The Origin and Architecture of Existential Indeterminates in Okinawan

In a number of languages, an indeterminate is combined with various particles to yield different indefinite pronouns. This has been called an indeterminate system (Kuroda 1965, Cheng 1991, Haspelmath 1997, Jayaseelan 2001). As Haspelmath (1997) and Jayaseelan (2001) observe, existential indeterminates are often built with disjunction markers. On the other hand, a disjunction particle and a question particle are often morphologically identical cross-linguistically (see Hagstrom 1998, Jayaseelan 2001). Thus, a question that I ask here is whether the alleged homophony between a disjunction marker and a marker that forms an existential quantifier is principled (Jayaseelan 2001, Szabolcsi et al. 2014) or coincidental (Haspelmath 1997, Cable 2010). In this paper, I argue that the observation about homophony is misguided and hence support Haspelmath’s hypothesis, based on the data obtained from my fieldwork on Okinawan, an endangered Ryukyuan language. I propose an analysis where existential indeterminates in Okinawan have a clausal structure of an embedded question and are derived by deletion.


Introduction.
In a number of languages, an indeterminate is combined with various particles to yield different indefinite pronouns. This has been called an indeterminate system (Kuroda 1965, Cheng 1991, Haspelmath 1997, Jayaseelan 2001. In Malayalam, Japanese, and Nanay, for example, an indeterminate 'who' is interpreted as existential 'someone', when combined with a disjunction particle (1b-3b). In Basque and Russian, an indeterminate 'who' receives a free choice interpretation when combined with a disjunction particle (4b-5b) (the data are cited from Haspelmath1997).
(1) a. aarə 'who' b. aar(ə)-oo 'someone' (Malayalam) (2) a. dare 'who' b. dare-ka 'someone' (Japanese) (3) a. uj 'who' b. uj-nuu 'someone' (Nanay) (4) a. nor 'who' b. edo-nor 'whoever' (Basque) (5) a. kto 'who' b. kto-libo 'whoever' As Haspelmath (1997) and Jayaseelan (2001) observe, existential indeterminates and free choice indeterminates are often built with disjunction markers. On the other hand, a disjunction particle and a question particle are often morphologically identical cross-linguistically (see Hagstrom 1998, Jayaseelan 2001. Thus, a question that I ask here is whether the alleged homophony between a disjunction marker and a marker that forms an existential quantifier is principled (Jayaseelan 2001, Szabolcsi et al. 2014 or coincidental (Haspelmath 1997). In this paper, I argue that the observation about homophony is misguided and hence support Haspelmath's hypothesis, based on the data obtained from my fieldwork on Okinawan, an endangered Ryukyuan language. I propose an analysis where existential indeterminates in Okinawan have a clausal structure of an embedded question and are derived by deletion.
(2) Japanese a. dare 'who' b. dare-mo 'any/everyone c. Taro-mo 'Taro, too' (3) Japanese a. dare 'who' b. dare-ka 'someone' c. Taro-ka Hanako-ka 'Taro or Hanako' While it has an additive conjunction particle n (the cognate of mo 'also' in Japanese), which is used to build a universal quantifier and an NPI, Okinawan lacks a simple morpheme that expresses nominal disjunction and the morpheme gana used for building existential indeterminates does not have a disjunctive function.
(4) Okinawan a. taa 'who' b. taa-n 'any/everyone' c. Taraa-n 'Taraa, too' On the other hand, the language does have a sentence-final question particle ga (for wh-question) and (m)i (for yes-no question), as (6a) indicates. This question particle is distinct from the yes-no question particle -mi as shown in (6b). Importantly, however, these particles cannot build an existential indeterminates (or disjoin phrases), either, as shown in (7).
Jiraa-NOM go Q 'Does Jiraa go?' Clausal disjunction in alternative question in Okinawan does not employ any overt marker.  Table 2. disjunction, question, and existential markers in Japanese and Okinawan These facts show that (i) Okinawan lacks a disjunction particle and hence that (ii) its question markers cannot build existential indeterminates, contrary to the languages listed in (1).

Existential indeterminates in Okinawan as clausal questions.
As noted in the previous sections, existential indeterminates in Okinawan employ the mysterious morpheme gana, which is not a disjunction particle.
(9) Okinawan Taa-gana-ga ch-an. who-GANA-NOM come-PST 'Someone came.' Although the morpheme gana is likely to be composed of the Q-particle ga and na, the status of the latter is not immediately clear. Putting aside gana for a moment, it is interesting to look at another form of existential indeterminate in Okinawan, as shown in (10).
(10) Okinawan Taa-gayara ch-an. who-GAYARA come-PST 'Someone came.' Importantly, this form is identical to an embedded self-question structure, as shown in (11). 3 Evidently, gayara in (11) is syntactically complex: ga is a Q-complementizer, yara is a conditional form of copula.
[Taa ga ya-ra] wakara-n. who Q COP-RA know-NEG.PRS 'I don't know who it is.' b. Taa ga ya-ra. who Q COP-RA '(I wonder/I don't know) Who it is.' As described in detail in Miyara (2000Miyara ( , 2007Miyara ( , 2015a, Karimata & Shimabukuro (2007), Shinzato & Serafim (2013), van der Lubbe (2017) Wh-question in (12a) is information-seeking. The Q-particle appears at the end of the clause as a Q-complementizer and requires a wh-phrase. In contrast, focus-concord question in (12b) is a self-question and hence not information-seeking. The Q-particle ga necessarily attaches to the focus of the sentence and the predicate takes the focus-concord suffix -ra (see Miyara 2000Miyara , 2007, 2015a for a detailed description). Thus, the existential indeterminate in (10) and the selfquestion in (11) are both derived from the focus-concord question construction in (12b). Interestingly, Okinawan employs this self-question structure in (11) and (12b) to build the existential indeterminate in (10). More specifically, I propose that what looks like an existential indeterminate in (10) is derived from an embedded self-question 'I wonder / don't know wh-it is' by dunno-deletion and question-internal ellipsis (or pseudo-sluicing) inside the embedded question (Ross 1969;see, Inamine 2005, Miyara 2007 [Uree Taa ga ya-ra] wakara-n shiga, chuu ndi doo. it.TOP who Q COP-RA know-NEG but come C.say SFP 'I heard that someone would come.' (lit.) I don't know who it is, (I heard) (he/she) will come.' 4 While the sentence-final Q-complementizer ga in Okinawan in (12a) requires a wh-phrase, the Q-particle ga in (12b) does not. 5 gayara, yarawan, and yatin are not always interchangeable, as the latter two are unconditional/free choice forms.

(14) Okinawan
[Uree Taa ga ya-ra] wakara-n shiga, chuu ndi doo. it.TOP who Q COP-RA know-NEG but come C.say SFP 'I heard that someone would come.' (lit.) I don't know who it is, (I heard) (he/she) will come.' The predicate wakara-n (shiga) 'I don't know', which takes the embedded self-question, is deleted. The embedded question has the following structure in (15). The topicalized nominal demonstrative, which refers to the antecedent clause, undergoes ellipsis. The Q-particle attaches to the wh-phrase taa 'who' and agrees with the sentence-final focus-suffix -ra (focus concord).
Haspelmath makes an important observation that existential quantifiers in some European languages have a sentential structure 'I don't know who', as their source. This is illustrated in (16).
Fin 6 … ya f. někǔto 'somebody' <ne vě kǔto 'I don't know who' (Old Church Slavonic) g. je ne sais quel 'some kind of' <cf. je ne sais (pas) quel 'I don't know which' (French) Haspelmath refers to this type of grammaticalization path of existential quantifiers as dunno-type and observes that it is restricted to the European languages, but our analysis in (13)-(14) demonstrates that it is a general mechanism, observed outside the particular language family (see also Section 5 on Japanese).
I argue that the lack of a disjunction particle and the question-based existential indeterminates provide strong evidence against the homophony between a disjunction particle and a particle that forms existential indeterminates. Rather, it must be the Q-complementizer that plays a crucial role in deriving existential indeterminates in Okinawan. 7 Finally, let us return to gana. As mentioned at the beginning of this section, ga should be a Q-particle of a focus-concord question. I propose that gana is derived from ga ya-ra in (17).
(17) Okinawan ga ya-ra Q COP-RA Suppose that the underlying structure of taa-gana 'someone' is taa ga ya-ra in (14). If deletion applies to the copula ya alone, leaving the focus-concord suffix intact, then we obtain the form *taa-gara in (17b). Assuming that /r/ undergoes phonological change to /n/, we obtain the form taa-gana in (17c).
(19) Okinoerabu Ryukyuan (Van der Lubbe 2017, 303) 9 Ichi ki-n gara waka-ra-n. when come-IND GARA know-NEG-IND 'I don't know when he will come.' Furthermore, the analysis in (18) is also independently supported by the fact that the counterpart of gana in Irabu is gagara (see footnote 7). 10 (i) Irabu (Shimoji 2008, 242-243) Tau-gagara-nu-du sïn-tar-ca. who-GAGARA-NOM-FOC die-PST-HS 'Someone has died, they say.' 5. Japanese existential indeterminates as embedded question. The analysis of existential indeterminates as hidden embedded question is corroborated by Japanese. Nakanishi & Hiraiwa (2019) and Hiraiwa & Nakanishi (to appear) argue that there is indeed evidence that they are clausal in its origin and the particle ka is a Q-complementizer. The existential indeterminate dare-ka in example (20a) has long been considered to be an indeterminate combined with a nominal disjunction particle ka. However, we also find a semantically equivalent clausal expression (20b) in Japanese, just like in Okinawan. If we apply dunno-deletion, we obtain (20). If the copula is further deleted, we obtain the form (20d), which is identical to (20a).
(20) Japanese (Hiraiwa & Nakanishi to appear) a. Two other forms yarawan and yatin can also be used for expressing what amounts to disjunction. They are also concessive conditional (or unconditional) clauses and the latter is also used as free choice expression (see table 1).
(ii) Okinawan a. Pan ya-ra-wa-n kume ya-ra-wa-n masiyasi kara irabee. bread COP-RA-COND-also rice COP-RA-TOP-also better from choose.IMP 'Choose what you like better, whether bread or rice.' b. Pan yati-n kume yati-n masiyasi kara irabee. bread COP.COND-also rice COP.COND -also better from choose.IMP 'Choose what you like better, whether bread or rice.' The forms in (21) cannot be explained if ka were a nominal disjunction particle, because then it would not be able to be selected by the verb 'know'.
From the perspective of the proposed analysis of existential indeterminates in Okinawan, the closest Japanese counterpart of the Okinawan example in (14) is (22) Just as the form ga-ya-ra in Okinawan is used as a self-question, the form yara is also used as a self-question in Japanese.
(23) Japanese (Hiraiwa & Nakanishi to appear, footnote 12) Dare-ga kuru no yara. who-NOM come C YARA 'I wonder/don't know who will come.' 6. Conclusion. In this paper, I have demonstrated that existential indeterminates in Okinawan have a clausal structure. The morpheme gana in Okinawan is neither a disjunction particle nor a Q-particle, lending support for the claim that the alleged morphological affinity between disjunction particles and morphemes building existential quantifiers is coincidental. Rather, I have argued that existential indeterminates in Okinawan are syntactically derived from an embedded focus-concord self-question wh-ga...ya ra through dunno-deletion. It is claimed, then, that the form gana has emerged from deletion of the copula ya and liquid-nasal alternation.