Zero derivation in Ainu

Ainu is supposedly a polysynthetic language, but displays a large number of characteristics that, at least when compared with the prototypical North American polysynthetic language, don’t really line up in my opinion.

I’ve long had a suspicion that Ainu has zero derivation, but didn’t have a lot of particularly good examples. One of my go-to ones up to this point has been the word omke ‘cough’, which is both a noun and a verb. The other night, I came across what I think is a much better example:

(1) a-un-kor-e-∅ ka somo ∅-∅-ki1 ruwe ∅-∅-ne wa.
INDF.A-1PL.O-have-CAUS-NMLZ even not 3.A-3.O-do FACT 3.A-3.O-COP EMPH
‘It was the case that we weren’t given it.’ (Nakagawa and Nakamoto 2004: 72)

I think there is one strong line of evidence for this being zero derivation of a verb to a noun, and one weaker one. The stronger of the two is the fact that this cannot be a relative clause, and the weaker is the non-occurrence of the complementizer hi.

Relative clauses in Ainu are modifier-head in their order. We have a verb, aunkore ‘we weren’t given any’, a potential relative clause but not a head and a particle ka ‘even’, along with the preceding verb, also a potential relative clause but not a head. Note that the adverb somo ‘not’ this negates the following verb (here, ki ‘to do’), so clearly, we have no potential heads for aunkore ka to be a relative clause that is the object of ki.

There is an alternate interpretation here, that the complementizer hi is just not expressed here for some reason. This would not be unheard of. For instance, compare the following in English:

(2) He said that I fell.

(3) He said I fell.

Both are perfectly grammatical. In Ainu, however, we do not find this variation. I was unable to locate any examples of the complementizer hi followed by even the lexical verb ki ‘to do’. So while it could be the case that we just lack a complementizer here, I think instead we are dealing with zero derivation.


1. As an aside, ki here functions as a light verb. I’m not entirely sure if in this capacity it is still transitive, but I have glossed it as such. Clearly, in its function as the lexical verb ‘to do’, we have good evidence that ki is indeed a transitive verb.

For instance, it takes transitive agreement markers (Izutsu 2003, a corpus of Asahikawa dialect Ainu, has examples of both an-∅-ki |INDF.A-3.O-do| and ci-∅-ki |1PL.A-3.O-do|, and no examples of *ki-an |do-INDF.S| or *ki-as |do-1PL.S|, and can be antipassivized (cf. i-ki-an |APASS-do-INDF.S|, also many examples in Izutsu 2003). But to the best of my knowledge, and certainly in Izutsu 2003’s corpus of Asahikawa Ainu, no distinction is made between the light verb ki and the lexical verb ki, so I’m just not sure.


1 – first person
3 – third person
A – Agent-like argument (“subject”) of a transitive verb
APASS – antipassive
CAUS – causative
COP – copula
EMPH – emphatic
FACT – factual/inferential evidential
INDF – indefinite person
NMLZ – nominalizer
O – Patient-like argument (“object”) of a transitive verb
PL – plural
S – single argument of an intransitive verb


Izutsu, Katsunobu. 2003. Ainugo Asahikawa Hōgen Kōpasu ni Motozuku Bunpōsho Hensan no Tame no Kiso Kenkyū [Basic Research for Compiling a Grammar Based on a Corpus of Asahikawa Dialect Ainu]. Asahikawa: Hokkaidō Kyōiku Daigaku Asahikawa-kō.

Simple verbs in Okinawan aren’t so simple

In the last post on this topic, I glossed over the development of the Modern Okinawan simple (imperfective) realis form from an earlier continuous realis form. This will be a short post simply describing what I think the series of phonological and morphological changes involved are.

For illustrative purposes, we’ll use the verb root √kak– ‘to write’. The original form would have been something like *kak-i+wor-um |write-INF+CONT-RLS|.

First, the evidence we have for the realis originally being *-um. In both Western and Eastern Old Japanese, we only find -u (or -i with r-irregular verbs) as what Vovin (2009: 595) calls the final predication form (what I call the realis form). However, the situation is different in the Ryūkyūan subgroup. In Okinawan, we find a number of forms ending in a nasal, as in the simple realis form, given in (1):

(1) kach-un |write-RLS| ‘[Someone] writes.’

The most crucial piece of evidence, however, is the realis form followed by the general question clitic =i. For instance:

(2) kach-um=i |write-RLS=GQ| ‘Does [someone] write?’

It appears that at some point in the history of Okinawan the distinction between syllable-final -m and -n was neutralized and we only find syllable-final -n. But in the form with the general question enclitic, there is resyllabification, and instead of being syllable-final, the underlying -m is now syllable-initial, and it is now licit for it to appear overtly as m. I will represent this underlying form as -uM.

The next point we need to explain is the palatalization of the verb root *√kak- with the realis suffix -un. There appears to be no overt synchronic motivation, but a diachronic explanation of a following infinitive form motivating the palatalization to *kat͡ɕ-i ( < *kakʲ-i. But -uM should not prompt the infinitive. The only reason we would expect the infinitive in this context is with an auxiliary verb. In Western Old Japanese, for instance, we find that nearly all of auxiliaries and some converbs following this pattern (such as the clause chaining converb -te, which always follows an infinitive), and even in Modern Japanese, we find compound verbs formed on this pattern (for instance, tob-i+kom-u |jump-INF+into-IPFV| ‘[Someone] jumps into [something]’).

Another neutralization involved the auxiliary verb wor- |CONT| (as a lexical verb ‘to exist’). This was caused by the addition of a pair of rules. The first mandated consonantal onsets to all morpheme-initial positions. For vowel-initial morphemes, this was satisfied by the addition of a glottal stop. Thus, PR *ame ‘rain’ (<  PJ *Samay1 ‘rain’) became pre-Modern Okinawan *ʔame ‘id.’. Additionally, the initial consonant of wo, especially after raising to wu, became optional and could be replaced by a hiatus. wor- is unaccented, and thus high register in all Okinawan varieties, but although short vowels tend to be higher in pitch than their long counterparts (an apparent phonetic universal), it is not clear how pitch would have interacted here (as the pitch accent pattern of a main verb + auxiliary verb compound would likely be different than the root verb).

The most contentious point is the V2 elision, due to the fact that V2 elision is a typological rare phenomenon. V2 elision is a process where a sequence of two vowels, V1 and V2, results in the second vowel, V2, being deleted and the first vowel, V1, remaining. I propose that in this case, the motivation is the relatively higher sonority of *i versus *o (or, if it raises in the Northern Ryūkyūan chain shift before this point, *u, which is even less sonorous). In any case, both the infinitive *-i and the entirety of the auxiliary of *or- must both delete, as there is no residue of either in Modern Okinawan.

As I just mentioned, the initial vowel of the auxiliary *or- |CONT| then deletes, and creates an illegal consonant cluster (all non-geminate consonant clusters are illegal in Modern Okinawan), which then simplifies by deleting the second consonant.

We are then left with just the realis form -uM, which becomes the new simple (imperfective) form of the verb, and does double duty contrasting with both the perfective aspect (an expansion from its original continuous aspect meaning), and with various other mood markers (like the adnominal2 -uru and the tentative -ura)

So schematically:

(3) *kak-i+wor-um
(introduction of contrast between o with a non-glottal onset, (w)o, and with a glottal onset, ʔo; hiatus optionally replaces glide)
(neutralization of syllable-final -n/-m contrast)
kakʲ-i+or-un (palatalization of k- due to following -i)
kakʲ-i+r-un (V2 elision of o)
kakʲ+r-un (deletion of due to palatalization of kʲ-)
kakʲ+un (deletion of r due to illegal consonant cluster)
kakʲ-un (reanalysis of auxiliary as suffix)
(affrication of kʲ- to t͡ɕ-)


(1) The *S stands for some sort of fricative. In certain compounds, ‘rain’ shows up with an initial s-. For instance, WOJ pîsamë ‘hail’ (< ‘ice’ +samë ‘rain’). Vovin (2010: 109) speculates that this may be PJ *z or *h.

(2) I consider the adnominal form a mood marker due to its interaction with so-called focus particles. In addition to being used to form relative clauses, it is obligatory when the emphatic focus particle =du is used in a sentence, like how the realis mood is typically obligatory, or how the tentative mood is used when the general question particle is unraised. Compare (4a), (b), and (c):

(4a) sumuchi=∅ kach-un
book=ACC write-RLS
‘[Someone] writes a book.’

(4b) sumuchi=du kach-uru
book=EFP write-ADN
‘[Someone] writes a book

(4c) sumuchi=ga kach-ura
book=IQ write-TENT
‘[Someone] writes a book?’


Vovin, Alexander. 2009. A Descriptive and Comparative Grammar of Western Old Japanese. Part Two: Adjectives, Verbs, Adverbs, Conjunctions, Particles, Postpositions. Folkestone: Global Oriental.

Vovin, Alexander. 2010. Koreo-Japonica: A Re-Evaluation of a Common Genetic Origin. Folkestone: Global Oriental.


∅ – null
– – morpheme boundary
+ – compound boundary
ACC – accusative case
ADN – adnominal mood marker
CONT – continuous aspect marker
EFP – emphatic focus particle
GQ – general question (yes/no-question) particle
INF – infitivie
IPFV – imperfective aspect marker
IQ – information question (wh-question) particle
RLS – realis mood marker
TENT – tentative mood marker

Kakari-Musubi — When a non-finite sentence is finite

In this post, I’d like to talk more generally about kakari-musubi (係り結び ‘binding’) constructions, which I’ll abbreviate KMCs.

So, first of all, a bit of a refresher as to what these phenomenon are. A prototypical KMC is one where a kakari particle (係助詞 kakari joshi) occurs, and where the matrix verbal of the sentence takes on a different marking than would otherwise be expected. Despite this definition, traditionally, the topic particle (=pa in Western Old Japanese) and the additive-scalar focus particle (=mö in Western Old Japanese) have also been considered kakari particles, despite the fact that they trigger the expected endings.

The following table summarizes the situation in Western Old Japanese:

[table class=”table table-striped”]
Particle, Meaning, Verb form
=pa, topic, FIN
=mö, additive-scalar (also/even) focus, FIN
=ya, polar (yes/no) question, ATTR
=ka, content (wh-) question, ATTR
=sö~=nzö, emphasis, ATTR
=namo, emphasis, ATTR
=kösö, strong emphasis, EV

We find a similar, though interestingly not the same situation in Okinawan. In particular, the two question markers behave quite differently than any of the other particles. I’ll deal with Okinawan in particular at a later date, but here a table equivalent to the one above for Okinawan:

[table class=”table table-striped”]
Particle, Meaning, Verb form
=ya, topic, FIN
=n, additive-scalar (also/even) focus, FIN
-i, polar (yes/no) question, FIN
=ga, content (wh-) question, TENT
=du, emphasis, ATTR

In all of these cases, we would generally expect the final form (or some non-finite form) of a verbal, rather than these alternative forms. Some examples:

(1) 庭尓敷流雪波知敝之久 (TOP + FIN)
NIPA-ni pur-u YUKÎ=pa ti-pê sik-u
garden-LOC fall-ATTR snow=TOP thousand-layer.CLF cover-FIN
‘The snow that falls on the garden covers [the ground] in many (lit. a thousand) layers.’ (MYS 17:3960)

(2) 奈尓乎可於母波牟 (CQ + ATTR)
nani-wo=ka omöp-am-u
what-ACC=CQ think-TENT-ATTR
‘What would [you] think?’ (MYS 17:3967)

(3) 安連曽久夜思伎 (EMPH + ATTR)
are=sö kuyasi-kî
I=EMPH be.regretable-ATTR
‘I am regrettable.’ (MYS 17:3939)

(4) 雪己曽波春日消良米 (EMPH + EV)
YUKÎ=kösö=pa PARU PÎ KIY-Uram
snow=EMPH=TOP spring day melt-NPST.TENT-EV
‘Snow melts [on] a spring day.’ (MYS 9:1782)

Note that transcriptions in ALLCAPS indicates semantographic, rather than syllabic writing. So the character 雪 YUKΠis used to mean ‘snow’ (it’s semantic value), rather than any syllabic value that might be associated with it. Compare this to 波 pa, which is used for its phonetic value, not its meaning, ‘wave’.

(4) is doubly interesting because while both kösö and pa occur, going off of the verb ending, kösö takes precedence. Also, PARU PÎ ‘spring day’ is not marked for case. I have translated it here as if it is a locative, as WOJ kiy- ‘to vanish, to disappear; to melt (of snow or ice)’ is an intransitive verb. While Western Old Japanese has both differential subject and differential object marking, where under certain circumstances the subject and object markers do not occur, this is not the case for any sort of locative marking.

An interesting, and likely related, phenomenon is that these “anomalous” endings can occur even without a kakari particle triggering them. While most of could be easily explained away due to the other sort of modal information the endings contain, the attributive can occur as a marker of final predication without any trigger:

(5) 我衣手乃干時毛名寸
1sg=POSS sleeve=GEN dry-ATTR time=ASF be.not-ATTR
‘There is no time at all for my sleeves to dry.’ (MYS X:1994) (Vovin 2009: 624; glossing and transcription modified to match my own).

That being said, Vovin (2009: 626) compares the semantics of this sort of construction to constructions in modern Japanese with a verb, the inferrential evidential の no, and the copula だ da, so here too it might be that there is some sort of modal meaning that is triggering the change. This is, in any case, a very interesting sort of insubordination—a formally subordinate clause which is used as a main clause—as it in a way lines up with some of the functions of insubordinate clauses presented in Evans (2007), like expressing modality and evidentiality, but doesn’t resemble them in many other ways.

Note that these are the final two lines of MYS X:1994. There is no potential head that …nakî could be modifying as a relative clause. Also note that the additive-scalar focus particle  is misspelled as  in (5). For those worried that this might invalidate this poem’s status as an example, Vovin (2009: 623–632) cites 14 more examples in WOJ, and 7 examples from EOJ.

An earlier version of the article had a typo, saying Okinawan =ga required the attributive ending. Non-final =ga requires the tentative ending, as I described in an earlier article.

Evans, Nicholas. 2007. “Insubordination and its uses.” In: Irina Nikolaeva (ed.). Finiteness: Theoretical and Empirical Foundations. Oxford: Oxford University Press. pp. 366–431.

Vovin, Alexander. 2009. A Descriptive and Comparative Grammar of Western Old Japanese. Part Two: Adjectives, Verbs, Adverbs, Conjunctions, Particles, Prepositions. Folkestone: Global Oriental.

Misbehavior in the mountains — The irregular structure of some verbs in Ainu

Ainu is fairly well-known for being a highly synthetic language (I dislike the term “polysynthetic” for a number of reasons), which has both noun incorporation and a set of three applicative prefixes.

Noun incorporation in Ainu is, with a few exceptions, fairly typologically typical. In the “template” of Ainu verbs, there is a slot where direct objects, whether they are in the form of pronominal agreement prefixes or incorporated nouns, fit. A very minimal transitive verb template is given in (1):


Where AAGR is any of the prefixes which mark agreement for the subject of a transitive verb, OAGR is any of the prefixes which mark agreement for the direct object of a transitive verb, OI.O is an incorporated noun, and V is the root verb.

(2a) and (2b) give more concrete examples:

(2a) Aunkore ka somo ki ruwe ne wa.
a-un-kor-e ka somo ∅-∅-ki ruwe ne wa
INDF.A-1PL.EXCL-give-CAUS even not 3.A-3.O-do FACT COP EMPH
‘The fact is, [she] didn’t even give us any.’ (Nakagawa and Nakamoto 2004)

(2b) To okari puy anetoyta kor.
to okari puy an--e-toy-ta kor
lake around marsh.marigold.root INDF.A-3.O-APL-earth-gather PROG
‘[I] was gathering marsh marigold roots from around the lake.’ (Izutsu and Tezuka 2006)

Both the subject and object slots seem to be obligatory (cf. Baker 1996), though in the third person there are no overt markers.

When nouns are incorporated, as one might expect, the valency of the verb is reduced by one. This means that only polyvalent verbs (transitives, causativized intransitives and transitives, etc.) should be able to incorporate nouns, as there cannot be a zero-valent verb. This also means that for these verbs to take new direct object, a valency increasing device—such as a causative or the applicative marker e-, which can function as a non-causative transitivizer (see Bugaeva 2010)—is needed. We see this in (2b).

(2b) also illustrates that applicatives alter the internal structure of a verb in Ainu. The root verb in that construction is ta ‘to gather (from under the surface)’, which is a transitive verb, so has the structure given in (1). With an incorporated noun, its structure is altered. Our best evidence for this is because the indefinite person and first person plural set of agreement markers have a tripartite alignment—that is, each has a separate marker for the subject of a transitive verb, the subject of an intransitive verb, and the direct object of a transitive verb. So we find the following sort of construction:

(3a) turep-ta-an
‘[Someone] gathers giant lilies.’

(3b) *an-turep-ta
‘[Someone] gathers giant lilies.’

As we can see in (2b), however, with the applicative e- and an incorporated noun, we now have a transitive verb again. The structure of a monomorphemic verb with an applicative is diagrammed below in (4a). A similar structural diagram for a verb with an incorporated noun and then an applicative is given in (4b).



To add to my abbreviations above, OAPL is the object agreement needed because of the addition of an applicative prefix and APL is the applicative prefix itself.

Some verbs, however, do not behave as we would expect. I have two examples: ekimne ‘to be in the mountains, to go into the mountains’ and eronne ‘to be on the ror (the seat of honor reserved for guests in a traditional Ainu home)’. The former is from e-kim-ne |APL-mountain-COP|, and the later is from e-ror-ne |APL-seat.of.honor-COP|, with the expected change of [ɾ] → [n] / _ [n] from one of the several sandhi rules found in Ainu. Some examples in context:

(4) ∅-∅-ne=no sir-an kor anakne, somo e-kim-ne-an=pe ∅-∅-ne.
3.S-3.O-COP=ADVZ weather-exist.SG PROG TOP NEG APL-mountain-COP-INDF.S=NMLZ 3.S-3.O-COP
‘When the weather had become so, it was that I didn’t go into the mountains.’ (Nakagawa 2001: 82)

(4b) o-tu kes pa-ta / o-re kes pa-ta / e-kim-ne-an kor
APL-two after year-LOC APL-three after year-LOC APL-mountain-COP-INDF.S PROG
‘After two years, / after three years, / I was in the mountains,[and…]’  (Nakagawa 2008: 296)

(This is from a yukar, an Ainu oral epic. The slashes here indicate metrical breaks.)

Instead of the pattern we see in (2b), that of a transitive verb de-transitivized by an incorporated noun and re-transitivized by an applicative, we see the same pattern as in (3a), a transitive verb de-transitivized by an incorporated noun, and—despite the presence of an applicative—no re-transitivization.

So their structure seems to be:


Where S is most of the prefixes for marking agreement with intransitive subjects, and SITR is the limited set of markers for intransitive subjects which are suffixes.

At this time, I have no good explanation for why this occurs. Simply put: these verbs misbehave.


Baker, Mark. 1996. The Polysynthesis Parameter. Oxford: Oxford University Press.

Bugaeva, Anna. 2010. Ainu applicatives in typological perspective. Studies in Language 34:4. pp. 749–801. DOI: 10.1075/sl.34.4.01bug

Izutsu, Katsunobu and Yoritaka Tezuka. 2006. Kiso Ainugo [Basic Ainu]. Sapporo: Sapporo Dōshoten.

Nakagawa, Hiroshi. 2001. Ainu Kōshōbungei Tekisuto Shū 1: Shirasawa Nabe Kōjutsu — Ōkami kara Nigareta Musume [Ainu Oral Literature Text 1: Narrated by Nabe Shirasawa — The Girl who Escaped from Wolves]. Chiba Daigaku Yūrashia Gengo Bunka Ronkōza [Chiba University Journal of Eurasian Languages and Cultures] 3, pp. 52–66.

Nakagawa, Hiroshi. 2008. Ainu Kōshōbungei Tekisuto Shū 8: Shirasawa Nabe Kōjutsu —Yukar Irupaye: Sinutapka-jin, Ishikari-jin to Tatakau [[Ainu Oral Literature Text 8: Narrated by Nabe Shirasawa — Yukar Irupaye: The Sinutapka Man Fights the Ishikari Man]. Chiba Daigaku Yūrashia Gengo Bunka Ronkōza [Chiba University Journal of Eurasian Languages and Cultures] 10, pp. 291–313.

Nakagawa, Hiroshi and Mutsuko Nakamoto. 2004. CD Ekusupuresu Ainugo [CD Express Ainu]. Tōkyō: Hakusuisha.

Relative clauses in Okinawan

Last time, I mentioned that Okinawan behaves very differently than Japanese in terms of how it morphologically marks relative clauses. In Japanese, the verbal form which modifies the head of the relative clause is marked the same as a verb which is the head of a matrix clause—that is, it takes tense-aspect-mood marking, but nothing else special. For instance:

(1) kak-u hon
write-IPFV book
‘the book that (somebody) writes’

(2) hon-o kak-u
book-ACC write-IPFV
‘(Somebody) writes a book.’

In both cases, we see the verb kak- ‘to write’ appearing with the same ending, the imperfective aspect marker -u.

The situation is markedly different in Okinawan:

(3) kach-uru sumuchi
write-ATTR book
‘the book that (somebody) writes’

(4) sumuchi kach-u-n
book write-IPFV-FIN
‘(Somebody) writes a book.’

Here, the verb kak- ‘to write’ has a very different form in the one construction versus the other. In the relative clause in (3), it takes what has been traditionally called the attributive form (or rentaikei 連体形). The attributive form is distinct from the final form (or shūshikei 終止形), as seen in (4), and the key difference between the two is that the attributive form almost exclusively appears on the final verb of a relative clause, while the final form appears exclusively on the final verb of a matrix clause.

I say “almost exclusively” for the attributive form, because there are some interesting cases of insubordination—that is, a seemingly non-finite clause used as a finite clause (see Evans 2007) in which it can occur. These are again kakari musubi (係り結び) phenomenon that we touched on last time, but with a different focus particle triggering them, and a different TAM marker resulting from the presence of said focus particle.

The focus particle du, which emphasizes the word or phrase it is attached to triggers the attributive form of the verb even in cases where we would expect the final form of the verb. Compare, for instance, (4) above with (5):

(5) sumuchi du kach-uru
book EF write-ATTR
‘(Somebody) writes a book.1

These sorts of phenomenon are interesting, and illustrate a little of my hesitation and struggle with finding a good label for the attributive and the final. We are not only dealing with the status of a verb in terms of being attributive—that is,  modifying things in the same way as a relative clause—versus final—that is, signaling the end a matrix clause, we are also dealing with issues of focus and epistemic modality.

Another interesting issue here arises when we compare the attributive forms of different verbs in Okinawan with their counterparts in Old Japanese:

[table class=”table table-striped”]
Okinawan, Old Japanese, gloss
tach-uru, tat-u, ‘to rise’
wata-y-uru, watar-u, ‘to cross’
chu-uru, k-uru (elision of *kö-uru), ‘to come’
chi-y-uru, kî-ru, ‘to wear’

Okinawan universally uses -uru to mark attributive forms, while Old Japanese has -u after consonant stem and r-irregular verbs, -ru after strong vowel stem verbs, -uru after other vowel stem verbs, and  after the copula n-. What exactly went on historically in Okinawan is not clear. Analogical leveling seems tempting at first glance, but we end up in these forms and in others with unexplained “residue”. Why, for instance, do we have palatalization of the second /t/ in ‘to rise’? Why are there seeming empty morphemes, |-y-|, with r-consonant stem and vowel stem verbs?

I’ll address this in a future post, as we’re getting pretty far from relative clauses.

1. This sort of focus would be indicated by intonation in English, represented here by the italics.

Evans, Nicholas. 2007. “Insubordination and its uses.” In: Irina Nikolaeva (ed.). Finiteness: Theoretical and Empirical Foundations. Oxford: Oxford University Press.

Questions in Okinawan

In Okinawan, there are multiple possible ways to mark a question. One key difference appears to be a distinction between polar (or yes-no) questions and content (or wh-) questions. Polar questions are marked by the verbal enclitic =i, while content questions are marked with the phrasal enclitic =ga. For instance, compare the following pair of sentences:

(1) maa=nkai ʔich-abii=ga
where=LOC go.IPFV-POL=CQ
‘Where are you going?’

(2) Naafa=nkai ʔich-abii-m=i
‘Will you go to Naha?’

However, unlike its cognate in modern Japanese, =ka, Okinawan =ga can also appear next to the phrase which it has scope over, not just the entire sentence. This in some ways resembles quantifier float. For instance:

(3) kunu yama=nu ʔuchee shichina-ʔudun=ga ya-yabii-ra
this forest=GEN inside\TOP Shichina-Udun=CQ COP-POL-TENT
‘Is Shichina Udun inside this forest?’ (Nakamatsu 1973: 78)

This is interesting in two ways.

First, the distinct between polar and content questions is lost. (3) is quite clearly a polar question, yet it does not use the polar question enclitic =i. Instead, it has the content question enclitic =ga. As far as I know, the enclitic =i is a verbal enclitic; that is, it can only attach to verb phrases. The enclitic =ga, on the other hand, can attach to a variety of different elements. At the least noun phrases, case phrases, and the infinitive form of verbs.

Second, there is a difference in the modal forms used in the sentences. There are a number of processes in Japonic languages, in the Japanese literature called kakari musubi (係り結び), where certain kinds of focus particles (including question particles) alter the mood of the main verb of the sentence. In (1), there is no mood suffix with =ga. However, with =i, as in (2), the final form -n ~ -m- (which marks the verb as being the matrix predicate of a sentence) is obligatory. Curiously, in (3)—and other cases where the content question clitic “floats”, the tentative mood marker -(u)ra is required.

I’m not really satisfied with any sort of explanation for what I’ve been calling the “final” suffix in any Japonic language. I think we have pretty good evidence that something mood-related is going on, as it is fairly clearly from all other cases of these sorts of kakari musubi phenomenon, in whatever variety of Japonic you want to look at, aside from this final suffix and another suffix called the attributive (which marks the verb phrase as being the modifier in a relative clause, as in (4a) and (b)), these are modal markers that are involved. However, what sort of mood we’re dealing with in terms of the final and the attributive I don’t think is very clear at this point.

(4a.) sumuchi yum-u-n
book read-IPFV-FIN
‘I read a book.’

(4b.) yum-uru sumuchi
read-ATTR book
‘the book that I read’

Nakamatsu, Takeo. 1973. Okinawago no Bunpō [Okinawan Grammar]. Naha: Okinawa Hōgen Bunka Kenkyū-zyo.