Marking in VPs

[ibarrere]

I posted this over at lingforum.com, but it looks like everybody has migrated here. Anyway:

It seems to me that most European (specifically case marking) languages would be classified as double-marked for VPs in that they generally have person/number agreement on the head (verb) and some form of case marking on the dependent (verbal arguments).

However, WALS says that "[t]he dependent-marked clause is common in Eurasia" ( http://wals.info/chapter/23 ) and indeed, the WALS map shows German, English, Finnish, etc as being dependent-marking. Is there something about the case marking in these languages that classifies them more as dependent-marked than head-marked?

[Daniel]

My initial reaction to your post was that you have a good point. Eurasian languages do seem to have double marking, at least in that they have some of it. It may be more dominant in one or the other. And there's a question of whether different kinds of constituents are considered, perhaps suggesting one is grammatically dominant.

Now I'm completely confused. Regardless of what seems like the best labeling strategy to me, I'd assume that they'd be consistent in their mapping. But they're not! In fact, for the languages I'm familiar with (Eurasian mostly) many of the datapoints seem flat out wrong:

French: no marking
That's perhaps the direction, but it's certainly found in the written language and sometimes pronounced.
English: dependent marking
Shouldn't it be the same as French? And how is English dependent marking? Pronominal case?
German: dependent marking
That makes sense, but it's very different than English, much more active with case.
Spanish (and Basque and Modern Greek): double marking
Seems correct to me (what you said), but then why not German (and probably English and French) too?
Egyptian Arabic: no marking
Case is gone from various spoken dialects of Arabic but as far as I know the verbs still do agree...



This is very strange. It's possible we're missing something, but I certainly don't get it at the moment.

[ibarrere]

In fact, for the languages I'm familiar with (Eurasian mostly) many of the datapoints seem flat out wrong

I've gotten that impression a few times with WALS. I wish they had a justification for all of their points as some of them seem to make no sense.

French: no marking
That's perhaps the direction, but it's certainly found in the written language and sometimes pronounced.
English: dependent marking
Shouldn't it be the same as French? And how is English dependent marking? Pronominal case?
German: dependent marking
That makes sense, but it's very different than English, much more active with case.
Spanish (and Basque and Modern Greek): double marking
Seems correct to me (what you said), but then why not German (and probably English and French) too?
Egyptian Arabic: no marking
Case is gone from various spoken dialects of Arabic but as far as I know the verbs still do agree...

I feel like all of these should be either head-marking or double-marking if we're talking about VPs. Every single language in that list has some sort of verb agreement with the subject, which is by definition head-marking (I think...). With that said, the ones that have no nominal case morphology would be head-marking while the ones with nominal case morphology (nominative and accusative at least) would be double-marking.

In fact, it seems like a huge majority of the world's languages would be double-marking or head-marking, but WALS seems to claim the opposite. I can think of very few languages which have no agreement on the verb at all. Lezgian comes to mind, as roles are determined purely by case assignment, but that sort of behavior seems pretty rare to me.

We must be missing something, but I can't imagine what. My understanding of head- vs dependent-marking is fairly solid. Perhaps it's an idiosyncrasy within WALS classification technique or something. I'd like to get to the bottom of this though.

[Daniel]

I've gotten that impression a few times with WALS. I wish they had a justification for all of their points as some of them seem to make no sense.

I have generally found it useful and assumed it was reliable, but now I'm questioning that. Specifically, justification would be incredibly helpful.
I am working on a project regarding coordination, more specific than their entry, for unrelated reasons, and I was just looking for some ideas of languages to check out a few weeks ago. The entry on NP/VP coordination was helpful sometimes. But it simplifies the data in a way that is unfortunate and, as far as I can tell, didn't make it easier for the authors. It's comparing whether NP and VP coordinators are the same form. It has three values: one when they're the same, one when they're different, and one when they're both null. What I'd love more information about are the cases when one is null and the other isn't, but those are just thrown into the general "different" category. (I was interested in, among other things, null VP coordinators). It would also be helpful to know whether there are multiple coordinators for each group and possibly if these are identical for VP/NP.
This wouldn't have to replace the current graphs, just supplement them.

The Atlas is impressive anyway, because it combines so much. More notes that didn't overwork the authors would be desirable, though. One reason they might not include more info is that they couldn't be thorough on all of it due to time or limited descriptions, but some of these details are part of what's needed to write anything at all (like whether a coordinator is overt or not).


Anyway, that was a bit off the topic of marking...

In fact, it seems like a huge majority of the world's languages would be double-marking or head-marking, but WALS seems to claim the opposite.

My biased sample of (mostly Eurasian) languages certainly follows what you're saying, and I feel like I've often come across similarly unsurprising systems in other languages I don't know well (random examples in a textbook, etc.). I am also very skeptical now about what the distributions in WALS are really claiming-- they may not be wrong, but they certainly aren't informing us of what we are expecting, such as whether these languages look like English or not.



In my personal opinion, it would be great if WALS became an open community like Wikipedia. The print version of the text is useful but has now been out for several years (which is why the website is free, I believe), and for continuity having just that info on the website is fine. But a WALS 2.0 with Wikipedia-style peer review* and added languages** and variables would be amazing! Every fieldworker should add in the data for their language. It would take almost no time at all and it would exponentially increase the information in and value of WALS (which is already great).
[*I would expect some verification of the info and probably not anonymous contributions]
[**There is some logic behind the distribution of languages they chose, and that is useful if you want to look at the world in that way. But other times a research project isn't about even distributions but about finding instances of a phenomenon for example, where all possible data is useful regardless of sampling method. Moreover, the set of languages included now is not consistent anyway, due to limited descriptive information, so perfect consistency/sampling can't be achieved.]

Do you think that's a conceivable idea? I wonder if they'd be interested in that.

[ibarrere]

I think that's a great idea. In fact, I was looking for a comments section of sorts for this particular component to try to get to the bottom of the marking question I brought here. There are a few existing comments (by who appear to be specialists) on the right side claiming that several data points are incorrect, although I'm not sure the comment interface is freely usable/something people even pay attention to (the comments are several years old).

I'm not sure converting the existing WALS installation into something quasi-publicly editable would be feasible, but creating a platform like that from scratch (reusing WALS data) would probably be a reasonable undertaking.

[ibarrere]

For what it's worth, I looked into the WALS data in a bit more detail and came up with this. WALS cites the "locus of marking in the clause" as follows:

In any kind of phrase, overt morphosyntactic marking reflecting the syntactic relations within the phrase may be located on the head of the phrase, on a non-head (i.e. on a dependent), on both, or on neither. In clauses, the arguments are dependents and the verb is the head.

The way I interpret that, head-marking languages will show some sort of agreement morphology on the verb and dependent-marking languages will show some sort of case marking on one or more arguments of the verb. As such, we can imagine that English would be a head-marking language when talking about VPs: "I see the table" vs "John sees the table", the only morphological difference is the 1sg vs 3sg agreement on the verb (see vs sees). The description from WALS for head-marking from Tzutujil backs that up:

jar aak'aalaa7 x-Ø-kee-k'aq aab'aj
the boys comp-3sg-3pl-throw rock

More-or-less the same thing as English, with the exception of subject and object agreement on the verb. Likewise, the dependent-marking example from the definitions is very un-English with its nominal morphology and lack of verbal agreement:

wutpu-nku uma-Ø ute-n
old.man-erg firewood.abs pick.up-pst

However, the map overtly cites English as a dependent-marking language. It says the same thing about Finnish, which definitely has person/number agreement on the verb (in addition to nominal morphology). There must be something wrong with this particular data set in WALS or else I'm seriously misunderstanding what's going on.
 

[Daniel]

I agree with what you said.

English is 'both':

Dependent marking: case is restricted to a few pronouns.
He saw her. / She saw him.

Head marking: only 3SG.PRES is marked (plus be.1SG PRES/PAST)
I see the table / John sees the table

Therefore, English is almost neither, but to the extent that it is one, it seems to be the other as well.

In terms of contrastive features, there's only:
Case: Nominative/subjective vs. Accusative/oblique
Agreement: either [±3SG], or [±SG, {1,2,3}]
Under a surface-level analysis, the complexity/extent of the marking appears to be roughly equivalent.
Arguable extra complexities:
--Genitive case in nouns
--1SG am/was [only] in verbs; [also, that agreement only applies in the present tense, except with 'be']

There must be something wrong with this particular data set in WALS or else I'm we're seriously misunderstanding what's going on.

Indeed.



Oddly enough, I wonder whether the better known languages might be less reliable in WALS. I imagine that the less documented languages simply say something in a grammar and it is coded into the database. Then for a language like English there is so much research and so many controversial ideas that perhaps the results are messier? This is obviously not acceptable anyway, but at least it would mean that to the extent that the maps are useful (which is, let's be honest, not really for English, Indo-European, etc. :p) and that we can rely on WALS for the languages we're not familiar with.
Or it could turn out that if it's this bad for English, it's just complete nonsense for the rest. I'm surprised though! Maybe we really are missing something, or maybe they really did switch the datasets...



Also, to be picky, the "no marking" category seems like it should be "word order only", and there should be a 4th (hypothetical) category "no encoding" or "pragmatic inference alone", which I doubt truly exists (no case, no agreement, no word order).
(Looking at the data there for Australia, where word order is often not much of a factor, I see no white dots. Dyirbal for example has case. Many languages have both. So it really does seem unlikely that a language would have no word order if it didn't have overt marking either.)
A 5th option could potentially be intonation alone, if that exists.

And what's "other" in this case? What's other than head or dependent marking? It's not neither, it's not both. I could understand "unknown" or "undefined".



Edit:
Looking at WALS again...

There is a place to submit comments. If our conversation doesn't make more sense of this, let's do that.

Looking at the two existing comments, I'm not sure I follow the logic in them either (the one on Salish is confusing, implying something about degree). Or maybe the issues we're seeing are just the tip of the iceberg.

Overall, one possibility is that this is just terminological. I'd hope they applied the definitions consistently, but it is possible that some are due to traditional classifications that don't actually match the data well. If "have" and "not have" can be interpreted as a sort of cognitive perception of the language, then who knows. What I mean is that we might not feel that agreement is "important" in English, so we could say it "doesn't have" agreement. That's a bad strategy, but could be the explanation. I'm still not seeing another.

[ibarrere]

The description for "other" is section 1.5 in this list. It does indeed account for a "low frequency" fifth option. Where did you see the comment section on WALS? I'll submit a question to it.

[Daniel]

Ah, I see. So the last option is floating (which is technically a kind of neither ), although I would perhaps think of that as a subtype of dependent marking: there's a relationship encoded non-locally, apparently somewhere near the dependent. But that works.

Rather than typologize clause locus based on the treatment of all three of subjects, direct or primary objects, and indirect or secondary objects, we have relied on just the treatment of direct or primary objects in this map.

This is found near the bottom of section 1.5.
I'm not clear on whether they mean that it applies that way only for the "other" type (which is odd), but I think that answers our questions if it applies for the whole map.

This is NOT about subject-verb agreement and case in general. This is strictly about direct objects. That clears up some of the inconsistencies.
French and English should still have the same value (case in only pronouns is found in each; there is no object agreement on verbs).
Spanish has differential object marking (the 'personal a') and so does Modern Greek I believe.
Hungarian has subject and object verb agreement. What about Finnish then?

I'm not clear on all of the details still, but the data is a little more plausible from that perspective.

If that's right, then the result is that this chapter is highly misleading and represents a very small (and possibly misrepresentative) subset of the data. It's fine for what it is, but what is it? Why would we care about geographic generalizations based on only direct objects? What if other arguments (like subjects!!) are encoded very differently? (And I think they are, as we've discussed here.)

[ibarrere]

Rather than typologize clause locus based on the treatment of all three of subjects, direct or primary objects, and indirect or secondary objects, we have relied on just the treatment of direct or primary objects in this map.

Wow. I love how that's the fine print, I didn't even catch it. That certainly does clear some of it up, at least.

French and English should still have the same value (case in only pronouns is found in each; there is no object agreement on verbs) ... What about Finnish then?

I'm not familiar with French, but English should still be classified as head-marking, if not more so with this in mind. Since the morphological changes taking place are only on the verb regardless of whether or not they're in concord with the subject or object it should still be considered head-marking, at least more than it should be considered dependent-marking. Finnish should still be double-marking as it has person/number agreement on the verb in addition to nominative/accusative/partitive marking on the subject and object(s).

If that's right, then the result is that this chapter is highly misleading and represents a very small (and possibly misrepresentative) subset of the data. It's fine for what it is, but what is it? Why would we care about geographic generalizations based on only direct objects? What if other arguments (like subjects!!) are encoded very differently?

Yeah, it's really quite silly. Only focusing on direct object agreement makes very little sense. Not only is the generalization pretty useless, but it entirely misses intransitive clauses as well.

[Daniel]

I'd be interested in at least info on subjects as well.

[ibarrere]

I'd be interested in at least info on subjects as well.

Yeah, definitely. I don't have a huge inventory of examples in my head, but object agreement? It seems like I only see that in polysynthetic, radically head-marking languages. Isn't subject agreement much more common?

[Daniel]

Isn't subject agreement much more common?

And therefore less interesting to some? It's highly misleading to disregard how common that is as if it isn't, but in terms of finding contrasts, perhaps it's just uninteresting.

radically head-marking languages

Indeed. And it's ideas like "radically" that seem to be relevant here-- it's not about literally what a language is doing, but some sort of interpretation of what a "head marking" language should be like (eg, a "radical" one). Since subject agreement is so common, the "real" or "interesting" test for whether a language is "really head marking" is whether it also marks other things.
I can't be certain this is what the authors had in mind, but I'm getting that feeling.

A very similar situation exists for "polysynthetic", which has absolutely no formalized conventional definition. It's a matter of impression, I guess.

Without definitions (that hopefully relate to something meaningful), distributions aren't much.

I don't have a huge inventory of examples in my head, but object agreement? It seems like I only see that in polysynthetic, radically head-marking languages.

Hungarian is the go-to example, and I don't know about others. I've seen examples here and there from various groups (I think one was Native American, but I'm just vaguely recalling the discussion), and I don't specifically remember any correlation to polysynthesis, but I wouldn't be surprised.

Depending on our definitions, we could include some other languages:
--Swahili has (definite) object morphemes: Ninakisoma Kiswahili 'I study-it Swahili' (The object might most often then be omitted.)
--Spanish and other Romance languages have object clitics: "Veo el banano" 'I see the banana' / "Lo veo" 'I see it'. Depending on morphological analysis that could possibly be considered agreement, at least when (it's rare) you can double the object and clitic: "Me lo da a mí" lit. 'me it he-gives to me', 'He gives it to me.'
--In Italian there's an interesting hybrid participial adjectival/verbal construction involving object agreement... in the event that a perfective form is transitive and has an object, the participle agrees with the object: "I have eaten.MASC.PL NOUN.MASC.PL". It's been a while since I studied that particular rule, but that gives you the idea. It's in a tiny pocket of the grammar, though. It's unclear whether these details are thought to matter.


Anyway, it seems sort of impressionistic about the "nature" of a language.

[ibarrere]

I typically use polysynthetic and polypersonal somewhat interchangeably even though they probably have at least vaguely formalized definitions which are distinct from one another; I should work on that I suppose. I certainly don't buy into the definition from Baker (1996) regarding polysynthesis. It's one of those weird descriptions that probably does a lot more good being used impressionistically.

At any rate, I'm not sure a lot more progress can be made on this particular subject. It's a shame that the WALS chapter is so misleading (and vague, and possibly erroneous), but I guess I'm glad we're on the same page regarding how head- and dependent-marking languages behave. Like so many other things in linguistics, we're probably better off approaching marking paradigms as a continuum rather than an on-off thing.

For what it's worth, the example I was thinking of for object agreement was Greenlandic. Eskimo-Aleut varieties typically agree on agent and patient, they're also good examples of typical head-marking languages when concerning VPs. But again, they mark ergative and absolutive cases on the nominals as well. So double-marking? :sigh:

[Daniel]

I typically use polysynthetic and polypersonal somewhat interchangeably even though they probably have at least vaguely formalized definitions which are distinct from one another; I should work on that I suppose. I certainly don't buy into the definition from Baker (1996) regarding polysynthesis. It's one of those weird descriptions that probably does a lot more good being used impressionistically.

By the way, I was just bringing that up as a general example, only coincidentally overlapping with your post.
I'm not sure that I agree it's good to use it impressionistically. Sure, we could say "Wow, that language looks polysynthetic", which is fine as an impression. But impressions don't work well on a map for example. The fact that people generally have the impression that North American languages are polysynthetic does not mean that all (or any percentage) of them are, or that languages elsewhere are not.

For what it's worth, the example I was thinking of for object agreement was Greenlandic. Eskimo-Aleut varieties typically agree on agent and patient, they're also good examples of typical head-marking languages when concerning VPs. But again, they mark ergative and absolutive cases on the nominals as well. So double-marking? :sigh:

Oh, I wasn't saying you're wrong about the two appearing together. I just don't know that there is any correlation. When I've seen object agreement (aside from Hungarian) it's been in the "exotic language" examples (y'know, the one that isn't English in the syntax textbook), but beyond that I can't recall any particular type of language.