word classes, and their behaviour can’t be discretised sensibly’. Word classes with concrete boundaries seem like a pretty good description for most languages.

OK, take a look at Classical Chinese. There's no morphology[*[, so you're entirely reliant on syntax. And individual words are really resistant to neat labeling. It's not lawless, of course. But it would be a severe test of your claim that word classes can be easily defined and have no exceptions or prototype effects.

[*] There probably was in pre-Chinese, but that's another discussion.

As discussed above, adjectives can morphologically be more nouny or verby, likewise infinitives. Participles have a way of turning into something else. … I think you've already admitted that you can't keep prototypes out of typology.

Sure. But these are inter-language comparisons, and as I’ve already intimated, prototype theory tends to excel at those. Applying it to the morphosyntax of a single language is much harder.

I've done just that for English morphosyntax. But I'll put this down to sloppy terminology; what I think you mean is your word classes.

Anyway, we seem to agree that word classes in typology exhibit prototype effects. But you want to deny them when looking at a particular language. I find that a bit odd— it's like saying that evolution is true for mammals, but not for otters. The overall theory should be true even for smaller instances or time depths, and can even point us to specific insights. It just won't be as spectacular.

And we do use typology. We're using it when we call something in a language a verb, or an aspect, or a dative. There's nothing wrong with saying that Latin adjectives are very nouny. (Well, you'd dress it up in more formal language.)

This can also be shown by expanding the time depth a little. E.g. much about English verbs and modals becomes clear if we understand that both were clear verbs a few centuries ago. The modals retained certain behavior of the older class (direct negation, question inversion). The "real" verbs lost that behavior, but retained inflections that the modals lost.

There are some good reasons to avoid history and typology at some stages of analysis. But we can look at those reasons and address them directly. The main one is that both can lead us to make dubious assumptions— e.g. "possessive s used to be a genitive case marking on nouns, so it still is", or "Because there's a Latin infinitive amare, the translation 'to love' is the English infinitive". But the lesson to draw is not "never look at history or typology", but not to make those types of declarations.

An example: consider the expression "Waste not, want not". From a synchronic point of view, this is worrying: are these modals? a new word class? But knowing the history makes it easier to decide that this is just an unproductive idiom, a quotation from an earlier era.

And in some languages it's more helpful to talk about verbal vs. nominal pathways rather than verbs and nouns.

I don’t understand what you mean by this. Could you elaborate please?

We talked about this before, e.g. the discussion of Quechua kawsay. Basically: in some languages, at least some roots can be inflected as nouns, or verbs, or something else.

You suggested that they be marked {N, V}. And maybe that's OK! But another possibility, especially if this is really common in a language, is that roots aren't specified for word class in that language. We'd have to see what makes for a better grammar. But I would suggest if (say) 75% of your lexicon is marked {N, V}, you'd be missing a generalization and improperly applying methods only suited to other languages.

This is why I insisted earlier on strictly pre-theoretic tests, along the lines of ‘this word can appear after the’ or ‘this word can take -ing’. This allows us to define word classes without reference to any sort of syntactic analysis. Then, those word classes may be used as the primitives on which we base further analysis of syntax. [...] My reasoning with regards to ‘baking’ is exactly parallel: I assume it is its own word class, unless there is a better description. In this case, there isn’t any — its distribution can’t easily be described as a union of that of any other set of English word classes. So, for lack of evidence to the contrary, I assume that it makes up its own, separate word class.

On "baking" in particular, you haven't actually given a reason to analyze it as one word; often people divide it into a more nouny word (gerund) and a more verby one (participle).

I understand what you're trying to do; my points all along have been a) it's not going to be as easy and ironclad as you think, and b) you're over-emphasizing the morphology.

Let's step back and pretend that we could analyze English from scratch, keeping the bins of morphology, syntax, and semantics hermetically separate. What do we get?

1. Morphology gives us a set of word classes— call them noun_m, verb_m, etc. It also leaves us a huge bin of uninflectable words.
2. Syntax gives us a much more complicated picture, with a much smaller trash bin (probably called "particles"). Part of this picture is the gradation I described; we can call the poles noun_syn and verb_syn.
3. Semantics gives us fuzzy clouds of things, but it does offer us a prototype agent— a body, probably human, which persists in time and does stuff— and a prototype action— a definite event in time which changes the objects around it, probably directed by an agent. We can call these noun_sem and verb_sem.

Now what is the relationship between these entities? Is noun_m = noun_syn = noun_sem? Or are they completely unrelated? Or related in a complex way?

I'm not quite sure what your answer would be, but it seems like you're expecting noun_m to be more basic (you called it a "primitive") and to drive the analysis of syntax. That's where I disagree.

What I was trying to say earlier is that noun_m / noun_syn / noun_sem are not the same thing, indeed can be frustratingly distinct— but that they are correlated in very interesting ways, which are best explained by prototype theory.

E.g., the "most nouny" frames in my graded list of examples happen to be precisely the situations where nouns are most inflected. The most verby frames are those where the verbs are most inflected. I don't think that's coincidence.

Also, the most semantically agentive situations correlate with noun inflection. E.g. in IE, m/f nouns but not neuters have a separate accusative. In Russian, there are subcases for animate nouns. In English, we still have 3s objective pronouns for animates but not inanimates. English mass nouns— less likely to be agents— lack a plural. English modals— which are not prototypical actions— lack number inflection. In many languages active verbs (closer to the prototypical action) have more aspect distinctions than stative verbs.

(This is not supposed to be a theory of everything... a lot of morphology is just weird for no known reason.)

[Moose-tache]

OK, take a look at Classical Chinese. There's no morphology[*[, so you're entirely reliant on syntax. And individual words are really resistant to neat labeling. It's not lawless, of course. But it would be a severe test of your claim that word classes can be easily defined and have no exceptions or prototype effects.

To give some illustration of what Zompist is saying, words in Old/Classical Chinese, much like English, had a lot of lexically determined notes about how they could be used and what connotations they acquired when used this way or that way. If you treated this as membership in a series of word classes, you'd need a lot of classes.

For example, 利 (profit) and 王 (king) were both nouns. As in English, they could be enverbled with impunity. But 利 as a verb meant "to consider something profitable," while 王 as a verb means "to be crowned." So do we have two word classes, one Yidong and one Shidong, to use the emic terminology? Or do we just have nouns with asterisks all over them?

EDIT, unrelated: I try to bring this up whenever I can, since it tends to get forgotten very easily, but English "possessive s" is not a direct descendant of the genitive case on nouns. It is in fact a remnant of the genitive case on pronouns! That's why it's written with an apostrophe; it's a contraction of "[noun] his." This also explains why it behaves like a clitic and not a case suffix; it would have originally intervened as the new head of the noun phrase. So like you said, history helps us understand why English syntax is the way it is.

EDIT, on topic this time: regarding the argument about "baking," it might be useful to point out that these are homonyms. English conflated two separate endings a thousand years ago, hence the identical pronunciation of the gerund and participle. But we could, if we really wanted to, treat them as two coincidentally identical forms. We've all made peace with the fact that English subjunctive verbs are usually identical to nominal derivations of those verbs ("We ask that he leave" vs "He is on leave"), without scratching our heads about how we reconcile these disparate uses. If we even insist on deciding whether "baking" and "baking" are one word or two in the first place, that is. Personally I'm not sure that question will help given the number of assumptions we'd hjave to agree on before we could get there.

[Estav]

EDIT, unrelated: I try to bring this up whenever I can, since it tends to get forgotten very easily, but English "possessive s" is not a direct descendant of the genitive case on nouns. It is in fact a remnant of the genitive case on pronouns! That's why it's written with an apostrophe; it's a contraction of "[noun] his." This also explains why it behaves like a clitic and not a case suffix; it would have originally intervened as the new head of the noun phrase. So like you said, history helps us understand why English syntax is the way it is.

I believe that account of the ultimate origin of -’s is currently considered discredited. While there is evidence of it being analyzed in some time periods as “his” (by being spelled that way, often only when it is already syllabic due to being after a sibilant) and that might be the reason why the apostrophe ended up being used, I think the evidence as a whole is taken to show survival of the Old English inflectional affix up to the point where it turned into the modern morpheme.