Mova0.6 pre-release

the flag of South Korea 21. Phrase structure

Middle Korean generally places nominal modifiers before the noun and the inflected predicate after its dependents. Case and other particles typically follow the expression they mark. The examples below use the royal preface to the 1459 Hunminjeongeum eonhae alongside constructed illustrations. The cited phrases omit tone marks and the Hangul reading glosses beside Chinese characters. The romanization represents ㆍ as o, ㅡ as u, ㅗ as wo, ㅜ as wu, and ㅸ as W. Glosses use ADN for adnominal, GEN for genitive, NMLZ for nominalizer, and NOM and ACC for nominative and accusative. (text) 1459, § Royal preface)

Order of modifiers inside nominal phrases

The head noun supplies the kind of entity the phrase denotes. A demonstrative picks out an instance or set, a possessor supplies a relation, an adnominal predicate supplies a property or situation, and a numeral supplies a quantity. Each of these dependents can precede the noun. A case particle or an additive or distributive particle follows the nominal expression. The last written syllable of that expression may already contain the particle.

ConstructionInternal orderCondition
Demonstrative phraseDemonstrative + nounThe demonstrative points to or retrieves the referent of the whole phrase.
Possessive or relational phraseGenitive dependent + nounThe dependent identifies the possessor or another entity through which the head is understood.
Descriptively modified phraseDescriptive predicate in adnominal form + nounThe predicate is inflected for use in front of a nominal head.
Relative or content clause with a nounAdnominal clause + nounThe whole clause modifies the head. Its own arguments and adjuncts belong inside the clause.
Prenumeral constructionNumeral in its adnominal form where applicable + nounThe numeral directly quantifies the noun. Counting-unit constructions also occur.
Particle-bearing nominal phraseNominal expression + particleThe particle follows its host. The host may itself contain a modifier, possessor, or quantity expression.

When modifiers combine, their order must allow the intended grouping. A demonstrative can stand before a descriptive modifier, as in the constructed example below. The demonstrative then selects a house from the context and the adnominal predicate describes the house. This pattern illustrates the scope of those two modifiers. The preface does not provide a full ordering paradigm for all combinations of demonstratives, genitives, clauses, and quantities.

FormRomanizationGlossTranslation
그 큰 집 (constructed)ku khun cipthat be.big-ADN house‘that big house’
“어린 百姓이” (text) 1459, § Royal preface)elin poyk syeng-ibe.unlearned-ADN common.people-NOM‘the unlearned common people’

In the second phrase, the adjective is an inflected descriptive predicate. The final -i marks the noun phrase headed by 百姓 as the subject. The sense ‘unlearned’ comes from the passage about people unable to express what they wish to say in writing. The same combination of adnominal inflection and phrase-final case is available when the modifier denotes a different property. (text) 1459, § Royal preface)

Two adjacent nouns require an analysis of their relation. A genitive ending overtly marks a dependent. A conventional two-noun expression may instead have a compound analysis. A counting unit next to another noun may measure it, and two noun phrases can stand in apposition when both identify the same referent. Precedence alone establishes none of these relations. In particular, the absence of an overt genitive ending licenses no automatic recovery of a missing genitive.

Genitives, demonstratives and adnominal clauses

A genitive dependent precedes the expression it determines. Its relation to the head may be ownership, association, origin, or another relation recoverable from the two nouns and the passage. The genitive on the dependent and the case of the full phrase serve different purposes. The genitive marks the internal relation. The case at the right edge marks the full phrase in a larger construction.

Genitive formDistribution
-ᄋᆡ, -의Used with ordinary animate dependents in the regular contrast. The choice of vowel follows the harmonic class of the stem.
-ㅅUsed with inanimate and honorifically treated dependents in the regular contrast. It is also found in established noun combinations, where it need not mark a freely constructed possessive.
Genitive forms of pronouns, including 제The genitive can be expressed in the form of the pronoun itself. Analyze the attested form before looking for a separate suffix.
FormRomanizationGlossTranslation
“나랏말ᄊᆞ미” (text) 1459, § Royal preface)nala-s malssom-icountry-GEN speech-NOM‘the country’s speech’
사ᄅᆞᄆᆡ 말 (constructed)salom-oy malperson-GEN speech‘a person’s speech’
“제 ᄠᅳ들” (text) 1459, § Royal preface)cey ttut-ulself.GEN intention-ACC‘their own intention’

In the first example, -s follows the dependent meaning ‘country’. Nominative -i follows the head meaning ‘speech’. The romanization shows the morphemes even where the written syllables run across their boundaries. In the third example, 제 refers back to the person whose intention is at issue. Accusative -ul marks that intention as the object of expressing it. The English plural in the glossed phrase reflects its generic context.

The basic demonstrative series distinguishes the speaker’s vicinity, the addressee’s vicinity or an already accessible referent, and a referent distant from both. Context can make accessibility a matter of discourse rather than physical position. In adnominal use, each form precedes the noun. Independent demonstrative expressions require their own analysis, since the form and function of a pronoun need not coincide with those of a modifier.

Adnominal formRomanizationUsual valueConstructed phraseRomanizationGloss and translation
이iNear the speaker이 사ᄅᆞᆷi salomthis person, ‘this person’
그kuNear the addressee or accessible in the discourse그 사ᄅᆞᆷku salomthat person, ‘that person’
뎌tyeDistant from both participants뎌 사ᄅᆞᆷtye salomyonder person, ‘that person over there’

A demonstrative-like expression may itself have adnominal morphology. In the preface, 이런 describes a reason as being of the kind just explained. The particle belongs after 젼ᄎᆞ, the head of the phrase. (text) 1459, § Royal preface)

FormRomanizationGlossTranslation
“이런 젼ᄎᆞ로” (text) 1459, § Royal preface)ilen cyencho-lobe.of.this.kind-ADN reason-RO‘for this reason’

An adnominal clause ends in an adnominal predicate and precedes the noun it modifies. A one-word descriptive modifier and a clause with its own dependents can therefore occupy the same position. With a descriptive predicate, an adnominal form in -ㄴ can describe a state holding at the relevant time. With an action predicate, an adnominal form in -ㄴ can present an event as completed relative to the noun or the larger situation. A form in -ㄹ can present an anticipated, intended, or otherwise unrealized event. The predicate class and the surrounding sentence determine the reading.

FormRomanizationGlossTranslation
“니르고져 호ᇙ 배” (text) 1459, § Royal preface)nilu-ko.cye hwol pa-isay-DESID do.ADN matter-NOM‘what they wish to say’

Here 배 contains the dependent noun 바 and nominative -i. The material before it describes what the people want to say. In the larger passage, the whole phrase is the subject of an existential predicate. A noun with little descriptive content, such as 바, can still head a phrase and carry case.

Where the noun supplies a participant understood inside the modifying clause, the construction has a relative interpretation. For example, what is wished to be said in the cited phrase is also what the noun 바 denotes. Other adnominal clauses provide the content of a noun or specify a situation associated with it, without assigning the noun a participant role inside the clause. The same position before the head is available to both. A finite or adnominal form on the last predicate gives evidence about the boundary of the modifying clause. It does not by itself resolve the relation of that clause to the head.

Quantities and particles at phrase boundaries

A numeral can directly modify a following noun. Some native numerals have special adnominal forms, so the shape used in counting on its own need not be the shape before a noun. A compound numeral can fill the same slot as a simple numeral. In the preface the compound numeral ‘twenty-eight’ precedes the noun ‘character’, and accusative case follows ‘character’. (text) 1459, § Royal preface)

FormRomanizationGlossTranslation
두 사ᄅᆞᆷ (constructed)twu salomtwo.ADN person‘two people’
“스믈여듧 字ᄅᆞᆯ” (text) 1459, § Royal preface)sumul yetulp ccong-loltwenty eight character-ACC‘twenty-eight characters’
“사ᄅᆞᆷ마다” (text) 1459, § Royal preface)salom-mataperson-DISTR‘each person’

A quantified nominal can contain a counting unit. The numeral specifies the quantity, the counting unit specifies what is counted as one, and the lexical noun identifies what is being measured or counted. When a numeral and counting unit follow a lexical noun, they can form a quantity expression associated with that noun. Some counting units are also ordinary nouns. Treat them as units in the contexts where they count or measure another expression. When the unit heads an independently referential phrase, it can take the same nominal dependents and particles as another noun. The preface’s 字 is the noun for the items being counted in its context.

Quantifying and distributing are distinct operations. A numeral gives a count. -마다 follows a nominal host and distributes what is said over its members or instances. Its English equivalent may appear before a noun even though the Middle Korean form appears after it. In the cited example, -마다 supplies the interpretation ‘each person’ without a numeral.

Right-edge environmentHostInterpretation
Case after a lexical nounThe head at the right edge of a nominal expressionThe marked expression bears a role in the larger clause.
Case after a dependent noun or nominalized predicateThe nominal expression at the end of the constructionThe full construction can act as a subject, object, or oblique according to its marking and use.
Distributive particle after a nominal expressionThe expression whose members or instances are involvedThe predication is distributed over those members or instances.
Topic, additive, or restrictive particle after an eligible hostThe constituent with that discourse or scope functionThe particle’s position identifies its host. The context helps determine the extent of the contrast, addition, or restriction.

Phrase structure helps distinguish a particle on an internal dependent from one on the larger phrase. In a genitive construction, the genitive marks the dependent even though another noun follows. In a quantified construction with case on the final noun, the case includes the quantity expression in its scope. With a particle sequence, identify the host of the inner particle before interpreting the outer one. A sequence can mark an oblique expression and then give that expression a discourse function. There is no requirement that every nominal dependent of a predicate carry overt case. The construction, the predicate’s valency, and the discourse can identify the role of a bare nominal.

A written boundary is insufficient evidence for separating a particle from its host or for deciding whether a quantity expression is independent. Conversely, a single written group can contain more than one syntactic relation. The two markers in 나랏말ᄊᆞ미 identify an internal genitive and the nominative role of the entire phrase, while the sole accusative marker in 스믈여듧 字ᄅᆞᆯ has scope over both the noun and its numeral. (text) 1459, § Royal preface)

Predicate groups and auxiliary order

The lexical predicate follows its objects and other local dependents in the ordinary ordering. Manner and degree expressions also precede the predicates they modify. A following auxiliary adds a value such as inability or intention to the lexical predication. The first predicate has a form licensed by the construction, and the last predicate bears the ending that relates the group to the surrounding syntax. The last predicate can take a clause-final, connective, or adnominal ending as required by that syntax.

ConstructionOrderCondition
Simple predicateDependents + inflected lexical predicateThe lexical predicate carries the ending for the construction.
Long negationLexical predicate in -디 + inflected negative predicateThe negative predicate takes scope over the preceding lexical predication. 아니ᄒᆞ- expresses negation, and 몯ᄒᆞ- can express inability or failure to carry out the event.
Desire or intention with ᄒᆞ-Lexical predicate in -고져 + inflected ᄒᆞ-The first predicate supplies the intended event. The second predicate carries the ending for the larger clause or adnominal construction.
Lexical predicate with a clausal complementComplement clause + selecting predicateThe second predicate has its own lexical sense and selects a situation, utterance, or other content expressed by the first clause.
Linked lexical predicatesPredicate with a connective ending + following predicationThe ending relates two predications through coordination or subordination. Each predicate can have dependents of its own.
FormRomanizationGlossTranslation
“펴디 몯ᄒᆞᇙ” (text) 1459, § Royal preface)phye-ti mwot-ho-lexpress-LINK cannot-do-ADN‘who cannot express it’
“ᄉᆞᄆᆞᆺ디 아니ᄒᆞᆯᄊᆡ” (text) 1459, § Royal preface)somost-ti ani-ho-lssaycommunicate-LINK NEG-do-CAUSAL‘because it does not communicate’

In the first example, the adnominal ending follows 몯ᄒᆞ-. The entire group modifies a noun for the person who cannot express an intention. The object belongs to the lexical predicate ‘express’. In the second example, the causal ending follows 아니ᄒᆞ-. It relates the entire negative predication to what follows. The positive lexical event is supplied in both cases by the predicate before -디. (text) 1459, § Royal preface)

The desire construction shows the same division of work between predicates. In 니르고져 호ᇙ, the event is ‘say’, its desired status is expressed by the combination with -고져, and the form of ᄒᆞ- makes the group adnominal. The independent verb ‘do’ and the second predicate of this construction share a form. The complement form and the meaning of the whole group determine the use in context. (text) 1459, § Royal preface)

Sequence alone does not establish a predicate group of the auxiliary kind. A connective predicate may express a separate event. A verb meaning ‘say’, ‘think’, or ‘order’ can select a clause and still head a lexical predication of its own. A separate subject is strong evidence for a clause boundary when the construction permits it. A shared subject also occurs in clause linking and in complements, so subject sharing alone does not establish auxiliary status. Look together at the first predicate’s ending, the lexical contribution of the second predicate, and which predicate licenses the participants.

Final position is a property of the local predicate construction. A lexical predicate may be followed by an auxiliary, and a subordinate predicate may be followed by the predicate of a larger clause. An adnominal predicate precedes the noun it modifies. These cases are compatible with the local ordering of a predicate after its own dependents. For instance, the noun after the desire construction is the head of the larger nominal phrase. It does not change the order of the lexical verb and the following ᄒᆞ- inside that construction.

Degree and manner phrases

Middle Korean descriptive predicates can head a predication or take adnominal endings before nouns. A degree expression precedes the descriptive predicate whose intensity or extent it specifies. In an adnominal phrase, the degree expression and the descriptive predicate together precede the noun. The adnominal ending goes on the descriptive predicate. The numeral in a phrase such as ‘two people’ quantifies the noun directly. A degree adverb such as ‘more’ in ‘a bigger house’ modifies the property of being big.

FormRomanizationGlossTranslation
더 큰 집 (constructed)te khun cipmore be.big-ADN house‘a bigger house’
“수ᄫᅵ 니겨” (text) 1459, § Royal preface)swuWi nikyeeasily learn.CONNECTIVE‘learning it easily’
“서르 ᄉᆞᄆᆞᆺ디” (text) 1459, § Royal preface)selu somost-timutually communicate-LINK‘communicate with one another’, in the negative construction
“날로 ᄡᅮ메” (text) 1459, § Royal preface)nal-lo pswu-m-eday-RO use-NMLZ-LOC‘in using it daily’

In the first example, 더 modifies 큰. The head noun 집 is outside that degree construction. The degree word can have a comparative reading even if the standard of comparison is supplied by the context. The possibility of a degree modifier depends on the meaning of the predicate or on a context that permits a scale. Adnominal inflection alone does not make a predicate gradable.

The preface uses more than one kind of preverbal expression. 수ᄫᅵ describes the manner of learning. 서르 describes a reciprocal relation associated with communicating and falls inside the lexical predication that is negated. 날로 supplies a temporal frequency to ‘use’. Its function is adverbial although its form contains a noun and a postposed particle. The locative on the nominalized ‘use’ expression marks that larger expression. It does not govern 날. (text) 1459, § Royal preface)

Some manner forms are lexical adverbs or are derived with an adverb-forming suffix, including -이 in words that admit that formation. An inflected predicate can also supply an adverbial dependent. In that construction the adverbial ending is on the predicate, and any dependents of that predicate stay within the adverbial clause. When an ending permits a result or purpose interpretation as well as a manner reading, the larger predicate and context decide which relation is intended. A particle-marked nominal used adverbially retains its nominal internal structure.

When several preverbal expressions occur, attachment follows meaning and construction. An expression of ease qualifies the event that can be performed easily. An expression of frequency can qualify a different event in the same sentence. A degree word next to an adnominal descriptive predicate can be part of a noun phrase even when the noun phrase is an argument of a later verb. Analyse the predicate each expression modifies before drawing conclusions from adjacency.

Postpositional phrases and proposed dependency analyses

A Middle Korean postpositional expression places the marker after its complement. The complement may be a simple noun, a modified noun phrase, or a nominalized predication. A particle such as locative -애 or -에 follows its host, with vowel choice determined by the form and class of the nominal. The -로 series can express such relations as means, direction, or cause, depending on the host and predicate. It can also occur in established adverbial expressions. The meaning of the relation is therefore determined in context. In 이런 젼ᄎᆞ로, for example, the marked noun means ‘reason’ and the phrase supplies a cause. (text) 1459, § Royal preface)

FormRomanizationGlossTranslation
“中國에” (text) 1459, § Royal preface)tywung kwuyk-eChina-E‘from China’, as the comparison term with ‘differ’
“文字와로” (text) 1459, § Royal preface)mwun ccong-wa.lowritten.character-WARO‘with written characters’
“ᄡᅮ메” (text) 1459, § Royal preface)pswu-m-euse-NMLZ-LOC‘in using it’
집 안해 (constructed)cip anh-ayhouse interior-LOC‘inside the house’

In 中國에, the predicate supplies the comparison reading. The English ‘from’ renders that relation in English and supplies no evidence for an ablative ending in the Middle Korean. In 文字와로, the sequence follows a single nominal and marks its relation to communicating. Its initial 와 does not by itself identify a coordination. In ᄡᅮ메, nominalization lets the event of using something act as the host of a locative. The spelling runs the nominalizer and case vowel together.

A spatial expression can have an internal nominal head meaning a region such as ‘inside’. In the constructed 집 안해, the house identifies the reference object and ‘interior’ denotes its relevant region. The locative follows the relational noun. When the relational element can carry nominal inflection and take a dependent of its own, those are grounds for analysing it as a noun. If a form functions as a grammatical marker selecting a complement, an adpositional analysis is available. Its position after another noun gives only part of the evidence. The same distinction applies to complex expressions built from source nouns whose distribution may differ across texts or constructions.

A Universal Dependencies analysis has to decide both what the syntactic units are and which one heads the construction. The following proposals use content-word heads. They apply after the relevant forms have been segmented and their functions identified. A bound case suffix can be represented by morphological features on its host under a tokenization policy that keeps the two in one word. The case relation applies when the marker is represented as a separate syntactic word. The head of the nominal or nominalized expression then bears the relation to the surrounding predicate.

ConstructionProposed analysisEvidence to use
Demonstrative before a nounAttach a determiner-like demonstrative to the noun with det.The form identifies the reference of an overt nominal head.
Genitive before a nounAttach the genitive nominal to the possessed or relational noun with nmod. A possessive subtype such as nmod:poss can capture that distinction if adopted.Genitive morphology supports the internal nominal relation. External case belongs to the larger phrase.
Simple descriptive modifierUse amod if the modifier is treated as adjectival in the annotation. Use acl where the modifier is analysed as an inflected predicate heading an adnominal clause.Predicate inflection and the ability to license dependents support a clausal analysis. The tagset and treatment of comparable constructions must be consistent.
Relative or other adnominal clauseAttach the clause’s head predicate to the noun with acl:relcl for a relative and acl for other clausal modifiers.Establish whether the noun supplies a participant understood inside the clause. Adnominal inflection is evidence for modification in either case.
Direct quantityAttach the numeral to the counted noun with nummod.The numeral gives the number of entities denoted by the noun.
Quantity with a classifierIn a construction with an overt numeral and a classifier dependent on it, attach the classifier to the numeral with clf.The classifier provides the unit for the numeral. An independently referential measure noun requires a nominal analysis.
Nominal expression with a separate postpositionAttach the postposition to the nominal head with case. Attach the nominal head to its governor according to the phrase’s role, for example with obl for a predicate-dependent comparison or cause.The postposition marks a nominal complement. The phrase’s attachment and the marker’s local attachment answer different questions.
Relational noun with an external case markerKeep the relational noun as head. Attach its nominal dependent with nmod and any separately represented case marker with case.The relational element still denotes a region or relation and behaves as a noun in the construction.
Predicate with a grammatical auxiliaryKeep the lexical predicate as head. Attach a separately represented auxiliary with aux.The lexical predicate licenses the arguments and the auxiliary contributes a grammatical value to that predication.
Lexical predicate with a complement or a linked clauseUse the appropriate clausal relation, such as xcomp for an open complement, ccomp for a clausal complement with independent subject licensing, or advcl for an adverbial clause.The predicates retain their lexical contribution and the construction determines the relation between them.
Degree or manner adverbAttach the adverb to the predicate it modifies with advmod.Identify the lexical predicate, descriptive predicate, or larger predication in whose interpretation the adverb participates.

The auxiliary proposal can make the lexical predicate the dependency head even when the clause’s final morphology is on the following auxiliary. The two choices represent different properties of the same construction. Likewise, the particle in a postpositional expression can determine its external interpretation while the noun heads it in the tree. For a nominalized clause with a following particle, first determine whether the annotation treats the nominalization as a clause or as a lexical noun. That decision governs the external relation and the treatment of its internal arguments. A following particle alone cannot settle it.

How Mova handles phrase structure

Mova has no level-2 parser model for Middle Korean (okm) and holds no annotated Middle Korean treebank. No treebank baseline is measured. At level 1, tools/mova-parse uses tools/langseg to give a language to sentences and spans. tools/langseg has no Middle Korean identification for Hangul and classifies the script as other. At level 2, a missing language model is reported in the # parse comment. For okm, LEMMA, UPOS, XPOS, FEATS, HEAD, and DEPREL are _, and MISC contains Parse=None. Mova does not tokenize, tag, lemmatize, or parse Middle Korean.

Mova’s en crate provides the tree parser and tagger. The baseline requested by What every language gets uses en xpos-fill before en train on the UD training split, then en parse-eval to measure UAS and LAS on development data. For Middle Korean phrase structure, the points needed first for tokenization, tagging, and that baseline include particle and ending boundaries, the distinction between adnominal and auxiliary constructions, and nominal head and attachment decisions. prag and world read English trees and have no parsed Middle Korean input.

Squid reads the manual as grammatical claims. Formalizable claims enter its queue for a place in Mova’s code. The phrase-structure rules and proposed analyses in this chapter have no implementation for okm yet.

Sources for this chapter

This page in Markdown → https://md.mova.elamur.ai/free/en/docs/languages/ko/okm/21-phrase-structure/