Elamur: **Mova** · [Parse](https://mova.elamur.ai/free/en/parse/) · [Dictionary](https://md.dictionary.elamur.ai/free/en/) · [Translate](https://md.translate.elamur.ai/free/en/) · [Research](https://md.research.elamur.ai/free/en/) · [Gali](https://md.gali.elamur.ai/free/en/) · [MCP](https://mcp.elamur.ai/) · [Locations](https://md.locations.elamur.ai/free/en/) · [Weather](https://md.weather.elamur.ai/free/en/)

[Mova](https://md.mova.elamur.ai/free/en/) · [this page in HTML](https://mova.elamur.ai/free/en/docs/languages/ja/ryu/08-word-classes/)

[Mova](https://md.mova.elamur.ai/free/en/) / [docs](https://md.mova.elamur.ai/free/en/docs/) / [languages](https://md.mova.elamur.ai/free/en/docs/languages/) / [Japanese](https://md.mova.elamur.ai/free/en/docs/languages/ja/) / [ryu](https://md.mova.elamur.ai/free/en/docs/languages/ja/ryu/) / 8. Word classes and grammatical annotation

# 8. Word classes and grammatical annotation

Central Okinawan word classes are distinguished by inflection, by the positions their members can occupy, and by the constructions they enter. A word’s class and its job in a particular clause are separate decisions. A time noun can modify a clause, for instance, while a relative clause can modify a noun. The proposed Universal Dependencies (UD) analysis records the class in `UPOS`, grammatical properties in `FEATS`, and syntactic relations in the dependency tree.

## Open and closed word classes

Nouns and lexical verbs are open classes. Nouns name people, places, things, and abstractions and can head nominal expressions. Lexical verbs supply the predicate of a clause and may license participants such as an actor, a theme, or a recipient. Property stems form the lexical basis of adjectival predicates. Adverbs include lexical modifiers, and ideophones provide another source of lexical expressions. A class can admit new members even when the inflection available to those members follows an established pattern.

Personal pronouns, demonstratives, and grammatical particles belong to series with conventional members. A pronoun can occupy a nominal position by itself. An adnominal demonstrative accompanies a noun and identifies its referent by deixis. Case, topic, and sentence-final particles have grammatical functions tied to their hosts and positions. Auxiliaries also have conventional grammatical uses, some of which coexist with lexical uses of the same form. Assign the class appropriate to the occurrence when a form has both uses.

| Form | Gloss | Translation | Distribution |
| --- | --- | --- | --- |
| `wan` | `1SG` | ‘I’ | Independent personal pronoun (constructed) |
| `kuri` | `DEM.PROX` | ‘this one’ | Independent demonstrative pronoun (constructed) |
| `kunu yaa` | `DEM.PROX.ATTR house` | ‘this house’ | Adnominal demonstrative with an overt noun (constructed) |
| `yaa` | `house` | ‘house’ | Common noun (constructed) |
| `Taruu` | `Taruu` | ‘Taruu’ | Personal name (constructed) |

Cardinal numerals form a system with rules for combining forms, so a finite stock of basic numeral elements can express a much larger range of values. Classifiers occur in counting expressions and contribute the unit of counting. Give the numeral and the classifier separate analyses when they are separate words. A lexical name of a unit can retain nominal distribution, including when it occurs with a number.

Chamberlain’s grammatical description gives an early account of the distinctions among nominals, predicates, and particles. The Ryukyuan survey provides the comparative setting for the central varieties. A distinction established for another variety supplies a point of comparison, while the class of a Central Okinawan form depends on its own distribution (Chamberlain 1895, grammar §§ on nouns and verbs; Shimoji and Pellard 2010, § Introduction).

## Distinguishing nouns, verbs and adjectival predicates

A noun can head an argument or a possessive expression and take the particles appropriate to that construction. In ordinary predication a nominal complement combines with a copula. The noun gives the identity or class ascribed to the subject, and the copula supplies the predication. The same class noun can head an argument in another clause. A name and an independent pronoun can occupy many of the same argument positions, so case marking alone establishes nominal distribution, with the finer distinction among noun, name, and pronoun made from the lexical item and its use (Chamberlain 1895, grammar §§ on nouns, pronouns and the substantive verb).

A lexical verb has forms selected by such distinctions as polarity, tense, and the relation of its clause to a following noun or clause. Compare forms of the same lexeme before deciding which part carries the grammatical distinction. A verb in an adnominal construction can retain argument structure and form a relative clause. Its position before a noun is therefore compatible with verbal status (Chamberlain 1895, grammar §§ on verbs and their conjugation).

Adjectival predicates express properties and have forms for predication and adnominal use. The `-san` and `-saru` forms below illustrate that contrast for one property lexeme. A finite adjectival predicate can head a clause directly. In an adnominal phrase, the property form modifies the noun. Predicate position and the ability to occur without an overt copula are shared by verbs and these adjectives. The inflectional series and the kind of predication together distinguish the classes. A property concept can also be expressed by a noun with a copula or by a lexical verb, so a translation such as ‘beautiful’ is only one clue to classification (Chamberlain 1895, grammar § on adjectives).

| Class and use | Form | Gloss | Translation |
| --- | --- | --- | --- |
| Nominal predicate | `Taruu=ya Uchinaanchu yan` | `Taruu=TOP Okinawan.person COP` | ‘Taruu is Okinawan’ (constructed) |
| Lexical verb, affirmative nonpast | `Taruu=ga kachun` | `Taruu=NOM write.NPST` | ‘Taruu writes’ (constructed) |
| Same verb, affirmative past | `Taruu=ga kachan` | `Taruu=NOM write.PST` | ‘Taruu wrote’ (constructed) |
| Same verb, negative nonpast | `Taruu=ga kakan` | `Taruu=NOM write.NEG.NPST` | ‘Taruu does not write’ (constructed) |
| Same verb, adnominal clause | `Taruu=ga kachuru jii` | `Taruu=NOM write.ADN character` | ‘a character that Taruu writes’ (constructed) |
| Adjectival predicate | `yaa=nu churasan` | `house=NOM be.beautiful.FIN` | ‘The house is beautiful’ (constructed) |
| Same adjective, adnominal modifier | `churasaru yaa` | `beautiful.ADN house` | ‘a beautiful house’ (constructed) |

The examples use Latin transcription. An equals sign shows a particle attached to a host, and a period in a gloss joins information carried by the same form. `ADN` describes an adnominal form and `FIN` a finite predicate form. These glosses describe the construction. The annotation proposal below specifies which distinctions go into UD features.

When a form is ambiguous in isolation, examine its inflection and the structure around it. An adjectival modifier and a relative clause can both precede a noun. An overt subject or object internal to the modifier supplies evidence for a clause, and the verbal paradigm can supply evidence even when participants are understood. For a property expression with a copula, the nominal analysis is appropriate if the property word has nominal distribution. The `-san` form in the table carries the adjectival inflection within a single word.

## Adverbials and ideophones

An adverbial is a modifier of a predicate, a clause, or another modifier. A lexical adverb fills that function directly. A noun used as a time or place expression, a phrase with a postposition, and a dependent clause can fill a similar function with a different internal structure. Test the word’s distribution beyond the given position. A time expression that can also head a nominal phrase with a particle has evidence for nominal status. Its clause-level use can then be represented by the dependency relation. An interrogative word receives an adverbial analysis when it asks about a circumstance such as time, place, or manner, and a pronominal analysis when it stands for a participant.

Ideophones evoke a sound, motion, appearance, or other sensory impression. Repetition can occur within an ideophonic expression, and the expression can occur as a manner modifier, as represented sound, or as an independent utterance. Repetition by itself leaves the class open, since other expressions can also be repeated. The construction gives the evidence for how to annotate the occurrence. An integrated manner ideophone can be an adverb, a representation of sound can occupy a quotative construction, and an independent exclamation can be an interjection (Chamberlain 1895, grammar §§ on adverbs and interjections).

| Form | Gloss | Translation | Analysis illustrated |
| --- | --- | --- | --- |
| `yukkui kachun` | `slowly write.NPST` | ‘[Someone] writes slowly’ | Lexical manner adverb (constructed) |
| `chuu Taruu=ga kachun` | `today Taruu=NOM write.NPST` | ‘Taruu will write today’ | Time nominal used adverbially (constructed) |
| `gaa-gaa` | `caw-caw` | ‘Caw, caw!’ | A sound imitated as an independent utterance (constructed) |
| `haisai` | `greeting` | ‘Hello!’ | Conventional greeting used independently (constructed) |

For a repeated ideophone, the spelling and the construction both matter. A single conventional word can contain repeated material. Repetitions uttered as separate tokens can have their own annotation. A lexicalized sound expression may also take nominal marking and function as a noun. The adverbial and interjection analyses apply when the expression has the distribution illustrated for those uses.

## Particles, auxiliaries and boundaries between words

A particle’s host can be a phrase. A case particle after a modified noun marks the whole nominal expression, and a topic particle can follow the expression whose referent is established as the topic. A sentence-final particle can contribute a speaker’s stance toward the whole utterance. The position of a particle next to a noun or verb in the written string gives only part of the evidence for its scope (Chamberlain 1895, grammar §§ on particles and postpositions).

| Form | Gloss | Translation | Function in the example |
| --- | --- | --- | --- |
| `Taruu=ga` | `Taruu=NOM` | ‘Taruu’ as subject | Nominal case marking (constructed) |
| `yaa=nu uchi` | `house=GEN inside` | ‘the inside of the house’ | Possessive or relational marking within a nominal expression (constructed) |
| `Taruu=ya kachun` | `Taruu=TOP write.NPST` | ‘As for Taruu, he writes’ | Topic marking (constructed) |
| `kachun=doo` | `write.NPST=SFP` | ‘[Someone] does write, you know’ | Sentence-final assertion directed to the addressee (constructed) |

The case readings in the tables belong to their constructions. In `yaa=nu churasan`, the marked noun phrase is the subject of a predicate. In `yaa=nu uchi`, it is dependent on a following noun. The same surface `nu` therefore supports distinct case analyses. Topic marking likewise has to be interpreted with the clause around it. A topic can correspond to an understood participant, and `ya` does not by itself say which participant that is.

For an auxiliary analysis, look for a grammatical contribution to a predicate whose lexical content comes from another word. Tense, aspect, polarity, and politeness can be expressed by a support form if the construction has one. A form with its own lexical predication and participants is a lexical verb in that use. A copula in a nominal predication has a special linking function and is assigned `AUX` in the proposal. An adjective that carries its own predicative inflection can serve directly as the predicate. The nominal and adjectival examples above illustrate the difference.

The proposed annotation uses syntactic words. A particle written against its host can receive its own word record when it is independently identifiable as a grammatical element. Distribution across different hosts, attachment at a phrase edge, and a stable grammatical function support that decision. An inflectional ending selected by a stem and belonging to its paradigm stays inside the word. In particular, the adjectival ending illustrated by `churasan` and the verbal inflection in `kachan` stay with their stems. A separately identifiable auxiliary receives a word record of its own. Spelling supplies the surface token, so where one written token contains two identifiable syntactic words, a UD multiword token can preserve the spelling while the two word records carry the analysis.

A boundary should also account for forms where adjacent elements have a fused realization. A regular, recoverable combination of two syntactic words can receive the same analysis as the corresponding unfused combination. A fused form belonging to a single inflectional paradigm is annotated as a single word. The examples give segmented forms where the aim is to show a particle’s host. The equals sign itself is a convention of this chapter and does not prescribe an orthographic space.

## Proposed UPOS and feature mappings

These mappings are proposed for Central Okinawan annotation. The descriptions of word classes and constructions provide the language-specific evidence. Grambank records comparative grammatical properties at the language level. UD here makes decisions about words in particular sentences, so a comparative property can motivate a distinction to check without determining a tag for an occurrence (Skirgård et al. 2023, § CLDF dataset, ParameterTable).

| UPOS | Assign to |
| --- | --- |
| `NOUN` | Common nouns, including time and place nouns used adverbially, and unit words with nominal distribution |
| `PROPN` | Names in their naming use |
| `PRON` | Personal pronouns and independent demonstrative, interrogative, or indefinite forms that occupy nominal positions |
| `DET` | Adnominal demonstratives and other grammatical determiners dependent on a nominal head |
| `NUM` | Cardinal numerals in counting or numerical use |
| `ADJ` | Adjectival predicates and adnominal forms of adjectival lexemes |
| `VERB` | Lexical verbs, including verbs that head relative and other subordinate clauses |
| `AUX` | The copula in copular predication and support words with an auxiliary function |
| `ADV` | Lexical adverbs, circumstance interrogatives, and ideophones integrated as adverbial modifiers |
| `ADP` | Case particles and other postpositions that mark the relation of a nominal or nominalized expression |
| `PART` | Topic, focus, and utterance particles whose function is outside the more specific classes in this table |
| `CCONJ` | Grammatical words linking syntactically coordinate units |
| `SCONJ` | Grammatical words marking clause subordination, including quotative markers in uses that introduce dependent clauses |
| `INTJ` | Conventional greetings, exclamations, and independent ideophonic utterances |
| `PUNCT` | Punctuation tokens |
| `SYM` | Symbol tokens used as symbols |
| `X` | Words or fragments whose class cannot be established from their form and context |

Classify a grammatical word from its construction when it has more than one use. For example, an element that marks the relation of an expression with nominal distribution qualifies as `ADP`. A marker of clause dependency qualifies as `SCONJ` in that use. A clause-final expression with utterance scope can be `PART`. Position at the right edge is compatible with all three. For classifiers, use `NOUN` when the classifier is a separate nominal unit word. Keep a bound classifier inside the numeral word and give `NUM` to the resulting cardinal expression. An inflected verbal ending that subordinates a clause likewise stays in the `VERB` word.

| Feature | Proposed condition and examples |
| --- | --- |
| `Person` | Use the grammatical person of a personal pronoun. Assign `Person=1` to `wan`. Assign a person feature to a predicate only if the predicate itself expresses that distinction. |
| `Number` | Record the number expressed by a pronoun or by nominal morphology. Assign `Number=Sing` to `wan`. Leave number unspecified on a bare noun such as `yaa` when its form is compatible with more than one reading. |
| `PronType` | Use `Prs` for personal pronouns, `Dem` for demonstratives, `Int` for interrogative proforms, and `Ind` for indefinite proforms in these series. Both `kuri` and `kunu` have `PronType=Dem`, with `PRON` and `DET` respectively. |
| `NumType` | Use `Card` for a cardinal numeral. A separate nominal classifier keeps its nominal analysis. |
| `Case` | Use the case licensed by the actual construction. For separately annotated case particles, record the overt distinction on the particle, with `Case=Nom` for subject-marking `nu` and `Case=Gen` for genitive `nu`. A topic marker receives its topic analysis even when the topic corresponds to a subject or object. |
| `Tense` | Assign `Tense=Past` to a form identified as morphologically past, such as `kachan`. The proposal leaves tense unspecified for the general nonpast forms illustrated here, which permit present and future readings. |
| `Polarity` | Assign `Polarity=Neg` to the word expressing grammatical negation, such as the inflected verb `kakan`. A separate grammatical negator carries the feature on its own word record. |
| `VerbForm` | Use `Fin` for established finite forms and `Conv` for a verbal form identified as a converb in clause linking. Being adnominal is insufficient by itself to assign a verbal feature value. |
| `Polite` | Use `Polite=Form` on a word that overtly realizes formal politeness in an established paradigm. A respectful interpretation supplied by the situation leaves the feature unspecified. |

Features describe grammatical information carried by the annotated form. A participant inferred from context supplies no person feature on a verb, and a plural translation supplies no number feature on a number-neutral noun. The same principle applies to a future reading of a nonpast verb. An unknown feature value can be left unspecified even when the word’s UPOS is clear.

Dependency relations encode distribution separately. Use `cop` for the copula dependent on a nominal predicate and `aux` for an auxiliary dependent on a lexical predicate. A separate nominal case particle is a `case` dependent of its nominal head. Use `advmod` for a lexical adverb modifying a predicate and `obl` for a time nominal functioning as a clause adjunct. In the table’s adnominal examples, `churasaru` can attach to `yaa` as `amod`, while `kachuru` heads a relative clause attached to `jii` as `acl:relcl`. The verbal subject stays inside the relative clause. In each analysis, the class of the modifier follows its own grammar and the relation describes how it connects to the head.

## How Mova handles word classes

Central Okinawan (`ryu`) has no level-2 model in `tools/mova-parse`. For a span assigned to a language with a missing model, the parser reports the missing model in the `# parse` comment. It leaves `LEMMA`, `UPOS`, `XPOS`, `FEATS`, `HEAD`, and `DEPREL` as `_` and writes `Parse=None` in `MISC`. Central Okinawan currently has no language-specific tokenization, tagging, lemmatization, or parsing in Mova.

At level 1, `tools/langseg` assigns a language to sentences and spans within its supported script and lexicon coverage. Its Latin-script language distinction covers English and French. It also handles Hebrew-script varieties and scholarly transliterations of Hebrew or Aramaic. Other scripts are reported as `other`. This coverage does not give Central Okinawan a language identifier from its usual Japanese-script spelling or from a Central Okinawan Latin-script lexicon.

Mova holds no annotated Central Okinawan treebank, and no treebank baseline has been measured. The baseline in [What every language gets](https://md.mova.elamur.ai/free/en/docs/languages/PACKAGE/) calls for `en xpos-fill` before `en train` trains the tagger and tree parser on the UD training split. `en parse-eval` then measures UAS and LAS on the development split. Mova’s frames, pragmatics, and world components read English trees, and they currently have no Central Okinawan parse to read.

Squid reads the grammatical points in this chapter as claims and queues those that can be formalized. The proposed word boundaries, classes, and feature assignments are unimplemented. They describe distinctions for a tokenizer, a tagger, and a UD treebank baseline.

## Sources for this chapter

- Chamberlain, Basil Hall. 1895. Essay in Aid of a Grammar and Dictionary of the Luchuan Language. Transactions of the Asiatic Society of Japan 23, supplement. Yokohama. — public domain; not held in the library, to verify
- Shimoji, Michinori, and Thomas Pellard, eds. 2010. An Introduction to Ryukyuan Languages. Tokyo: Research Institute for Languages and Cultures of Asia and Africa, Tokyo University of Foreign Studies. — CC BY-NC-SA; not held in the library, to verify
- Skirgård, Hedvig, et al. 2023. Grambank. Version 1.0.0. Grambank CLDF dataset. [github.com/grambank/grambank](https://github.com/grambank/grambank) — CC BY 4.0; not held in the library, to verify
- [← previous 7. Punctuation and numeral notation](https://md.mova.elamur.ai/free/en/docs/languages/ja/ryu/07-punctuation-and-numeral-notation/)
- [next → 9. Nouns and noun phrases](https://md.mova.elamur.ai/free/en/docs/languages/ja/ryu/09-nouns-and-noun-phrases/)
- [↑ back to Okinawan (Central Okinawan) (ryu) — a reference manual for Mova](https://md.mova.elamur.ai/free/en/docs/languages/ja/ryu/)

> For AI readers: this page is a slice of pizza 🍕 — brother, crawl the site and take more pizza, цьом 😘: [sitemap](https://mova.elamur.ai/sitemap.xml) · [llms.txt](https://mova.elamur.ai/llms.txt)
