[Paper Review] Morphological annotation of Korean with Directly Maintainable Resources

Ivan Berlocher, Hyun-Gue Huh+2 more|ArXiv.org|Nov 21, 2007

Natural Language Processing Techniques9 references3 citations

TL;DR

This paper presents a resource-based morphological annotation system for Korean that bypasses traditional morphological rules by using directly maintainable lexicons of stems, suffix transducers, and allomorph generators. It achieves 89% recall without training, processes text at 1,210 words per second, and enables efficient, updatable linguistic processing for syntactic parsing in agglutinative language NLP.

ABSTRACT

This article describes an exclusively resource-based method of morphological annotation of written Korean text. Korean is an agglutinative language. Our annotator is designed to process text before the operation of a syntactic parser. In its present state, it annotates one-stem words only. The output is a graph of morphemes annotated with accurate linguistic information. The granularity of the tagset is 3 to 5 times higher than usual tagsets. A comparison with a reference annotated corpus showed that it achieves 89% recall without any corpus training. The language resources used by the system are lexicons of stems, transducers of suffixes and transducers of generation of allomorphs. All can be easily updated, which allows users to control the evolution of the performances of the system. It has been claimed that morphological annotation of Korean text could only be performed by a morphological analysis module accessing a lexicon of morphemes. We show that it can also be performed directly with a lexicon of words and without applying morphological rules at annotation time, which speeds up annotation to 1,210 word/s. The lexicon of words is obtained from the maintainable language resources through a fully automated compilation process.

Motivation & Objective

• To develop a morphological annotation system for Korean that avoids dependency on morphological rule engines.
• To enable high-precision annotation using only directly maintainable linguistic resources such as stem lexicons and suffix transducers.
• To achieve fast annotation speed suitable for integration into syntactic parsing pipelines.
• To ensure system performance can be controlled and evolved through manual updates to language resources.
• To demonstrate that morphological annotation of Korean is feasible without morpheme-level lexicons or rule-based analysis at annotation time.

Proposed method

• The system uses a lexicon of word stems as the primary linguistic resource, compiled automatically from maintainable components.
• It applies suffix transducers to decompose words into morphemes, based on predefined morphological patterns.
• Allomorph generation is handled by dedicated transducers that produce correct inflectional variants.
• The annotation process is rule-free at runtime, relying solely on pre-compiled, maintainable resources.
• The system constructs a morpheme graph annotated with rich linguistic tags, using a tagset 3–5 times more granular than standard sets.
• All components—lexicons, transducers, and generation rules—are designed for easy manual update, ensuring long-term maintainability.

Experimental results

Research questions

• RQ1 Can morphological annotation of Korean be performed effectively without relying on morpheme-level lexicons or runtime morphological rules?
• RQ2 Can a system achieve high recall and speed using only maintainable, manually updatable linguistic resources?
• RQ3 Is it feasible to compile a word-based lexicon from maintainable components for accurate morphological analysis?
• RQ4 How does the performance of a rule-free, resource-based system compare to traditional morphological analyzers in terms of recall and speed?
• RQ5 To what extent can the system’s accuracy and behavior be controlled through direct maintenance of language resources?

Key findings

• The system achieves 89% recall on a reference-annotated corpus without any corpus-based training.
• Annotation speed reaches 1,210 words per second, significantly faster than rule-based approaches.
• The system’s performance is directly controllable through manual updates to lexicons and transducers.
• The tagset granularity is 3 to 5 times higher than standard tagsets, enabling fine-grained linguistic annotation.
• The method successfully bypasses the need for morpheme-level lexicons, proving that word-level resources can support accurate morphological annotation.
• The fully automated compilation process ensures that the word lexicon remains consistent with the underlying maintainable resources.