Constituency Structure over Eojeol in Korean Treebanks

TL;DR

This paper advocates for an eojeol-based constituency representation in Korean treebanks, separating morphological details from phrase structure to improve interpretability and cross-resource comparability.

Contribution

It introduces an eojeol-based annotation scheme that maintains constituency clarity and enables effective cross-treebank comparison and conversion.

Findings

Sejong and Penn Korean treebanks are representationally equivalent at the eojeol level.

An annotation scheme supporting interpretability and cross-resource comparison is proposed.

The scheme facilitates conversion between constituency and dependency structures.

Abstract

The design of Korean constituency treebanks raises a fundamental representational question concerning the choice of terminal units. Although Korean words are morphologically complex, treating morphemes as constituency terminals conflates word internal morphology with phrase level syntactic structure and creates mismatches with eojeol based dependency resources. This paper argues for an eojeol based constituency representation, with morphological segmentation and fine grained part of speech information encoded in a separate, non constituent layer. A comparative analysis shows that, under explicit normalization assumptions, the Sejong and Penn Korean treebanks can be treated as representationally equivalent at the eojeol based constituency level. Building on this result, we outline an eojeol based annotation scheme that preserves interpretable constituency and supports cross treebank comparison and constituency dependency conversion.