PulseExploreJournal ClubDebatesTrendingResearchersJournals
Instagram
HomeExploreJournal ClubTrending
Synapse
⌘+K
Synapse
June 28, 2026KIISE Transactions on Computing Practices0 citations

Enhancing Korean Semantic Role Labeling Performance Using Large Language Models with Self-Reasoning and Self-Refinement

View Full Paper
HHHyunsun HwangYJYoungjun JungCLChangki Lee

Key Points

  • The aim is to enhance the performance of Korean semantic role labeling using large language models' reasoning capabilities.
  • Utilized self-reasoning and self-refinement techniques with large language models for Korean semantic role labeling.
  • Implemented in-context learning by providing examples to generate reasoning processes.
  • Evaluated the performance improvements with the gemma-2-27b-it model, focusing on reasoning and review.
  • Self-reasoning improved performance by 6.12%p with gemma-2-27b-it model.
  • Self-refinement led to an additional improvement of 0.58%p.

Abstract

최근 대규모 언어 모델(LLM)은 사전 학습한 후, 추가 학습 없이도 프롬프트(prompt)에 예제를 제공하는 방식만으로 다양한 작업을 수행할 수 있게 되었다. 특히 Chain-of-Thought(CoT) 기법은 프롬프트에 문제 풀이 과정을 명시하여 수학과 같이 추론 과정이 중요한 작업에서 우수한 성능을 보여주고 있다. 본 논문에서는 자연어처리 작업 중 하나인 한국어 의미역 결정에 LLM의 자가 추론(Self-Reasoning) 및 자가 재검토(Self-Refinement) 능력을 이용하여 성능 향상을 시도하였다. In-context Learning(ICL)의 예제에 대해 LLM이 정답을 기반으로 추론 과정을 자동 생성하고, 이를 LLM이 재검토한 뒤 예제와 함께 입력으로 제공하여 추론을 유도하였다. 실험 결과, gemma-2-27b-it 모델의 경우 자가 추론 적용 시 6.12%p의 성능 향상을 보였으며, 자가 재검토 적용 시 추가적으로 0.58%p의 성능 향상을 보였다.

Ask AI
Helpful
Bookmark
Share
View Full Paper

Cite This Study

Hwang et al. (2026) studied this question.

synapsesocial.com/papers/6a40b92b61bb0a67205c592ehttps://doi.org/10.5626/ktcp.2026.32.6.235
Ask AI
Helpful
Bookmark
Share
View Full Paper

Also Consider

Synapse has enriched 5 closely related papers on similar clinical questions. Consider them for comparative context:

  1. 1Enhancing LLM Reasoning Performance Based on a Mathematical Knowledge Graph: Focusing on Quadratic Equation Problems2026
  2. 2Experimental Study on Cross Task Generalization Phenomenon in Korean small Large Language Models (sLLM)2026
  3. 3Large Language Models have Intrinsic Self-Correction Ability2024 · 2 citations
  4. 4Toward Self-Improvement of LLMs via Imagination, Searching, and Criticizing2024 · 4 citations
  5. 5When Does Self-Correction Work for LLMs in Content Analysis? The Role of External Structure and Task Complexity2026