Mova0.6 pre-release

Old Javanese: the red and white stripes of Majapahit 8. Word classes and annotation diagnostics

Old Javanese word classes are identified from the forms a word takes and the positions it occupies. A root, a word formed from it and a particular occurrence of that word provide different kinds of evidence. The diagnostics below connect that evidence to proposed Universal Dependencies (UD) tags and features. The examples are already romanized. Small capitals in the glosses give grammatical information.

Roots, derived words and grammatical words

A root is the base shared by related forms. A derived word is classified by its own morphology and distribution. An occurrence of the root without an overt affix must likewise be classified in context. A dictionary headword can identify a family of forms even when the headword itself has a restricted distribution. Van der Tuuk uses a passive form as the entry form for certain verb families and distinguishes it from the corresponding active forms. His discussion also shows how a spelling of an active form can suggest the wrong root (Van der Tuuk 1897, § 6).

The word family below illustrates the difference. The glosses give the sense relevant to each citation. ACT and PASS indicate active and passive forms, and 3 identifies a third-person form with number left open. The alternatives in the family establish contrasts in use, without supplying a complete inflectional paradigm.

FormGlossTranslationEvidence
“uni”sound‘sound’Van der Tuuk 1897, s.v. uni, § I
“muni”sound.ACT‘to sound, utter a sound’Van der Tuuk 1897, s.v. uni, § I
“mangunyakën”utter.ACT‘to utter, recite’Van der Tuuk 1897, s.v. uni, § I
“inunyakënira”utter.PASS=3‘uttered by that person’Van der Tuuk 1897, s.v. uni, § I
“uni”earlier‘earlier, before’Van der Tuuk 1897, s.v. uni, § II

The nominal ‘sound’ and the temporal ‘earlier’ have the same written form. A phrase in which uni names a sound supports a nominal analysis. A clause in which it locates an event in the past supports an adverbial analysis. The voice contrast between mangunyakën and the verbal host of inunyakënira is evidence for verbal morphology. The final -ira in the latter example identifies a participant in the uttering. Its interpretation follows from its construction with a passive verb.

Grammatical words have distributions of their own. A nominal introducer combines with a nominal expression, an adposition relates its complement to another part of the clause, and a clause particle has scope over a predicate or clause. Short forms can also be roots or affixes. Classify a short form by what it combines with and by the scope it has in that occurrence.

Distributional diagnostics for nominals, verbs and property words

Nominals

A common noun can head an expression that refers to a person, thing, place or other entity. Evidence includes occurrence with a demonstrative or nominal introducer, with a possessor or descriptive modifier, or as the head of an adposition's complement. Proper nouns identify individual people, places or other named entities. A title before a name can retain its common-noun use. A personal pronoun can occupy an argument position on its own and identifies its referent through person and discourse context.

FormGlossTranslationEvidence
ikang umahDET house‘the house’(constructed)
riṅ umahLOC house‘at the house’(constructed)
wwaṅ sugihperson rich‘a rich person’(constructed)

Argument position alone is insufficient for identifying a noun. A nominal introducer can also make a property or a larger expression refer to a person. The expression “sang sugih” has the gloss HON rich and means ‘the rich person’. The introducer supplies the nominal environment and sugih describes the referent. The use supports treating sugih as a property word within the expression (Van der Tuuk 1897, s.v. hîna, p. 21).

Verbs and property words

Verb diagnostics include participation in a voice alternation, verbal affixation and a regular pattern of arguments licensed by the predicate. A verb can be intransitive, and its arguments can be left implicit where the context supplies them. The existence predicate hana illustrates a verbal use without an object. Its dictionary entry records the base and several suffixed or prefixed forms. Each derived form still requires its own analysis (Van der Tuuk 1897, s.v. hana).

FormGlossTranslationEvidence
hana wwaṅ riṅ umahexist person LOC house‘There is someone at the house.’(constructed)
sugih ikang wwaṅrich DET person‘The person is rich.’(constructed)

A property word can describe a nominal head or serve as a predicate. In the two examples with sugih, the nominal and clause structures distinguish attributive from predicative use. Both uses support an adjective analysis. Predication by itself is weak evidence for a verb. For a derived form denoting a state or a change of state, look for its actual verbal morphology, related voice forms and argument pattern before assigning VERB. For a property expression used with a verb, check whether it describes a participant or specifies how the event unfolds. A manner interpretation does not by itself change the lexical category of a property word. The temporal use of uni has independent evidence for an adverbial reading (Van der Tuuk 1897, s.v. uni, § II).

When tests conflict, give more weight to morphology whose function is established for that word family and to clear syntactic relations than to English translation or word order alone. In a short passage, a form may have too little context for a secure distinction between a property predicate, a state verb and a nominal predicate.

Function words and forms with several grammatical uses

Function words are classified by their relation to the surrounding phrase or clause. The relevant questions are whether a form introduces a complement, joins parallel units, identifies a referent, or contributes a meaning such as negation within a single clause. Short forms with similar spellings can have different functions.

Form or constructionGlossTranslationDiagnosticEvidence
riṅ umahLOC house‘at the house’The locative form takes a nominal complement.(constructed)
ikang umahDET house‘the house’The introducer identifies a referent together with a nominal head.(constructed)
aku lawan kita1SG and 2‘you and I’In the coordination reading, the two pronouns have parallel status.(constructed)
“jan tan datènga”if NEG come.IRR‘if [he] had not come’jan, here read as yan, introduces the condition. tan negates its predicate.Van der Tuuk 1897, s.v. a, § II
“tambis ta nghulun matya”almost PTCL 1SG die.IRR‘I would almost have died’ta occurs within the clause, following tambis.Van der Tuuk 1897, s.v. a, § II
“aho”EXCL‘oh!’In this sense the form is an exclamation.Van der Tuuk 1897, s.v. aho, exclamatory sense, p. 10

In the coordination use of lawan, each conjunct can fill the same grammatical slot. Where lawan instead introduces an accompanying participant dependent on a predicate, an adpositional analysis fits that use. A subordinator such as conditional yan marks a dependent clause, including one with an unexpressed subject. A clause particle such as ta can follow material within its own clause. That position alone gives no basis for treating it as a conjunction. An exclamation can stand apart from the construction of the predicate. The dictionary also records senses of aho other than the exclamation, so the spelling alone does not identify an interjection (Van der Tuuk 1897, s.v. aho, p. 10).

A demonstrative that fills an argument slot on its own supports PRON. A demonstrative used to identify the referent of an overt nominal head supports DET. A nominal introducer such as sang can occur before a descriptive expression, as in sang sugih. In that use the whole expression occupies the nominal slot, and the introducer retains its role. The presence of a human referent or an honorific alone does not establish that a word is a name (Van der Tuuk 1897, s.v. hîna, p. 21).

Affixes, clitics and the limits of manuscript word boundaries

An affix belongs to the morphological word. Analyze a marker as an affix when it forms a stem or a grammatical word form as part of an identifiable pattern. Verbal voice morphology stays within the verbal token. The same principle applies to a grammatical ending on a noun or verb. A clitic depends on a neighboring form for its realization but can contribute its own syntactic relation. A person form may, for example, express a possessor with a noun or an actor with a passive verb. The host and the clause together determine that relation.

FormMorphological readingGlossTranslationEvidence
“matya”mati with final -adie.IRR‘would die’Van der Tuuk 1897, s.v. a, § II
“tambaka”tambak with final -adam-AS‘for use as a dam’Van der Tuuk 1897, s.v. a, § II
“sihungnya”sihung=nyafang=3‘its fang’Van der Tuuk 1897, s.v. a, § II
“inunyakënira”inunyakën=irautter.PASS=3‘uttered by that person’Van der Tuuk 1897, s.v. uni, § I

The final -a in matya has a modal reading in its clause. After a noun it can express intended use or capacity, as in tambaka. This nominal construction supplies no basis for assigning verbal mood to every form ending in -a. Initial a- has a separate distribution. The dictionary distinguishes a verbal-prefix use alongside ma- from a negative prefix in words of Indic origin. Identify the base and construction before assigning any grammatical significance to a written initial a (Van der Tuuk 1897, s.v. a, §§ II–III).

Use spelling as evidence for the text of a witness and test word boundaries against recurring forms, alternations and syntactic positions. A sequence written together may contain a complement or participant marker. An apparent boundary in a romanized edition can also reflect editorial practice. Van der Tuuk uses a single spelling for Old Javanese dictionary entries, discusses divergent manuscript readings, and notes that some spellings in Balinese witnesses obscure distinctions needed to identify the base. His examples of lost adpositional vowels and shortened forms of -um- also draw on comparative material. They establish the need to verify a proposed split in the relevant variety and passage (Van der Tuuk 1897, §§ 5–8, discussion preceding § 15).

When an attached element has a recoverable pronominal form and its own syntactic relation, the proposed analysis represents it as a separate syntactic word. UD can retain the combined surface form when it records that split. Where the reading, segmentation or function remains uncertain, preserve the available form and flag the uncertainty in the analysis. An editor's spacing, a convenient gloss and a resemblance to an affix each provide insufficient evidence on their own.

Proposed Universal Dependencies tags and features

Assign UPOS to the syntactic word in context. Use the table as a decision guide. Its rows cover the UD tag inventory, and a conditional row supplies a criterion for using a tag without asserting the existence of a distinct Old Javanese class for it.

UPOSProposed diagnostic
NOUNA common nominal head, including a derived word that names an entity or activity and has nominal distribution.
PROPNA form used as an individual name. Examine the use of a title separately from the name it accompanies.
PRONA personal or other proform filling a nominal slot, including an independently analyzed person clitic with a syntactic relation.
DETA demonstrative or nominal introducer that determines reference within a nominal expression.
NUMA word or numeral sign that expresses a cardinal value in counting or quantity. A lexical ordinal with adjectival distribution follows the adjective analysis.
ADJA property word with attributive or property-predicate distribution, including one used in a nominal expression such as sang sugih.
VERBA lexical verb identified by its morphology or construction, including an existential predicate such as hana.
AUXA separate word grammatically dependent on a lexical predicate and expressing a category such as mood, aspect or voice. This analysis requires evidence from the construction.
ADVA lexical modifier of a predicate or clause, for example temporal uni ‘earlier’.
ADPA word that marks the relation of a nominal complement to another expression, for example locative riṅ.
CCONJA coordinator of parallel words, phrases or clauses.
SCONJA marker introducing a subordinate clause, for example conditional yan.
PARTA grammatical particle with a function such as negation or clause marking, including tan and the illustrated use of ta.
INTJAn exclamation or other independent reaction form in its interjectional use.
PUNCTA sign functioning as punctuation in the annotated text.
SYMA symbol token with a function distinct from punctuation or a written numeral.
XA recoverable token whose class cannot be assigned after examination of the form and context.

Features describe grammatical distinctions supported by the token in its construction. They complement UPOS. For example, two verb forms can share VERB and differ in Voice. A pronoun can carry person even when the same information is absent from the verb. The examples of voice and person in inunyakënira illustrate why assigning a feature to the correct token depends on segmentation (Van der Tuuk 1897, s.v. uni, § I).

FeatureProposed use
PronType=Prs, PronType=DemUse Prs on personal pronouns and Dem on demonstrative pronouns or determiners. Require evidence of deixis or demonstrative reference before using Dem on a nominal introducer.
Person=1, Person=2, Person=3Use the value encoded by the personal form. For a separately analyzed pronominal clitic, place person on the clitic.
Number=Sing, Number=PlurUse when the form marks the number contrast. Leave number unspecified for a form compatible with both values.
Poss=YesUse on a pronominal or determiner form when its analysis identifies it as possessive. An attached person form with an actor reading on a passive verb has a different relation.
NumType=Card, NumType=OrdDistinguish quantities from rank in an ordered sequence, including when an ordinal has ADJ as its UPOS.
Voice=Act, Voice=PassUse for established active and passive forms of a verb. The absence of an overt affix does not by itself establish a voice value.
Mood=Irr, Mood=ImpUse Irr for a verbal form established as irrealis and Imp for a form established as imperative. A final -a, a conditional context or a directive translation alone is insufficient.
Polarity=NegUse on a negative marker such as tan in its negating use. Record the scope through the syntactic analysis of the clause.

Leave a feature unspecified when it can only be inferred from the translation or the situation. Adpositions can mark the role of nominal expressions without furnishing morphological Case on the nouns. A noun's real-world referent can be plural without a marked Number value. A time adverb can locate an event without furnishing morphological Tense on the verb. Likewise, a property word used in a comparison requires evidence of a grammatical degree distinction before it receives a Degree value.

How Mova handles word-class claims

Mova has no level-2 parser model and no annotated treebank for Old Javanese (kaw). When tools/mova-parse receives a span identified as kaw, the # parse comment reports the missing model. The tokens have _ in LEMMA, UPOS, XPOS, FEATS, HEAD and DEPREL and Parse=None in MISC. Mova does not tokenize, tag, lemmatize or parse Old Javanese. Language identification at level 1 depends on the script or lexicon support available in tools/langseg.

Squid reads the manual as grammatical claims. Formalizable rules, closed lists and statistics wait in its queue for implementation. The word-boundary diagnostics, UPOS assignments and feature proposals in this chapter are unimplemented. The first points needed for a baseline are those used by a tokenizer, a tagger and a UD treebank. Mova measures no treebank baseline for kaw, and its frame, pragmatics and world components have no Old Javanese trees to read.

Sources for this chapter

This page in Markdown → https://md.mova.elamur.ai/free/en/docs/languages/jv/kaw/08-word-classes/