How does case inflection work on the head noun in internally headed relative clauses?

Question

I have read that some languages use internally headed relative clauses; so, for example, instead of saying "the man that we met yesterday went home today"; they position the shared noun in a relative clause, so they use a structure that literally translates into English as "we met the man yesterday went home today"; but I am wondering how does case declension work in those languages? I know some of those languages have cases (for instance I know Tibetan and Navajo use internally headed relative clauses and Tibetan has cases as well); but what I am wondering is how do they inflect the head noun so that its role in both the relative clause and the main clause is clear? it is easy to conceive of situations in which the noun although occurring in both clauses does not play the same role in them both; so how do languages with internally headed relative clauses express both cases?

Answer 1

Hittite puts the noun inside the relative clause, then puts a pronoun in the main clause. The case of this pronoun indicates the role of that noun in the main clause.

ÌR.MEŠ=YA=wa=za ku-ēs dā-s nu=war=as=kan kattanta pehute-t nu=war=as=mu arha upp-i
subjects=my=QUOT=INT which-N.PL take-2S.PST and=QUOT=3P.ACC=MOD away lead-2S.PST and=QUOT=3P.ACC=1S.DAT back send-IMP.SG
"My subjects that you have taken, you have led them away, send them back to me!"

Or in idiomatic English: "Send back my subjects that you have taken and led away!"

nasma=tta ^URUKÙ.BABBAR-sas ZAG-as ku-is BĒLU maninkuwan nu ERIN₂.MEŠ ANŠU.KUR.RA.MEŠ apē-dani wek-ti
or=2S.LOC Hattusa-GEN border-GEN who-N.SG lord-N.SG near and soldiers chariots that-DAT.SG request-2S
"Or the lord of the borders of Hattusa (NOM) who is near to you, ask that one (DAT) for soldiers and chariots"

Or in idiomatic English: "alternately, ask the border-keeper who is closest to you for soldiers and chariots". The noun is in the nominative since it's the subject of the relative clause; the pronoun is in the dative because it's the indirect object of the main clause.

If necessary, the noun can also be repeated in the main clause; this is helpful when there are several relative clauses in a row, to emphasize which one goes where.

^dU-as kuē-dani UD-ti hatuga tethi-ski-t […] ANA ^ĜIŠGIGIR-ya=kan kuē-dani apē-dani UD-ti ar-hahat […] ^ĜIŠGIGIR-ya tūriyan apātt=a dāi-r
stormgod-N.SG which-LOC.SG day-LOC.SG fearsomely thunder-ITER-3S.PST […] LOC chariot=MOD which-LOC that-LOC day-LOC stand-1SG.PST […] chariot-ACC harnessed there-and take-3PL.PST
"The day on which the Storm-God thundered furiously, the chariot on which I stood on that day, they took the harnessed chariot"

Or in idiomatic English: "they took the harnessed chariot that I stood on on the day when the Storm-God thundered furiously".

Answer 2

I think the Wikipedia article on relative clauses has a fairly good summary of different types of relative clause. I will refer to some of its categorizations below, but use them differently.

The term relative clause seems to cover the strategy which most languages can use to pragmatically and syntactically combine two full predications that share a common noun into one sentence. In many languages, the noun that heads the relative clause remains in the matrix clause. In other languages, the head noun remains in the subordinate clause. The latter seems to be what is called "internally headed relative clause" and what Wikipedia calls a relative clause of the unreduced type. Some languages use both types.

From my somewhat limited knowledge, I have not found any characteristic that is exclusive to internally headed relative clauses. Languages employing such clauses appear to use more or less the same range of strategies as languages with externally headed clauses.

Somewhat independently of where the head noun remains, different languages have different strategies of semantically combining the clauses: gapping, pronoun replacement, and nominalization. Syntactically, there may or may not be some additional overt morpheme added to the relative clause to indicate the subordination.

With the gapping strategy, used in Mandarin and English, for example, the listener/reader can recover the function of the noun missing from the relative clause by noticing what is missing. An example from English is: "I saw the man that I was working on the project for." The listener/receiver notices the gap appearing after the preposition "for" and knows that what is missing refers to the head noun.

If the referent is low on the accessibility hierarchy, the language may require another strategy to improve "accessibility," such as adding a co-referring pronoun or changing the form of the verb to change its valency. In English, for example, we can front prepositional phrases that refer to a head noun as in "I saw the man for whom I was working on the project."

With the pronoun-replacement strategy, a co-referring pronoun or related morpheme can replace the shared noun in the other clause to reveal its syntactic role in that clause. In languages with relative pronouns, these pronouns also indicate syntactic subordination.

Some languages have explicit correlative morphemes, allowing the shared noun to appear in either or both clauses in connection with such morphemes. Latin, for example, sometimes uses a correlative strategy, but sometimes just has the head noun in whatever clause is first. (See, Allen & Greenough) The following is an example of a Latin sentence with an internal head that also uses a gap strategy. :

quae prīma innocentis mihi dēfēnsiō est oblāta suscēpī; (Sull. 92)

(which first of-an-innocent to-me defense was offered I-undertook)

I undertook the first defense of an innocent man that was offered me.

Such a gapping strategy is necessary in this sentence since suscēpī (undertook) is a transitive verb and is lacking an overt object in the required accusative case in the matrix clause. The relative adjective quae is in the nominative case and modifies defensiō, the subject of "was offered." Most commonly in such cases, Latin would have a correlative pronoun in the main clause, especially in cases where the clauses are not so compact. An example is:

“quae pars cīvitātis calamitātem populō Rōmānō intulerat, ea prīnceps poenās persolvit ” (B. G. 1.12)

(which part of-the-state calamity to-the-people to-the-Roman had-brought, that first the-penalties paid)

That part of the state which had brought disaster on the Roman people was the first to pay the penalty.

The words quae and ea are correlatives, sharing the same gender, number, and underlying morpheme. I think such a strategy is less used in later Latin, but is the norm in Sanskrit.

With nominalization, a language can make use of any of its normal noun modifiers in addition to the basic verbal form that is nominalized to help disambiguate the noun's function in the relative clause. An example of a noun modifier might by a word or phrase identifying an additional agent, such as in English if we turn "The warrior who killed his enemy..." into "the warrior who, upon his killing of the enemy,...." An example of modifying the underlying verb form would be to change it from active to passive. In English we could change "the warrior who killed his enemy" into "the warrior who, with the enemy having been killed by him,...."

According to a grammar I have, Tibetan uses a nominalization strategy to create relative clauses that optionally can result in an internally headed relative clause. I have only flipped through a grammar of the language, but it seems to use both a gapping strategy and different nominalizers optionally to disambiguate the noun's function in the relative clause. Some of the nominalizers are derived from nouns that mean or meant such things as "cause" or "place," but can also indicate "agent," for instance. It seems that normally such precision is not felt to be necessary. For more details, you can refer to Manual of Standard Tibetan by Nicolas Tournadre & Sangda Dorje that discuss relatives in modern Lhasa Tibetan and in the literary language.

Vai, which can be considered a "northern Mande" language, is spoken in Liberia and Sierra Leone and also uses internally headed relative clauses. The word order in the language is: (topic) + subject (+ TAM auxiliary) (+ object) + verb + (time/place noun/phrase). It generally uses a strategy of using resumptive pronouns in the matrix clause and using a relative marker (mú) after the shared noun in the relative clause. An example using modified IPA (the apostrophes below represent an optional [l/] that is usually unpronounced) might be:

Kpɔ̀ŋ fà'á'à kɔ̀ŋkpɔ́ mú 'á kó'à, à wá tó'à à ɖá 'ɔ̀.

(monkey died [at] nut which {possessive-marker} sake, he/him/it {intensfier} remained [at] he/him/it mouth inside.

(The monkey died for which nut's sake, that remained in its mouth)

The nut the monkey died for remained in its mouth.

In this sentence, the relative clause acts like a topic before the matrix clause. Notice that the word for nut remains internal to its clause, preceded by a verb and followed by a "postpositional phrase," and that the matrix clause contains two pronouns. Notice also that the two identical pronouns (i.e., "à") in the second clause refer to different entities, and yet the pragmatics makes clear which one is the subject of the verb meaning "to remain" (i.e., "tó'à) and which one acts as a possessive pronoun for the inalienable noun "mouth" (i.e., "ɖá").

I am not certain about the situation in Navajo; however, I believe that both independent verbs and verbs in relative clauses in that language have obligatory argument markers that would normally disambiguate the function of the head noun for each verb. In other words, it obligatorily uses pronoun morphemes in both verbs to co-refer to the noun and disambiguate the syntax that applies to each clause.

One way in which internally headed and externally headed relative clauses may differ is in allowing for different quantification of the noun head. This linked paper compares such clauses in Navaho (which may not allow externally headed clauses) and Japanese (which can apparently have both types).