The diachronics of noun classes

[LoneWolf]

Diachronically speaking, how do noun classes come into being?

I have read on a few occasions that noun classes can sometimes originate from classifiers or again from the cliticization of distinct sets of pronouns ... but aside from that I haven't a clue

[Yng]

All sorts of ways. For example, a bi-class system could come into being if we had, say, two large classes of nouns that end in the same way (let's go with -a and -o for funzies). If you then get adjective agreement - which is possible - based on the ending of the noun, and then all of the nouns that don't share the endings get analogised into another class, or merged into the two existing classes - then you have masc vs fem (or class 1 vs class 2, more accurately).

Otherwise - classifiers, pronouns, etc etc.

[merijn]

All sorts of ways. For example, a bi-class system could come into being if we had, say, two large classes of nouns that end in the same way (let's go with -a and -o for funzies). If you then get adjective agreement - which is possible - based on the ending of the noun, and then all of the nouns that don't share the endings get analogised into another class, or merged into the two existing classes - then you have masc vs fem (or class 1 vs class 2, more accurately).

Otherwise - classifiers, pronouns, etc etc.

How do you get adjectival agreement in your scenario?

[Yng]

What I said - based on the ending of the noun. The noun's suffix is generalised to adjectives.

[merijn]

What I said - based on the ending of the noun. The noun's suffix is generalised to adjectives.

Do you have any examples where that happened in any language? I cannot see a reason why a language would change that way.

[alice]

What I said - based on the ending of the noun. The noun's suffix is generalised to adjectives.

Do you have any examples where that happened in any language? I cannot see a reason why a language would change that way.

Have a look at n-stem nouns, from whose endings the entire weak declension of adjectives developed in Germanic.

[merijn]

That is not the same. It would be the same if the form with the-n- endings were only used before nouns with -n- endings.

[roninbodhisattva]

What I said - based on the ending of the noun. The noun's suffix is generalised to adjectives.

Do you have any examples where that happened in any language? I cannot see a reason why a language would change that way.

Language change does not often have an obvious reason. It's just something you have ton deal with it. This shit is arbitrary.

It's a perfectly plausible change, especially if adjectives are very closely related/closely pattern with nouns.

[linguoboy]

It's a perfectly plausible change, especially if adjectives are very closely related/closely pattern with nouns.

I think that's the key: the distinction between nouns and adjectives in Indo-European languages is very weak when compared to other languages like Chinese or Mixtec where "adjectives" are a verbal subclass. There's no reason to think that a language like this would be amenable to the kind of development YngNghymru describes.

[roninbodhisattva]

It's a perfectly plausible change, especially if adjectives are very closely related/closely pattern with nouns.

I think that's the key: the distinction between nouns and adjectives in Indo-European languages is very weak when compared to other languages like Chinese or Mixtec where "adjectives" are a verbal subclass. There's no reason to think that a language like this would be amenable to the kind of development YngNghymru describes.

There's also absolutely no reason to think that the language wouldn't be amenable to such a change. Nothing in his description suggests that the adjectives would have been a verbal subclass. And even if it were, it could certainly still happen. All you need is some kind of adjectival like modifier in the noun phrase that then has the noun ending generalized to it.

[Yng]

It's a perfectly plausible change, especially if adjectives are very closely related/closely pattern with nouns.

I think that's the key: the distinction between nouns and adjectives in Indo-European languages is very weak when compared to other languages like Chinese or Mixtec where "adjectives" are a verbal subclass. There's no reason to think that a language like this would be amenable to the kind of development YngNghymru describes.

So? The OP didn't specify a specific type of language in which this could occur.

[merijn]

To clarify myself: what I don't find likely (and I think has never been attested) is the following scenario:
1) there are some nouns that end in -a, and some nouns that end in -o
2) Our of nowhere, adjectives start to show agreement with the final vowel of the noun; adjectives before nouns ending in -a start to end with an -a an adjectives before nouns ending in -o start to end with -o.
3) Other nouns are reanalyzed in one of the classes.
I find step 2 very unbelievable.

What I do find believable is the following:
1) A group of animate nouns has marking for gender; -o is male and -a is female.
2) What western languages do with adjectives, this language does with noun apposition (perhaps amongst other strategies). So where we would say, "this big meter maid", speakers of this language would say, "this meter maid, a giant". Semantics demand that we would use the female variant of giant, say gianta for meter maid, but the male variant, gianto for a Beatle.
3) the nouns in apposition are reanalyzed as adjectives and the gender marking as agreement marking.

[murtabak]

Something like Chinese measure words can also develop into noun classes, am I right? This seems a more plausible origin of very complex noun-class system (as in Bantu), whereas what YngNghymru said might develop into two-way or three-way distinction, at most. Anyway, is there any study on how the Bantu system originate?

[TaylorS]

What I said - based on the ending of the noun. The noun's suffix is generalised to adjectives.

Do you have any examples where that happened in any language? I cannot see a reason why a language would change that way.

It is thought to have happened in PIE

[linguoboy]

What I said - based on the ending of the noun. The noun's suffix is generalised to adjectives.

Do you have any examples where that happened in any language? I cannot see a reason why a language would change that way.

It is thought to have happened in PIE

As I understand it, PIE didn't have a formal morphological distinction between nouns and adjectives. Thus, as merijn says, you may be harkening back to a situation where apparent modifiers were actually appositives.

[Radius Solis]

To clarify myself: what I don't find likely (and I think has never been attested) is the following scenario:
1) there are some nouns that end in -a, and some nouns that end in -o
2) Our of nowhere, adjectives start to show agreement with the final vowel of the noun; adjectives before nouns ending in -a start to end with an -a an adjectives before nouns ending in -o start to end with -o.
3) Other nouns are reanalyzed in one of the classes.
I find step 2 very unbelievable.

Well, I'd disagree on instinct that (2) is impossible, though it's true I have no direct examples to hand. But for circumstantial evidence, we have systems like the Bantu noun classes - just about every Bantu sentence is full of class-marking prefixes, which appear on nouns, adjectives, verbs, and numbers. I find it a very big stretch to assume that copy-morpheme systems this extensive must have arisen without some spontaneous application of morphology to word classes that didn't previously take it. Besides which, optimality theory predicts it should be possible. There are constraints motivating agreement of various types, and we'd expect these constraints to be capable of rising to a ranking that dominates I/O fidelity constraints. This could produce spontaneous extension of existing class/gender morphology to word classes that didn't previously take it.

[merijn]

To clarify myself: what I don't find likely (and I think has never been attested) is the following scenario:
1) there are some nouns that end in -a, and some nouns that end in -o
2) Our of nowhere, adjectives start to show agreement with the final vowel of the noun; adjectives before nouns ending in -a start to end with an -a an adjectives before nouns ending in -o start to end with -o.
3) Other nouns are reanalyzed in one of the classes.
I find step 2 very unbelievable.

Well, I'd disagree on instinct that (2) is impossible, though it's true I have no direct examples to hand. But for circumstantial evidence, we have systems like the Bantu noun classes - just about every Bantu sentence is full of class-marking prefixes, which appear on nouns, adjectives, verbs, and numbers. I find it a very big stretch to assume that copy-morpheme systems this extensive must have arisen without some spontaneous application of morphology to word classes that didn't previously take it. Besides which, optimality theory predicts it should be possible. There are constraints motivating agreement of various types, and we'd expect these constraints to be capable of rising to a ranking that dominates I/O fidelity constraints. This could produce spontaneous extension of existing class/gender morphology to word classes that didn't previously take it.

OK, you make to points. 1) Bantu languages form circumstantial evidence that it happened, and 2) OT predicts it. I will start with the second one before tackling the first one.

OT is a theory about phonology, not about morphosyntax. It is true that there have been people trying to adapt OT for syntax, but not everybody agrees that it is attempts are successful. However, even if we do assume OT for syntax, and even if we assume that there are constraints that want agreement between nouns and adjectives (let's limit ourselves to adjectives). Well, here is the following problem, how many noun classes does a language have before these constraints rise? Well one, which includes all nouns. So after this agreement constraint rises, how many noun classes are there available for triggering agreement, still one. For this scenario to work you need noun classes first, so it can't be an account of noun class genesis, it can be an account of some type of agreement genesis though.

Your second point, the Bantu languages. Noun classes in Bantu languages go back a long way, and are usually constructed for proto-Niger-Congo, which dates a long way back. Proto-Bantu is thought to have three patterns of agreement where all other patterns of agreement are based on. One is the adjectival agreement for adjectives, the second is subject agreement that forms the basis for most types of agreement, including subject agreement, object agreement, the pronouns, the demonstratives, several quantifiers including "all" and "alone", and a third one called enumerative agreement which has morphological features of both adjective agreement and subject agreement, and which is thought to have been used with a handful of determiners and probably some numbers. As you can see, most types of agreement are based on the subject agreement (with the caveat that noun class 1 is in many cases irregular). Subject agreement is however the type of agreement that is the furthest away from the noun class prefixes. Many noun classes, including the most common ones, the subject agreement is different from the noun class marker. Of the first 10 noun classes, 7 are commonly reconstructed with difference between the noun class prefix and the subject agreement, though the less common higher noun classes are usually reconstructed with similar noun class prefixes and subject agreements.
We also don't need spreading of morphemes to explain the wealth of agreement patterns in Bantu. If we assume that the subject agreement was originally a demonstrative, I think you can explain all the different types of agreement that are based on the subject agreement diachronically.
One last remark, especially in some Niger-Congo languages people have argued that the opposite happened to what you are suggesting. Nouns had lost their noun class marking but an article agreeing with noun class was attached to it and as it lost its meaning was reanalyzed as a noun class affix.

[Bedelato]

Noun declension classes come about when a single homogeneous paradigm interacts with different noun stems through sound change. Then analogy works its course.

Here's a simple example. Even though this is L&L, I'll be writing from a conlanging point of view here, since there's applications in both areas.

Say a language has the following noun paradigm:

Code: Select all

       Sg        Pl
Dir    --        -on
Gen    -us       -i
Dat    -a        -u

Let's say this language has the following lexicon, with declensions (no meanings given, since we're only concerned with form here):

veti: veti, vetius, vetia, vetion, vetii, vetiu
med: med, medus, meda, medon, medi, medu
igu: igu, iguus, igua, iguon, igui, iguu
ova: ova, ovaus, ovaa, ovaon, ovai, ovau

See all those vowel clusters? Let's introduce some monophthongization.
Here are the resulting paradigms. Macrons mark vowel length, and <y> = /y/.

veti: veti, vetȳs, vetia, vetin, vetī, vetȳ
med: med, medus, meda, medon, medi, medu
igu: igu, igūs, igō, igūn, igȳ, igū
ova: ova, ovōs, ovā, ovōn, ovē, ovō

Yep, we just created four declension classes from a single paradigm, through a simple sound change.

Then analogy will work its course, eventually reducing the number of declensions to two or three. For example, the direct plurals of veti and med might become vetȳn and medun by analogy with the pattern in the other two words, with the same vowel being used in GEN-SG, DIR-PL, and DAT-PL.

Semantic associations will be made based on common elements of meaning within different classes. Words will start being reanalyzed into one class or another based on phonosemantic properties, and this way a gender system might develop.

And that's basically how noun declension classes develop. There's obviously much more to it (e.g. pronoun agreement), but yeah. The same thing can happen to other types of words too.