Relative clauses: cross-linguistic comparison

[zompist]

I've always recommended to conlangers that they consider four possible relative situations, where the target noun is

(1) subject in both clauses: The player who complains about his team will lose.
(2) subject in main, object in subclause: The player that Widowmaker shot was too visible.
(3) object in main, subject in subclause: Everyone hates the player who complains about his team.
(4) object in both clauses: Mercy resurrected the player that Widowmaker shot.

Looking more closely at several natlangs, though, it seems the subclauses in (2) and (4) are the same, and so are the ones in (1) and (3). That is, the structure of the subclause (including case roles) doesn't depend on anything in the main clause.

Does anyone have a natlang counterexample at hand? That is, an example where the subclause itself varies depending on whether it's attached to a subject or an object?

[Vijay]

Latin.

Also, Ancient Greek. From Language Universals & Linguistic Typology (1989, 2^nd ed.) by Bernard Comrie, p. 153-154:

"One phenomenon that is found sporadically is that known to traditional grammarians of Latin and Greek as attraction, whereby the case marking of the head noun in one clause is attracted into that of the other clause. Persian provides particularly clear exemplification of attraction from the relative clause...In Ancient Greek, attraction usually works the other way round, an expected accusative relative pronoun in the relative clause being attracted into the case of its antecedent:

ek tõn póleōn [hõn éxei] (41)
from the cities-GENITIVE which-GENITIVE he-has
'from the cities which he has'

The preposition ek requires the genitive case, so the genitive case of tõn póleōn is as expected in the main clause; the verb éxei ' he has ', however, would be expected to have an accusative object, but instead the relative pronoun has been attracted into the case of the noun phrase within the main clause."

Also, this happens in English sometimes! From here:

The book is one of those rare novels that owes its (owe their?) popularity more to word-of-mouth recommendation than to the publisher’s advertising.
(cf. Those novels that owe their (???owes its) popularity more to word-of-mouth recommendation than to the publisher’s advertising aren't worth reading.)

EDIT: Hmm. That last one probably occurs in some other languages, too, doesn't it?

[jmcd]

More suggestion that anythng, but surely the alignement will affect things? Ergative languages will do relative clauses differently, as will Austronesian alignement languages, I'd expect.

[Vijay]

They do relative clauses differently, but that still doesn't necessarily mean that the structure of the subclause depends on anything in the main clause, right?

This is an example Comrie has in Malagasy (p. 159 of the same book I quoted in my last post; I've changed the gloss slightly):

ny vary izay novidin'ny vehivavy ho an'ny ankizy
'the rice that the woman bought for the children'

Is there any other way to say that in Malagasy? Does it matter whether the rice is the subject or the object of the main clause?

[zompist]

The Greek example is great, though keep 'em coming.

I have the Comrie and I had a note to myself to re-read it, so I will make sure I do that.

Malagasy only allows relativization of the subject of a clause. To get anything else you need to change the voice (fortunately there's at least three).

[Vijay]

Malagasy only allows relativization of the subject of a clause. To get anything else you need to change the voice

Right, and he says so himself! But I still don't see how that would affect what you're looking for.

(fortunately there's at least three).

Wait, you mean there might be more?

Anyway, I'll see what if anything else I can come up with, but another thing I thought of was the use of the subjunctive in (some varieties of?) Spanish when referring to a theoretical/nonspecific entity, e.g. Búscate un hombre que te quiera 'get yourself a man who loves you' (this is apparently the name of a song) vs. Un hombre que te quiere en serio, no sugiere tales acciones de tu parte 'a man who seriously loves you doesn't suggest such actions on your part' (source: https://www.soycarmin.com/buenavida/Fra ... -0024.html).

EDIT:

The Greek example is great, though keep 'em coming.

Thanks!

[Xephyr]

Hopi has something called the "subject constraint" which says that a relativized noun can't be the subject-in-the-main-clause unless it's also the subject-in-the-relative-clause. So sentences 1, 3, and 4 are permissible in Hopi, but not 2.

[Circeus]

While we're on the topic, I'm curious to ask: how "normal" are relative clauses where the relativized noun is not an argument at all cross-linguistically? (are they even relatives proper ) Stuff like "the reason why", "the place where/whence", "the year when" etc.

[Vijay]

Comrie seems to suggest that making subclauses of all four types is more problematic in sentences where there is more than one subordinate clause. From the end of his chapter on relative clauses (p. 160-162):

"In section 7.3.1, we looked only at simplex sentences, and only at a single position at a time. However, one might venture to claim that the kinds of generalization made in section 7.3.1 can be extended if one also considers more complex constructions. [...] However, there are also problems in generalizing these extensions, indicating that further work needs to be done in some of these areas. For instance, a reasonable extension would be to claim that the subjects of subordinate clauses should be more accessible to relativization than non-subjects (just as subjects of main clauses are more accessible than non-subjects). However, all the evidence suggests just the opposite. In English, non-subjects of subordinate clauses are freely relativizable, while subjects can be relativized only if there is no conjunction:

the girl [ that you think (that) I love ] (66)
the girl [ that you think (*that) loves me ] (67)

For some speakers of Hungarian, quite irrespective of conjunctions, subordinate subjects cannot be relativized, whereas non-subjects often can:

a pénz, [ amit mondtam, hogy a fiú elvett ] (68)
the money which-ACCUSATIVE I-said that the boy took-away
' the money that I said (that) the boy took away '

*a fiú, [ aki mondtam, hogy elvette a pénzt ] (69)
the boy who I-said that took-away the money-ACCUSATIVE
'the boy that I said took away the money'

(Elvette is used with a definite direct object, otherwise elvett. Other speakers of Hungarian would find (69) grammatical.) In Imbabura Quechua, it is possible to relativize a non-subject of an embedded clause using the gap type, but not the subject of an embedded clause:

[ Marya Juan wawa -ta riku-shka -ta ni-shka ] llugshi-rka. (70)
María Juan child ACCUSATIVE see NOMINALIZER ACCUSATIVE say NOMINALIZER leave PAST-3SINGULAR
' The child that María said that Juan saw left. '

*[ Marya warmi Juan -ta riku-shka -ta ni-shka ] llugshi-rka. (71)
María woman Juan ACCUSATIVE see NOMINALIZER ACCUSATIVE say NOMINALIZER leave PAST-3SINGULAR
' The woman that María said saw Juan left. '

There is thus good cross-linguistic evidence for the surprising generalization that subordinate non-subjects are easier to relativize than subordinate subjects, but apparently no good explanations as to why this should be so."

[jal]

but apparently no good explanations as to why this should be so."

Topicality? Subjects are often topics, and though it would make sense to expand on that topic by relativization, it would make less sense (and more difficult for the mind to process) to try to kinda shift the topic mid-sentence and in a sub-clause at that.


JAL

[Vijay]

But doesn't this lead to a situation where changing the subject from one clause to another is exactly what you do?

[jal]

But doesn't this lead to a situation where changing the subject from one clause to another is exactly what you do?

Yes, but since that subject is actually an object (in the matrix clause) that seemingly doesn't matter that much. Note I was just brainstorming before, it's not a set theory :).

I also wonder if the word order plays a role. In a SVO language, it's easier to endlessly add subclauses on the object than the subject, since there's no need for difficult syntactic juggling to make it fit.

As for your example sentences 66 and 67, in Dutch it works quite different, that probably being the reason I have some difficulties with them:

het meisje waarvan jij denkt dat ik van haar hou (66)
het meisje waarvan jij denkt dat ze van mij houdt (67)

In both sentences, "waarvan" means "of whom" (or "of what") and "dat" is a subordinating conjunction (not pronoun "dat" as in "het meisje dat ik zie"), followed by a full clause with both subject and object. The gapping going on in English is impossible.

[hwhatting]

the girl [ that you think (that) I love ] (66)
the girl [ that you think (*that) loves me ] (67)

Is it sure that the sub clauses are really two consecutive relative clauses and not a complement clause ("that you think") embedded in a relative clause? Would they work using who(m)? (It's of course possible that I just have difficulties seeing them as two relative clauses because of my native German, which works like the Dutch in JAL's examples).

[Richard W]

Does anyone have a natlang counterexample at hand? That is, an example where the subclause itself varies depending on whether it's attached to a subject or an object?

In my usual English, 'who' with antecedent 'I' has 1s concord, but with antecedent 'me' it has 3s concord. This rule also works when 'I' or 'me' is the complement, so it is probably an attraction rule rather than anything deeper.

[zompist]

Does anyone have a natlang counterexample at hand? That is, an example where the subclause itself varies depending on whether it's attached to a subject or an object?

In my usual English, 'who' with antecedent 'I' has 1s concord, but with antecedent 'me' it has 3s concord. This rule also works when 'I' or 'me' is the complement, so it is probably an attraction rule rather than anything deeper.

Sounds neat, but to make sure I'm following, can you give some example sentences?

[Richard W]

Complements:

It's I who do that.
It's me who does that.

These two could also be a difference of register, but I'm not very good at register.

Subject v. object:

I, who loathe the idea, will resist it.
Why have they asked me, who doesn't approve of the plan, to implement it.

These are continuative relative clauses, but 'I' and 'me' are usually pretty well defined by their very nature.

[Vijay]

Those first two examples remind me of things like "I say so" vs. "says me." (I'm not sure to what extent the previous sentence makes sense to non-Americans. I hope it at least sort of makes sense to other Americans ).

[Travis B.]

Forms like "The system just crashed, says me." are grammatical to me, even though I cannot figure out how they are, because when by itself the subject ought to be I, and only becomes me (and only in everyday speech, not in higher registers) when there are multiple subjects including a 1S subject.

[zompist]

It's I who do that.
It's me who does that.

I, who loathe the idea, will resist it.
Why have they asked me, who doesn't approve of the plan, to implement it.

Thanks for the examples. Neat and a little weird!

I think I'd agree with you on the first pair. I'd probably say the second in preference to the first. It's certainly not *It's me who do that.

The second pair are stranger. Neither sounds terribly wrong to me, but if I wanted to say something like that, I think I'd reword it.

[So Haleza Grise]

Ergative languages will do relative clauses differently

Will they though? Switch reference markers seem universally to mark *subject*, not agent or patient, even in languages that exhibit ergativity. Of course, switch reference is not the only way to form relative clauses.

Of course, IME some ergative languages will have antipassives that can be used to chain clauses, but not every ergative language uses antipassives. And that's not to mention 'syntactic' versus 'morphological' ergativity - there are more examples of languages of the latter type than the former. At least, so I understand.

Really, I would like to see more examples of how different languages with ergativity use relative clauses. I feel like I haven't come across enough of them to understand them fully.

[Cedh]

Apparently, of the languages said to have syntactic ergativity, some indeed seem to have syntactically ergative relative clauses, while others have syntactically accusative relative clauses:

[quote="Heaton et al. 2016, "The Status of Syntactic Ergativity in Kaqchikel", Lingua 170, pp. 35–46"]
[...] some syntactically ergative languages are reported to treat relativization and wh-movement in different ways. One such language is Chukchi (Paleo-Siberian), in which the agent argument of a transitive verb can be directly questioned, but can be relativized only with the help of antipassivization (Polinsky, in press).

(3) a. Relativization (antipassivization required)
[ _ məlgr-epə ine-kune-lʔ-ən] ənpənačg-ən
_ gun-ABL AP-buy-PTCP-ABS old.man-ABS
‘The old man that bought a gun’

b. Wh-movement (no antipassivization)
Mikəne [ _ milger kun-nin]?
who.ERG _ gun.ABS buy-AOR.3SG.SUBJ.3SG.OBJ
‘Who bought a/the gun?’

Based on facts such as these, Polinsky suggests that relativization is the most reliable test of syntactic ergativity.

A possible challenge to this proposal comes from the K’ichean Mayan language Kaqchikel. Uncontroversially ergative, Kaqchikel has long been assumed to require detransitivization (in the form of agent focus) in contexts that call for the agent argument of a transitive verb to be relativized or questioned. As illustrated below, the agent focus pattern employs absolutive agreement morphology, signaling detransitivization.

(4) a. Relativization with agent focus:
Ri retal k’o pa ruwi’ ri ala’ [ri n-Ø-q’et-en ri xtän _ ]
DET sign be PREP top DET boy REL INCOMPL-3SG.ABS-hug-AF DET girl _
‘The arrow is above the boy who is hugging the girl’

b. Wh-movement with agent focus:
Achike [n-Ø-q’et-en ri xtän _ ]?
WH INCOMPL-3SG.ABS-hug-AF DET girl _
‘Who is hugging the girl?’

However, observations by the first author suggest that, contrary to the traditional view, direct relativization of the agent argument of a transitive verb may in fact be possible, even though no such option is available for wh-movement. Thus the relative clause in (5a) differs from the wh-question in (5b) in being acceptable, even though both are fully transitive, as shown by the presence of ergative agreement morphology [underlined] for the subject and of absolutive agreement for the direct object.

(5) a. Relativization without detransitivization:
Ri retal k’o pa ruwi’ ri ala’ [ri Ø-ru-q’et-en ri xtän _]
DET sign be PREP top DET boy REL 3SG.ABS-3SG.ERG-hug-PERF DET girl _
‘The arrow is above the boy who is hugging the girl’

b. Wh-movement without detransitivization:
*Achike [n-Ø-u-q’et-ej ri xtän _]?
WH INCOMPL-3SG.ABS-3SG.ERG-hug-TV DET girl _
‘Who is hugging the girl?’
(acceptable only when interpreted as a direct object question, with the interpretation ‘Who is the girl hugging?’)

[...]

As stated at the outset, our goal has been to use the technique of elicited production to explore a fundamental issue in the syntax of relative clauses and wh-questions in Kaqchikel: do either or both of these patterns show signs of syntactic ergativity? Our findings were robust, permitting a strong conclusion for each pattern.

On the one hand, we found little evidence of syntactic ergativity in the case of relative clauses. Although a detransitivization strategy was sometimes used to relativize the subject argument of a transitive verb, participants showed an overwhelming willingness to produce subject relative clauses that contained a transitive verb with ergative agreement morphology. The fact that this tendency was evident in all our participants, regardless of age, suggests that the accusative character of relativization is firmly established in the language.

On the other hand, we found very strong evidence for syntactic ergativity in the case of wh-questions. There, only younger speakers produced subject wh-questions that contained a transitive verb with ergative agreement morphology. Speakers 40 and older invariably had recourse to a detransitivization strategy that involved either an agent focus pattern or antipassivization---classic signs of syntactic ergativity.

It is unclear how and why Kaqchikel has reached this point. At least some other Mayan languages seem to manifest a similar asymmetry, including Mam, Ixil, and Sipacapense, all of which belong to the Eastern branch of the Mayan family. Beyond Mayan, however, the only other instance of split syntactic ergativity involving A-bar movement appears to be the mirror image of what we have reported for Kaqchikel: as reported by Polinsky (in press), Chukchi shows signs of ergativity in its relative clauses, but not in its wh-questions.

Taken together, the data from Kaqchikel and Chukchi raise new questions for the study of ergativity. Most obviously, they call into question the possibility [...] that relative clause patterns might offer a criterial test for whether a language is syntactically ergative. In contrast to Chukchi, syntactic ergativity in Kaqchikel is manifested in wh-questions, but not in relative clauses.
[/quote]

[So Haleza Grise]

Fantastic info! Thank you very much for this.

[Arzena]

I've always recommended to conlangers that they consider four possible relative situations, where the target noun is

(1) subject in both clauses: The player who complains about his team will lose.
(2) subject in main, object in subclause: The player that Widowmaker shot was too visible.
(3) object in main, subject in subclause: Everyone hates the player who complains about his team.
(4) object in both clauses: Mercy resurrected the player that Widowmaker shot.

Looking more closely at several natlangs, though, it seems the subclauses in (2) and (4) are the same, and so are the ones in (1) and (3). That is, the structure of the subclause (including case roles) doesn't depend on anything in the main clause.

Does anyone have a natlang counterexample at hand? That is, an example where the subclause itself varies depending on whether it's attached to a subject or an object?

Let's see how Modern Standard Arabic handles them:

اللاعب الذي يشتكي من فريقه سيخسر
al-lā3ibu alāḏī yaštakī min fariqihi sayaxsar
al-lā3ibu alaḏī ya-štakī min fariq-i-hi sa-ya-xsar
DEF-player-NOM REL.masc.sg. 3sg.IND-complain from team-GEN-3sg.masc FUT-3sg.masc-lose
The player who complains about his team will lose.

اللاعب الذي أطلق المرمل النار عليه كان بائنا بالكثير
al-lā3ibu alāḏī aṭlaqa al-Murammil al-nāra 3alayhi kān bā'inan bi-l-kaṯīr
al-lā3ibu alaḏī aṭlaqa al-Murammil al-nāra 3alayhi kān bā'inan bi-l-kaṯīr
DEF-player REL.masc.sg fire DEF-widow.maker DEF-fire upon-3sg.masc be.PRF.3sg.masc appear-ACC in-DEF-much
The player that Widowmaker shot (al-Murammil 'causes to be a widow') was too visible (Arabic speakers: I had trouble deciding whether to use bā'in, wāḍiḥ, or manẓūr here, thoughts?

يكره الجميع اللاعب الذي يشتكي من فريقه
yakrahu al-jamī3u al-lā3iba alāḏī yaštakī min fariqih
ya-krah-u al-jamī3-u al-lā3ib-a alaḏī ya-štakī min fariq-i-h
3sg.masc-hate DEF-collective DEF-player-ACC REL.masc.sg 3sg.masc-complain from team-GEN-3sg.masc
Everyone hates the player that complains about his team.

بعث عبدالرحمان من الموت اللاعب الذي أطلق المرمل النار عليه
ba3aṯa min al-mawti 3abdu al-raḥmān al-lā3iba alaḏī aṭlaqa al-Murammil al-nāra 3alayhi
ba3aṯa min al-mawti 3abdu al-raḥmān al-lā3iba alāḏī aṭlaqa al-Murammil al-nāra 3alayhi
send.3sg.PRF from DEF-death-GEN Abdelrahman DEF-player-ACC REL.masc.sg. fire.PRF.3sg Widowmaker DEF-fire-ACC upon-3sg
Abdelrahman resurrected (Arabic gamers how do you express this?) the player who Widowmaker shot.

The only instance in which the subclauses differ is by the number and gender of the referent nominal. Ie, feminine singular nouns take alātī, plural alawātī; masculine plural alāḏīna. The colloquial forms of Arabic simply this schema further: the relative pronoun becomes a single illī for all referents, regardless of gender or number:

كلنا نكره اللاعب اللي بيشتكي من فريقه
killenā nekrah il-lā3ib illī byeštekī min farigo
kill-nā ne-krah il-lā3ib illī b-yeštekī min farig-o
all-1.pl 1.pl-hate DEF-player REL IND-3sg.-complain from team-3sg.masc
We all hate the player who complains about his team.

البنت اللي اني مشتاق لها لاعبه فورتنايت
il-binit illi ani meštag laha lā3ibet Fortnite
DEF-girl REL 1sg love to-3sg.fem player-fem.CONS Fortnite
The girl that I love is a Fortnite player.

أنا وصحابي عايزين نقتل اللاعبين اللي غشوا
ana we soḥābī 3ayzīn ne'tol il-lā3ibīn illi ğaššū
1sg and friends.pl-1sg want-pl. 1pl.SUB-kill DEF-player-pl. REL cheat-PRF.3sg.pl.
My buddies and me want to kill the players who cheated.

[Yng]

It's alladhī not alādhī. I think مرئي جدا is probably the best translation for 'too visible' although out of context the English sentence sounds a bit funky to me too. But yeah, definite Arabic relative clauses are formed by a... subordinator? that acts very much like a definite substantive (it even has the definite article, al- or il- respectively, and is probably related to words like dhū X 'possessing X'). Although it doesn't have much case variation, in the cases where it distinguishes case it does so based on the main clause, not the subordinate clause, i.e. the 'relative pronoun' is in the main clause not the subordinate clause:

الرجلان الذان رئيتهما
ar-rajul-āni l-ladh-āni / raʾay-tu=humā
DEF-man-DU.NOM DEF-REL-DU.NOM see.PST-1sg=3du
The two men I saw

هل تقصد الرجلين الذين رأيتهما؟
hal ta-qṣud-u r-rajul-ayni l-ladh-ayni / raʾay-tu=humā
INT 2sg-mean.PRES-IND DEF-man-DU.ACC DEF-REL-DU.ACC see.PST-1sg=3DU
Do you mean the two men I saw?

But I'm not sure this is of direct relevance to the original question, since if we take the subclause as starting after the slash, the structure remains the same; it only differs (arguably) based on the role of the head noun in the subclause itself (if anything other than subject, a resumptive pronoun is required - but Arabic grammar in fact analyses both resumptive pronouns and verbal person marking as two categories of pronoun anyway, for fairly good syntactic reasons not worth going into here, making them structurally identical).