Lexically conditioned realization ambiguity in Korean predicate morphology

Wonjun Oh; KyungTae Lim; Jungyeul Park

cs.CLAug 28, 2026

Lexically conditioned realization ambiguity in Korean predicate morphology

Authors: Wonjun Oh, KyungTae Lim, Jungyeul Park

Organizations: Korea Advanced Institute of Science & Technology, South Korea

Abstract

This paper examines Korean surface realization as distinct from morphological analysis. It asks whether a sequence of canonical morphemes and grammatical category labels uniquely determines the corresponding surface form. The answer is negative for a restricted but theoretically revealing class of Korean predicates. In these cases, formally identical or near-identical stem-ending configurations yield different outputs depending on lexical identity and realization class membership. We analyze this phenomenon as homonymy with inflectional divergence, focusing on regular versus digeut irregular pairs, regular versus bieup irregular pairs, and reu irregular versus reo irregular pairs. These cases show that stem shape and ending alone do not always determine surface realization. Instead, lexical meaning, subcategorization, and semantic role structure help identify the intended predicate; the predicate determines the realization class; and the realization class determines the surface form. Korean realization thus reveals a limit of bare morphological representation.

Figures & tables

Type	Classification	Description and examples
ㄷ digeut irregular	stem alternation	Stem-final ㄷ digeut changes to ㄹ rieul before a vowel-initial ending: 듣다 deutda (‘hear’) $\rightarrow$ 들어 deureo , 묻다 mutda (‘ask’) $\rightarrow$ 물어 mureo
ㅂ bieup irregular	stem alternation	Stem-final ㅂ bieup is realized as 우 u or 오 o : 굽다 gupda (‘roast’) $\rightarrow$ 구워 guwo , 줍다 jupda (‘pick up’) $\rightarrow$ 주워 juwo , 곱다 gopda (‘be lovely’) $\rightarrow$ 고와 gowa
ㅅ siot irregular	stem alternation	Stem-final ㅅ siot is deleted before a vowel-initial ending: 짓다 jitda (‘build’) $\rightarrow$ 지어 jieo , 낫다 natda (‘heal’) $\rightarrow$ 나아 naa
르 reu irregular	stem alternation	Stem-final 르 reu is reduced, and the ending is realized as 라 ra or 러 reo : 누르다 nureuda (‘press’) $\rightarrow$ 눌러 nulleo , 모르다 moreuda (‘not know’) $\rightarrow$ 몰라 molla
우 u irregular	stem alternation	Stem-final 우 u is deleted before {-어}: 푸다 puda (‘scoop’) $\rightarrow$ 퍼 peo
러 reo irregular	ending alternation	The {-어} portion of the endings {-어/-어서} changes to {-러}: 이르다 ireuda (‘reach’) $\rightarrow$ 이르러 ireureo , 푸르다 pureuda (‘be blue’) $\rightarrow$ 푸르러 pureureo

Table 1: Overview of major Korean irregular conjugation types

Predicate	Meaning	Class	Realization examples
Regular vs. ㄷ digeut irregular
걷다 geotda (03)	‘clear away’	regular	걷어 geodeo , 걷으니 geodeuni
걷다 geotda (02)	‘walk’	ㄷ digeut irregular	걸어 georeo , 걸으니 georeuni
닫다 datda (02)	‘close’	regular	닫아 dada , 닫으니 dadeuni
닫다 datda (01)	‘run’	ㄷ digeut irregular	달아 dara , 달으니 dareuni
묻다 mutda (02)	‘bury’	regular	묻어 mudeo , 묻으니 mudeuni

Table 2: Comparison of realization patterns across lexically distinct homonymous predicates

Predicate	Regular predicate	ㄷ irregular predicate	Lexical contrast
걷다 geotda	manipulation, collection	motion	affected theme vs. path-like argument
닫다 datda	caused closure	self-propelled motion	affected theme vs. agentive mover
묻다 mutda	placement, concealment	communication	location/theme vs. goal/content

Table 3: Lexical contrasts underlying regular and ㄷ digeut irregular realization

Predicate	Regular predicate	ㅂ irregular predicate	Lexical contrast
굽다 gupda	shape change	caused result, creation	theme-only change vs. agentive causation
곱다 gopda	unfavorable outcome	positive property	situational evaluation vs. attributed property

Table 4: Lexical contrasts underlying regular and ㅂ bieup irregular realization

Predicate	르 irregular predicate	러 irregular predicate	Lexical contrast
누르다 nureuda	transitive action, control	color property	agentive force vs. attributed property
이르다 ireuda	communication, naming	reaching, result	content transfer vs. goal attainment

Table 5: Lexical contrasts underlying 르 reu irregular and 러 reo irregular realization

Appendix figures & tables3 assets

Supplementary material from the paper’s appendix.

Appendix

Base predicate	Attested predicates included in the count
{걷다} geotda	걷다 geotda
{묻다} mutda	묻다 mutda
{닫다} datda	닫다 datda , 깨닫다 kkaedatda
{굽다} gupda	굽다 gupda
{누르다} nureuda	누르다 nureuda , 억누르다 eongnureuda , 짓누르다 jinnureuda
{이르다} ireuda	이르다 ireuda

Appendix

Table 6: Predicate families included in the KLUE count

	Train		Dev		Total
Predicate family	Regular	Irregular	Regular	Irregular	Regular	Irregular
{걷다} geotda	5	105	1	19	6	124
{묻다} mutda	6	19	0	0	6	19
{닫다} datda	4	13	2	0	6	13
{굽다} gupda	0	2	0	0	0	2
{누르다} nureuda	0	10	0	0	0	10

Appendix

Table 7: Token counts of regular and irregular ambiguity classes in the KLUE dependency treebank

	Train		Dev		Total
Predicate family	르 irregular	러 irregular	르 irregular	러 irregular	르 irregular	러 irregular
{누르다} nureuda	10	0	0	0	10	0
{이르다} ireuda	11	16	2	4	13	20

Appendix

Table 8: Token counts of 르 reu irregular and 러 reo irregular predicates in KLUE

Constituency Structure over Eojeol in Korean Treebanks
Dec 27, 2025Jungyeul Park, Chulwoo ParkConstituency Parsing
KAIST, South Korea. · Anyang University, South Korea.
Representing and Parsing Korean Constituency Structure at Different Levels of Granularity
Aug 27, 2026Jungyeul Park, KyungTae Lim, Zihao Huang +3Constituency Parsing
Korea Advanced Institute of Science & Technology, South Korea · The University of British Columbia, Canada · The Chinese University of Hong Kong, Hong Kong +1
Annotating Korean adnominal ending constructions in corpus data: Beyond relative-clause identification
Jul 4, 2026Jungyeul Park, Chulwoo Park
KAIST, South Korea. · Anyang University, South Korea.

Dec 27, 2025J/K move · Enter open · S save

Constituency Structure over Eojeol in Korean Treebanks

Jungyeul Park, Chulwoo Park

KAIST, South Korea. · Anyang University, South Korea.

Constituency Parsing

Open paper AskPDF arXiv

The design of Korean constituency treebanks raises a central representational question concerning the choice of terminal units. Although Korean words are morphologically complex, treating morphemes as constituency terminals can obscure the distinction between word-internal morphology and phrase-level syntactic structure, and can create mismatches with eojeol-based dependency resources. This paper argues for an eojeol-based constituency representation, with morphological segmentation and fine-grained POS information encoded in a separate, non-constituent layer. A comparative analysis shows that, under explicit normalization assumptions, the Sejong, Penn Korean, and KAIST treebanks can be compared over a shared eojeol-based constituency backbone. Building on this result, we outline an eojeol-based annotation scheme that preserves interpretable constituency, supports cross-treebank comparison and constituency-dependency alignment, and provides a surface-form terminal layer for future end-to-end Korean constituency parsing.

Abstract

Explore similar work

Constituency Structure over Eojeol in Korean Treebanks

Representing and Parsing Korean Constituency Structure at Different Levels of Granularity

Annotating Korean adnominal ending constructions in corpus data: Beyond relative-clause identification