Linguistic Miscellany Thread

[Salmoneus]

I wasn't trying to promulgate jargon - I don't know how widely used those phrases are as technical terms - just to make a common-sense conceptual distinction in plain words.

If your language distinguishes two types of words distinguished by how they can be used, then clearly you have lexical differences in word class, and syntactic differences in word class.


I suppose my distinction can be summed up in the claim: most languages have different sets of words that are used in different ways from one another. This is only true of a language if a) the language distinguishes different ways to use a word (syntactic classes) and b) those different ways of using words require different sets of words.

The point being, even a language that genuinely did not have different sets of words, but could use all words in all ways - i.e. had limitless and neutral zero-derivation - would still probably use those words in a discrete and largely-distinguishable set of ways.

Or to put it another way: there's a difference between whether you can distinguish nouns and verbs in the dictionary and whether you can distinguish nouns and verbs in sentences. If you can do the first, you can do the latter, but the opposite is not necessarily true.

(though many people think it is true in practice for all actually-known human languages).

[zompist]

Or to put it another way: there's a difference between whether you can distinguish nouns and verbs in the dictionary and whether you can distinguish nouns and verbs in sentences. If you can do the first, you can do the latter, but the opposite is not necessarily true.

I'm struggling to understand your concept of a "lexical class." I'm not sure how to translate it to linguistic terminology. The above statement seems closest to an operational definition, though.

The immediate problem is that dictionaries are not data; they are themselves the product of a linguistic analysis. Something about the language has to tell you what to write in the dictionary!

(The rest of this is not really a reply to Sal, though it may help explain why I don't grasp his term. It's intended to be generally helpful, but I understand it may explain too much to some people, and not enough for others.)

Traditionally, there's three approaches:

1. Semantic; e.g. nouns are "people, places, and things." Linguists usually fervently reject this, for good reasons: a) lots of words don't fit the definitions; b) it's IE-centric; c) it's of no use when it comes to minor categories. On the other hand, the idea might be rescued by prototype analysis, which does not require that all members of a class fit a "definition". Anyway, if you do a proper syntactic analysis, you usually end up with a couple categories that sure look like nouns and verbs, and what else are you going to call them?

2. Morphological: e.g. in Quechua, something with a case ending must be a noun; something with a tense must be a verb. This can't always be done, or can only be done in part (e.g. in Mandarin, plural -men applies only to pronouns and nouns), but it's nice when it happens.

3. Syntactic: a word can only occur in certain frames. E.g., in Mandarin, "adjectives" don't pattern like nouns, but like verbs— they can take adverbials, they can be perfective. And "verbs" can be used in comparative constructions. (There may be something that distinguishes verbs from adjectives, but I can't think of it offhand.)

The analysis you get from (2) or (3) is usually similar. But not always— e.g. Slavic or Hindi verbs often look like adjectives, because they're derived from participles. It's not clear a priori what an "infinitive" is— in (say) French we usually treat it as a verb form, but in Quechua it's a noun and takes case endings.

Zero-derivation can make it very hard to say what class the root belongs to. I'm not sure the question makes sense, unless we're smuggling in a semantic definition. Dictionaries of roots are kind of hard to use, for Westerners— so, for instance, this Quechua dictionary here has entries pukllay 'play' and pukllana 'toy'. These are very obvious and easy derivations from a root puklla-, but it would be unhelpful to organize the dictionary that way... you'd have to give the derived words anyway just so English speakers can understand it. But a Quechua-Aymara dictionary could be organized by root.

(In Quechua, like English, just one inflection usually tells you what morphosyntactic category you've got. Languages like Nuuchahnulth are far less tractable— you can easily throw on tense or modal affixes, yet end up with a noun.)

[Salmoneus]

Or to put it another way: there's a difference between whether you can distinguish nouns and verbs in the dictionary and whether you can distinguish nouns and verbs in sentences. If you can do the first, you can do the latter, but the opposite is not necessarily true.

I'm struggling to understand your concept of a "lexical class." I'm not sure how to translate it to linguistic terminology. The above statement seems closest to an operational definition, though.

The immediate problem is that dictionaries are not data; they are themselves the product of a linguistic analysis. Something about the language has to tell you what to write in the dictionary!

...yes, and I'm saying that it's lexical class that tells you what to write in a dictionary.

More technically, lexical class is a property of words qua lexical items (the things you try to list in a dictionary). Syntactic class is a property of words qua instances in an utterance (I'm assuming we're treating morphology as part of syntax in this instance). Or, another way, I'm talking about the difference between words as types (lexical items) and words as tokens (instances of those items occuring in syntax).

But to be less technical: you use the concept of "frame", which is very closely related.

Within a sentence (/corpus of sentences, etc), it is possible to identify different syntactic functions. If I say "The dog kalishes the cat every wednesday", you can identify that 'kalish' is functioning as a verb. There is, as you say, a particular frame here, and from the fact it fits in that frame, you know what function the word has, given the syntax of the language.

So having identified two different functions, you can then ask if that implies two different lexical sets, one for each function. Or, if you prefer, whether there are two different lexical sets, one appropriate for each (function-relevant) frame.

In English, "ocelot" is a noun - it fits in the noun frame. "Accelerate" is a verb - it fits in the verb frame. "Harbour", however, can fit in EITHER frame.

If we imagine a language in which ALL nouns are like 'harbour', and can equally easily fit in either the verb frame or the noun frame, we could say that the language does not have a lexical distinction (or a 'distributive' distinction if you prefer) between nouns and verbs. There would be no words - in the type sense - that would be characteristically 'nominal' or 'verbal'.

However, that would not mean that the language did not distinguish nouns and verbs at the syntactic level - that is, individual words as tokens could still be identified as functioning as nouns or verbs.

Take these English sentences:
The captain harbours bears
Harbours bear captains
The bear captains the harbour

In each sentence, we have no difficulty identifying which words are functioning as nouns and which are functioning as verbs. English has syntactic rules that distinguish between these two functions, and by seeing which rules are being applied to which words (here related to word order, articles, and the limited morphology of -s), we can tell which are the verbs. And yes, you can use frames as tests for this. Within these sentences, a distinction can be made between two functions, one prototypically nominal and one prototypically verbal.

However, all three substantives here can be quite naturally placed in either frame. Therefore there is no 'distributional' basis on which to call them specifically nouns or specifically verbs. So dictionary makers either have to list them as both, or list a pair of homophones for each.

Now, in some languages, it's been argued that all substantives are like this. That is, there is no word - as type, as item in a hypothetical lexicon - that can be specified as being only a verb or only a noun - all substantives can equally well fit in either frame. This is what I mean by lacking "lexical class" - there is no such thing in these languages, purportedly, as a 'class' that can be assigned as a lexical property (i.e. a property of the word qua type).

But as we can see from the English, even when the type of the word is distributionally noncategorical, the token of the word can still be assigned a syntactic category or function: that is, even if the word can fit in either frame, you can still say which frame it's fitting in in a given instance. And you can still do that even if every type-word is distributionally noncategorical (although, to be sure, a scholastic argument then arises about which frame-set should be called 'nominal' and which 'verbal', easily resolved in practice, one would assume, by resort to semantics - although one could if one really insisted just number them or something instead).


So crucially that discussion, about whether a language has distributional categories as properties of lexical types (what I've been calling 'lexical class'), is different from the discussion about whether a language has syntactic categories as properties of lexical tokens (what I've been calling 'syntactic class'. Although, of course, if a language doens't have any syntactic categories to begin with, it can't have distributional categories either, since it's the syntactic categories that types are distributed to (/in).

However, there can be questions about whether there are syntactic categories too. For instance, in some languages there's allegedly free zero-derivation between verbs and agents of verbs, coupled with zero copula. So if we imagine a sentence "John xan fish", where we know this sentence conveys the idea that John eats fish, we may not know whether to call 'xan' a verb or a noun in this instance (should we, in English paraphrase, say that this is more like "John eats fish" or like "John is an eater of fish"?) In other words, there is a frame here, but we don't know whether it should be grouped with the noun-identifying frames or the verb-identifying frames. Now, in a language where there are no distributional categories, we can imagine that a great many frames have this ambiguity, to the extent that we cannot really divide out two different sets of frames and call one an indicator of verbs and one an indicator of nouns. In the same way that, more prosaically, we can't distinguish adjectives as separate from verbs.

Whether any languages lack a syntactic distinction between nouns and verbs is much more controversial than whether some may lack a distributional distinction between nouns and verbs. It's apparently relatively accepted that many Austronesian languages have no distributional distinction between nouns and verbs (or adjectives in many cases) (i.e. no lexical classes 'noun' and 'verb' distinct from one another), but they probably do have distinct nominal and verbal syntactic categories (though this has also been questioned in some cases) - that is, in many if albeit not in all sentences, it's possible to identify some words as acting as 'verbs' and others as 'nouns' in that sentence, even though there are no distributional word classes.



I suppose I still haven't explained it well enough, but I think that's the best I can do.

[KathTheDragon]

Is it really necessary to posit limitlessly productive zero-derivation to explain when an entire class of words shows more than one distinct syntactic usage? Why not just accept that the traditional categories of part of speech just don't fit every language, and that what you really have is one uniform class that can be used in two distinct manners?

[Salmoneus]

Is it really necessary to posit limitlessly productive zero-derivation to explain when an entire class of words shows more than one distinct syntactic usage? Why not just accept that the traditional categories of part of speech just don't fit every language, and that what you really have is one uniform class that can be used in two distinct manners?

Well, that's what I said, so...

[zompist]

OK, we agree on syntactic categories, so that part's cool.

I don't see that the lexical categories exist, or are needed.

In some languages, the morphosyntax tells you how to label the lexeme, so we're just using the (morpho-)syntactic category.

To use my example, nothing tells us if Quechua puklla- "is" a noun or a verb. But I don't see that there's a question to answer. We can list the root as a lexeme, or we can list two roots... it's a lexicographic question, not a linguistic one.

Or to put it the other way, tokens of "ocelot" are always syntactically nouns. But there's no need to posit an additional claim that it's a "lexical noun." The concept "always a syntactic noun" is all we need (so far as I can see). Labeling it 'n.' in the dictionary is a shorthand for this.

[zompist]

This may be a side issue, but it's kind of interesting to look at Latin for instances of N/V ambiguity. Latin is almost the poster child for words being very clearly one part of speech or another. But it's not at all hard to find roots that can be used for either nouns or verbs, or adjectives or verbs:

altern- (be) alternate
caed- fell(ing)
cant- sing, song
cav- hollow
cens- count
claud- limp(ing)
color- color
coqu- cook
cumul- heap
cur- care
don- present
dur- hard(en)
duc- lead (cf. dux/ducis)
festin- hasten/hasty
fid- trust
foed- foul

Any given morph is going to be more determined, but that's true of Quechua too.

Of course, it's more common to have some kind of derivational suffix— conjuro, conjuratio; claro, claritas, etc.

[TomHChappell]

.... there's a difference between whether you can distinguish nouns and verbs in the dictionary and whether you can distinguish nouns and verbs in sentences. ....

Thanks! That I understand.
And I particularly enjoy the bear captain harbor example!

Zero-derivation can make it very hard to say what class the root belongs to.

And that too!

[Kuchigakatai]

OK, we agree on syntactic categories, so that part's cool.

I don't see that the lexical categories exist, or are needed.

In some languages, the morphosyntax tells you how to label the lexeme, so we're just using the (morpho-)syntactic category.

To use my example, nothing tells us if Quechua puklla- "is" a noun or a verb. But I don't see that there's a question to answer. We can list the root as a lexeme, or we can list two roots... it's a lexicographic question, not a linguistic one.

I don't think Salmoneus is talking about assigning word classes to roots at all, but rather about assigning word classes to words in a language like English or Mandarin where roots can often be used in bare form as words in a sentence.

If such a language allows every and any semantically noun-y word (people, places, things) to be both a syntactic noun "X" and a syntactic verb meaning "be an X, be the X", why should the syntactic verb be considered a derived separate word at all, instead of an inflection of sorts? Rather, you could talk about a Noun-y Lexical Class that when used in a particular sentence is either a (Morpho)Syntactic Noun or a (Morpho)Syntactic Verb.

This is different from the traditional semantic approach to word classes you described above because what holds the Noun-y Lexical Class together is not that it refers to people, places or things, but the observation that the semantics of the (Morpho)Syntactic Verb change in a regular way when compared to the (Morpho)Syntactic Noun. Ultimately it's semantics that holds a word and its forms together even in the traditional approaches anyway. With Salmoneus' distinction we just recognize that some "inflections" of sorts may do the most drastic of changes (changing verbal morphology/behaviour to nominal morphology/behaviour).

This could also apply to languages with bound morphemes too. The Latin gerund is very noun-like as it inflects for case and, unlike participles and infinitives, makes no tense distinctions, no voice distinctions and no subject agreement at all. It remains syntactically verbal-ish because it can take direct objects in the accusative case (something no normal noun can do), but otherwise we might as well consider it a derived noun. Seeing it skirts the word classes of Latin as it does, we could probably consider the Latin gerund to be morphosyntactically noun-like while still belonging to the Verb Lexical Class.

[Kuchigakatai]

In my grammar for my conlang Arpien, the word-and-phrase-classes I call “parts-of-speech” are “distributional” classes.

[...]

I don’t know whether what I came up with should be called lexical categories or lexical classes or syntactic categories or syntactic classes* or word categories or word classes. (*Probably syntactic classes, if I understand Salmoneus’s post about them.)
But I do know that, whatever they should be called, the distinction between one of them and another one is quite clear.

I'd say those are lexical classes, as they are groups of distributions. The individual places where they can go would be your syntactic classes.

[zompist]

I don't think Salmoneus is talking about assigning word classes to roots at all, but rather about assigning word classes to words in a language like English or Mandarin where roots can often be used in bare form as words in a sentence.

If such a language allows every and any semantically noun-y word (people, places, things) to be both a syntactic noun "X" and a syntactic verb meaning "be an X, be the X", why should the syntactic verb be considered a derived separate word at all, instead of an inflection of sorts? Rather, you could talk about a Noun-y Lexical Class that when used in a particular sentence is either a (Morpho)Syntactic Noun or a (Morpho)Syntactic Verb.

But we already have this concept— we call it zero-derivation. There's absolutely no problem with inflections changing syntactic category— e.g. white > whiten, note > notate, real > realize, heal > health. Zero-derivation just says you don't need an actual inflection.

Now, you could totally talk about roots that aren't specified for syntactic category— in English, Mandarin, Quechua, or (as I showed above) Latin. But I don't see the need to decide that they belong to a "noun-y lexical class", and I don't see a principled way to decide what class that is. Sometimes it's clear etymologically, but etymology is not accessible to most speakers.

Ultimately it's semantics that holds a word and its forms together even in the traditional approaches anyway.

As I said, if you think in terms of prototypes, the traditional definitions make more sense. You can use syntax alone, abstractly, to define syntactic categories, but deciding which ones to call "nouns" and which are "verbs" is basically semantic.

The Latin gerund is very noun-like [...] Seeing it skirts the word classes of Latin as it does, we could probably consider the Latin gerund to be morphosyntactically noun-like while still belonging to the Verb Lexical Class.

Or we admit that categories are not absolute but fuzzy.

[MacAnDàil]

Yes, the concept of zero-derivation exists, but, AFAICT, it basically posits an invisible modification of the word. So I prefer the analysis of just saying there is a large degree of polysemy, that one word can mean multiple parts of speech.

[Moose-tache]

What do we do with words like "dog," "take," and "access?" The first is clearly a noun turned verb, and vice-versa for the second, while the third has no obvious "default."

If we assume that each root has no inherent part of speech, like some Platonic form immune to classification, we're bound to run into problems with words where one part of speech is clearly dominant. For example, both "dog" and "take" serve as nouns, but only "dog" can then be used as a noun-turned-adjective. A phrase like "dog inspector" makes perfect sense, but "take insurance" (meaning "an insurance policy on a share of a sum of money") is confusing, even in context. Esperanto runs into this problem all the time by pretending that each root is neutral in terms of part of speech, but secretly having non-predictable defaults. One example is broso > brosi (a brush > to brush) but kombi > kombo (to comb, an act of combing).

On the other hand, if we assume that each root has a "default," and is subject to zero derivation, then what do we do with cases like "access," or "comb," or "handle?" Some will easily accommodate separate definitions (like the various meanings of "table"). But do we have any non-arbitrary criteria for when to separate derivations like this? Also, it feels somewhat like cheating; we always have the option to list every permutation of every root/word as a separate entry, but a big part of lexicology is figuring out how to not do that.

[zompist]

These strike me as lexicographic problems. English lexicographers always, so far as I know, separate N and V definitions, not least because you get extra senses that aren't equally ambivalent. E.g. "dog (N)" can be a wretched fellow, a hot dog, a bad product or performance, none of which apply to the verb. So it's easier to have a (N) section which lists all the noun senses, then list the (V) senses.

The Esperanto examples are interesting, because the process is obviously arbitrary. As you say, there seems to be a 'secret real answer', often revealed in the derivational morphology. My Esperanto dictionary just lists these-- e.g. it has entries komb/i, bros/o. I do wonder how they decided that 'rain' was pluv/o rather than pluv/i.

Verbs-turned-nouns certainly be used in compounds: hunt coordinator, sniff test, cruise ship, riot cop, fry cook, tickle fight. "Take insurance" is confusing only because it's not obvious what such a thing would be. That seems more a fact about the world than about the word.

[MacAnDàil]

We don't need to assume a zero-derivation in order to say that a root has one meaning preferred over another. For that, we can simply turn to the concept of prototypes: the prototypical meaning of the word 'dog' is a noun, and among its noun usages, the most prototypical is its usage for canines, more specifically the domestic dog. And some domestic dogs are more prototypical than others. The concept of prototype is already readily used for the last of these specifications but, in this particular example at least, the usage of it seems to me more clearcut for the earlier specifications.

Sure, 'dog' got turned into a verb from a noun, but this is just a larger degree of semantic change. It already underwent semantic change from Old English 'docga' "hound".

It is equally possible to do the same process with other lexemes.

[Moose-tache]

Verbs-turned-nouns certainly be used in compounds: hunt coordinator, sniff test, cruise ship, riot cop, fry cook, tickle fight. "Take insurance" is confusing only because it's not obvious what such a thing would be. That seems more a fact about the world than about the word.

The reason I used "take insurance" is subtler than that. If you saw a phrase like "dog fnarg" in isolation, you wouldn't assume it meant "to aggressively bother fnarg." You would assume that "fnarg" must be a word for poop, or kibbles, or something that could go with the nominal meaning of "dog." On the other hand, if you saw a (as you say, meaningless) phrase like "take insurance" in isolation, you would assume it means "to take out an insurance policy." This was what I meant about the combination "take insurance" as an insurance policy on a share not being possible: simply that a reader won't parse it correctly. The examples you gave, like hunt coordinator don't really apply in this case. For one thing, it's not clear to me at all whether "hunt," "riot," etc. are inherently verbal or nominal. And also in most cases there is no distinction. The coordinator of a hunt is no different than the coordinator of some hunting. So they can't really illuminate what people are thinking when they read them. Are readers assuming that "cruise" was first a verb, which was then nominalized before modifying ship? Or is it just a fundamental noun compounding with another noun? The same problem occurs with phrases like "slap fight." Hence I deliberately used pairings ("dog inspector" and "take insurance") which could only make sense as N:N and V:N combinations respectively, showing that both N:N and V:N combinations would not be available for every case. Ergo, not every root that can serve as a noun and a verb does so in the same way.

Your point about the lexicographical problem is well taken, but I'm not really concerned with how the words are listed in the dictionary. You could always solve that problem by organizing your dictionary by roots, with infinite room to expostulate on the various forms each root can take. I am more concerned with how the derivational system of a language is described. Do we say that a language contains roots with no inherent part of speech that simply takes nominal or verbal endings, or do we say that a language contains nominal roots and verbal roots, which can be transmogrified with null derivational affixes? This may seem trivial, but I think it could be important in some circumstances (Just go ask the people who start every tree diagram with fourteen layers of nothing). It seems like a lot of hand-waving to call it just a dictionary problem.

[akam chinjir]

On the other hand, if you saw a (as you say, meaningless) phrase like "take insurance" in isolation, you would assume it means "to take out an insurance policy." This was what I meant about the combination "take insurance" as an insurance policy on a share not being possible: simply that a reader won't parse it correctly.

Though with any actual context, or even without context if it's spoken, it'd be hard for "take insurance" to be genuinely ambiguous. And if you went to work in an office where they had something called they called take insurance, you might need to be told what exactly it is, but the expression wouldn't cause you any grammatical confusion, I don't think.

I've got nothing useful to add on the larger issue, though.

[KathTheDragon]

I think the biggest problem with "zero derivation" is the second word. After all, does anyone actually have a definition of "derivation" that is consistent across all languages? Haspelmath has one - it's just the less productive end of the productivity spectrum for grammatical constructions. hence if "zero derivation" is allowed to be limitlessly productive, as you'd require for defending a noun-verb contrast even when all words can be both, it's not derivation anymore.

And honestly, the fact that word-class-changing inflection exists (see Haspelmath's paper) undercuts the logic for staunchly separating word classes anyway.

[zompist]

I am more concerned with how the derivational system of a language is described. Do we say that a language contains roots with no inherent part of speech that simply takes nominal or verbal endings, or do we say that a language contains nominal roots and verbal roots, which can be transmogrified with null derivational affixes? This may seem trivial, but I think it could be important in some circumstances (Just go ask the people who start every tree diagram with fourteen layers of nothing). It seems like a lot of hand-waving to call it just a dictionary problem.

I guess I don't see much practical difference. You start with an ambiguous root, and you end with a clear noun or verb. It sure looks like something changed— the ambiguity disappeared. Assigning it to a rule, or to a hidden inflection, or a hidden polysemous part of speech, are different ways of saying the same thing.

I'll grant you that the idea of zero-derivation makes the most sense if there's a lot of derivation anyway. As I pointed out, there are N/V or A/V ambiguous roots in Latin. But more typically you get an N>V or V>N inflection. So zero-derivation fits in with everything else and doesn't introduce a new concept.

It's definitely more of a stretch in Mandarin, which has compounding but very little inflection. That doesn't mean the roots are all fluid, but here the polysemy approach would fit the language more.

[Vijay]

My dad trying to sing this under his breath, knowing that he doesn't know the words:

[kʊt͡ʃɨ naː kəhoː kʊt͡ʃɨ biː naː kəhoː
kjaː kɛhɛna heː kjaː sʊnɨna heː
mut͡ʃɨko pəɾaː heː koːjiː nəhiː heː]
(Repeats from beginning continuously until he gets sick of it and just goes back to singing Christian hymns in Malayalam instead)