PaperlayerView exact revision on arXiv

Source-linked AI summary

Lexically conditioned realization ambiguity in Korean predicate morphology

Wonjun Oh, KyungTae Lim, Jungyeul Park

arXiv:2608.27966v1cs.CL

TL;DR

The paper asks whether morphemes and grammatical categories uniquely determine Korean surface forms. It argues that they do not generally, because Korean realization reflects interactions among stem shape, lexical identity, and realization class.

  • Problem

    The paper examines whether a segmented morphological representation determines a unique Korean surface form, testing the sufficiency of morphological representation for surface realization.

  • Method

    The paper analyzes Korean irregular conjugation as homonymy with inflectional divergence, tracing surface realization through lexical-semantic and argument-structure information.

  • Results

    The analysis finds that morphemes and grammatical categories do not generally determine a unique surface output, revealing interactions among stem shape, lexical identity, and realization class.

  • Takeaways & Limitations

    Korean irregular conjugation cannot be adequately understood from stem shape and endings alone; surface realization requires a fuller account of lexical and morphological information.

  • Takeaways & Limitations

    The paper does not claim that morphology reduces to semantics or that semantic roles replace morphophonological description.

Abstract

from arXiv · show

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.

1 Introduction

The paper asks whether a morpheme sequence with grammatical-category labels uniquely determines Korean surface form. It argues that predicate realization can remain ambiguous until lexical identity and realization class are fixed.

  • Core problem: Formally identical or near-identical predicate stems can yield different surface forms because they belong to different lexical items and inflectional classes.The ambiguity concerns a restricted class of homonymous or near-homonymous predicates.
  • Scope and contribution: The paper names this pattern homonymy with inflectional divergence and examines regular versus digeut irregular, regular versus bieup irregular, and reu versus reo irregular pairs.These ambiguity classes connect lexical identity with inflectional behavior.
  • Explanatory proposal: Lexical meaning and argument structure help identify the intended lexeme, which determines the realization class and thereby the surface form.The paper preserves a morphology-first account while treating lexical-semantic information as necessary for disambiguation.
  • Contribution: The contribution is to formulate predicate realization as an ambiguity problem that tests whether bare morphological representation is sufficient for surface recovery.The appendix adds corpus support from the KLUE dependency treebank and a surface reconstruction experiment.
  • Research question: The paper reframes Korean surface generation as morphological realization from an analyzed morpheme sequence, rather than simple segmentation or concatenation.The target output is the overt eojeol produced from canonical morphemes and grammatical categories.

2 Morphological analysis and surface realization in Korean

Korean morphological analysis decomposes eojeol into ordered morphemes, but surface realization may involve predictable alternations or remain underdetermined. In the latter cases, lexical identification selects the realization class that determines the output.

  • Analysis and realization: Korean analysis represents an eojeol as an ordered sequence of morphemes associated with grammatical categories, while realization produces the overt form.The analyzed representation and surface string are therefore distinct levels.
  • Transparency and adjustment: Nominal combinations are often relatively transparent, whereas predicate forms can depart from linear concatenation through contraction and other morphophonological adjustments.For example, {학교+에} yields 학교에, while {가+아} yields 가 and {오+아} yields 와.
  • Types of mapping: Realization ranges from nearly transparent, to locally predictable given lexical and morphophonological conditions, to underdetermined by the analyzed sequence.The paper illustrates these relations with 먹었다, 나섰다, and alternative realizations of {걷+었+다}.
  • Realization ambiguity: The same analyzed stem-ending configuration can support multiple outputs when distinct lexical items with similar stem shapes belong to different realization classes.The example {걷+었+다} can yield 걸었다 or 걷었다 depending on the intended lexeme.
  • Disambiguation: Argument structure and lexical meaning help identify the intended lexeme, but the lexeme—not semantic roles directly—determines the realization class and output.General grammatical rules and the analyzed morpheme sequence alone do not suffice in these cases.

3 Irregular conjugation and the limits of formal predictability

Korean irregular conjugation comprises multiple realization patterns rather than one exceptional mechanism. Its limits become clearest when class membership cannot be recovered from stem shape and ending alone.

  • Lexically conditioned irregularity: Korean irregular conjugation is organized by lexically indexed realization classes, so a possible morphophonological alternation does not itself determine whether it applies.Class assignment depends on the lexical item.
  • Regular versus ㄷdigeut irregular pairs: In digeut pairs, some predicates change final ㄷ to ㄹ before vowel-initial endings, while regular predicates such as 닫다 preserve ㄷ.Thus {듣+어} yields 들어, whereas {닫+아} yields 닫아.
  • Regular versus ㅂbieup irregular pairs: Bieup predicates likewise diverge: irregular items may realize final ㅂ as 우 or 오, while other predicates preserve the consonant.The alternation becomes predictable only after lexical realization-class assignment.
  • 르reu versus 러reo patterns: Reu and reo-related patterns show that formally similar stems can participate in distinct realization classes with different stem-ending relations.The examples include 이르러 and 푸르러 alongside reu-irregular reduction patterns.
  • Broader pattern: Other irregular types, including siot, u, yeo, o, and hieut patterns, support a system of multiple lexically conditioned classes.These include ㅅ deletion, 우 deletion, 하다-related 여 forms, 다오, and adjective forms such as 까매 and 노래.
  • Theoretical implication: Traditional class descriptions presuppose that the predicate has already been assigned, whereas this paper addresses cases where the analyzed input does not determine that assignment.The resulting problem is broader than the regular-versus-irregular opposition.

4 Ambiguity in surface realization and lexical disambiguation

Korean predicates with identical or near-identical stem shapes can diverge in surface realization because lexical identity determines realization class. Across ㄷ digeut, ㅂ bieup, and 르 reu/러 reo contrasts, lexical meaning and argument structure help disambiguate the intended predicate.

  • Core ambiguity: Identical or near-identical stem-ending configurations can yield different surface forms because predicates belong to different realization classes.The ambiguity is therefore a realization problem, not merely a list of homonymous dictionary entries.
  • Regular versus ㄷ digeut irregular pairs: In ㄷ digeut pairs, 걷다, 닫다, and 묻다 show regular and irregular outputs tied to distinct lexical meanings and argument structures.The contrasts include 걷어 versus 걸어, 닫아 versus 달아, and 묻어 versus 물어.
  • Regular versus ㄷ digeut irregular pairs: The ㄷ digeut irregular predicates denote motion or interaction, whereas their regular counterparts involve affected themes, closure, placement, concealment, or related transitive events.Argument-structural patterns diagnose lexical identity rather than independently triggering the surface alternation.
  • Lexical disambiguation: The proposed pathway is argument-structural evidence to lexical identity, lexical identity to realization class, and realization class to surface output.Surface realization is thus not determined by phonological stem shape alone.
  • Regular versus ㅂ bieup irregular pairs: Regular versus ㅂ bieup irregular pairs likewise diverge by lexical identity, including externally caused resultative events versus internally unfolding changes of shape.For 굽다, the contrast is 구워 versus 굽어; lexical differentiation also distinguishes the relevant evaluative predicates in 곱다.

5 Implications for the description of Korean morphology

The paper argues that Korean surface realization cannot be fully described from segmented morphemes and grammatical categories alone. Lexical identity, realization-class membership, lexical meaning, and argument structure help determine otherwise ambiguous outputs.

  • Morphological analysis and realization: Morphological analysis is an abstraction over surface form, not a reversible encoding that uniquely determines realization.The same analyzed stem-ending configuration can remain compatible with more than one surface form until lexical identity is fixed.
  • Lexically specified irregularity: Korean irregularity is a property of lexical items in realization classes, not merely stem shape plus the following ending.Predicates with identical or nearly identical segmental shapes may belong to different realization classes, sometimes correlating with lexical meaning.
  • Lexically specified irregularity: A fuller account must distribute realization patterns across more articulated, lexically specified classes rather than rely on the coarse regular–irregular opposition.The ambiguity pairs identify formal sites of alternation but do not determine which realization applies.
  • Morphology, meaning, and argument structure: Lexical meaning helps identify the predicate, the predicate carries a realization class, and that class determines the surface output.This dependency does not reduce realization to semantics or replace morphophonological description with semantic roles.
  • Morphology, meaning, and argument structure: Argument structure and semantic roles provide evidence for distinguishing homonymous predicates when their meanings and inflectional realizations diverge.The suggested architecture lets morphology, lexical semantics, and argument structure interact when a lexical item is selected for realization.
  • Broader descriptive implications: Surface realization should be analyzed in its own right because it exposes the lexicon, inflectional classes, and morphology–meaning interaction.Decomposing an eojeol does not guarantee descriptive adequacy for realization; asking what information yields a unique output reveals where categories require enrichment.

6 Conclusion

Korean surface realization is underdetermined by bare morphological representation in a restricted but theoretically significant class of predicates. Lexical identity and argument-structural information help resolve realization-class ambiguity beyond stem shape and grammatical categories alone.

  • 6 Conclusion: Korean predicate morphology exhibits a complex interaction among stem shape, lexical identity, realization class membership, and morphophonological conditioning.This interaction shows why Korean irregular conjugation cannot be understood adequately as only a set of local phonological alternations.
  • 6 Conclusion: Identical or near-identical predicate stems can belong to different realization classes and yield divergent surface forms in the same formal environment.The conclusion identifies these cases as homonymy with inflectional divergence rather than ordinary homonymy or irregular conjugation alone.
  • 6 Conclusion: The paper groups the ambiguity into regular–digeut, regular–bieup, and reu–reo irregular contrasts.These recurrent classes are treated as a coherent surface-realization problem rather than scattered observations about irregular conjugation.
  • 6 Conclusion: Lexical meaning, subcategorization, and semantic role structure help identify the predicate before realization-class membership determines the surface output.The explanatory path runs from lexical-semantic and argument-structural evidence to lexical identity, then to realization class and surface form.
  • 6 Conclusion: Morphologically analyzed morpheme and part-of-speech input cannot always be reconstructed by simple concatenation or local spelling normalization.Ambiguous cases require access to realization classes and, in context, lexical identity.
  • 6 Conclusion: The relevant predicate families are attested in KLUE, and surface reconstruction improves substantially when explicit realization constraints are modeled.The reported corpus and reconstruction results support practical consequences for Korean morphological processing without replacing the linguistic analysis.

A Corpus counts of ambiguity predicates in KLUE

The KLUE counts attest the paper’s ambiguity classes, while showing uneven, lexically conditioned distributions. Corpus classification required contextual lexical identification because morphological annotation alone could not recover the relevant realization contrasts.

  • The count targets three ambiguity classes: regular versus digeut irregular, regular versus bieup irregular, and reu irregular versus reo irregular predicates.
  • Each corpus token was classified manually in sentential context according to its contextual lexeme, not surface form alone.The counting unit was the corpus token, with each occurrence counted once regardless of inflected form.
  • Several ambiguity classes are attested in KLUE, but their frequencies are highly uneven across predicate families.The counts provide a corpus-based check that the contrasts are attested rather than merely constructed paradigm examples.
  • The 걷다geotda family is frequent and strongly skewed toward the ㄷdigeut irregular motion predicate, while 묻다mutda and 닫다datda have both regular and irregular tokens.
  • 굽다gupda, 누르다nureuda, and 이르다ireuda appear on only one side of the regular-versus-irregular count, whereas 이르다ireuda has both reu irregular and reo irregular realizations.No 곱다gopda token was found in the searched splits.
  • The corpus counts support attestation, not estimates of grammatical productivity or lexical completeness, because KLUE is not a lexicographic resource for inflectional ambiguity.Low frequency or absence in KLUE does not imply marginality in Korean.
  • Surface reconstruction over KLUE improved from a target-eojeol F1 of 64.96 for the previous rule-based baseline to 92.56 for the refined rule-based system.The refined system also achieved 85.59 sentence-level accuracy and used constraints on boundaries, part-of-speech configurations, lexical stem classes, and realization classes.
  • Remaining errors show that rule-based constraints alone do not fully solve reconstruction and may require finer-grained lexical and contextual information.The computational task operationalizes the claim that morphological analysis is not always a reversible encoding of surface form.
Loading 2608.27966v1…