Sungpil Wang
Sungpil Wang

Sungpil Wang Text AI · NLP · Computational Humanities

PhD Candidate, KAIST GSCTLecturer, Korea University

An NLP researcher who builds language-model pipelines end to end.

I research and share knowledge in Text AI, NLP, and Computational Humanities.

I analyze and generate text across many domains — literature, law, and broadcasting.

연구

Research

Selected and ongoing work

Narrative Structure in Flash Fiction

Analyzed the sentence-level flow of meaning in Korean flash fiction to reveal five narrative types and a recurring pattern of stories breaking and then leaping forward.

Narrative AnalysisNLPLiterature

Prose Poetry vs. Flash Fiction

Showed that prose poetry and flash fiction can be told apart purely from the shape of their trajectory through meaning space, with every genre cue removed.

EmbeddingsNarrative AnalysisPoetry

Flash Fiction Generation Agent

Extracted nine narrative archetypes from human-written flash fiction and trained a Korean language model (Upstage) on them. An agent then diagnoses and rewrites its own output in a closed loop.

Generative AIAI AgentReinforcement Learning

Plot Structure & Story Retrieval

Compared whether a traditional Korean plot structure or a Western one helps retrieve stories better, and found that how much a query is compressed matters more than the plot structure itself.

Information RetrievalNLPNarrative Analysis

Broadcast Speech Transcription

Built a pipeline that separates speech from 15 TV and radio channels and transcribes it to text, designed to stay reliable even when music and noise overlap. (KOCCA)

Speech-to-TextAudioPipeline

Word–Color Association

Developed a way to measure how strongly words associate with colors from large-scale web images, validated on the world’s top 100 brands and color symbolism in 20th-century Anglo-American poetry.

MultimodalComputer VisionComputational Humanities

Legal Network Analysis

Modeled the entire body of Korean law (1948–2024) as a network and classified citations as reinforcing or conflicting, showing that legal coherence declined as special laws proliferated in the mid-2000s.

Network AnalysisNLPLaw

Reliable Topics in Korean Prose Poetry

Used topic modeling on 4,642 Korean prose poems (2000–2024) to isolate six motifs that survive every perturbation (sea, flora, rain, poem-making, music, cat), while about half the corpus resists reduction to any topic.

Topic ModelingComputational HumanitiesPoetry

Dancing Bbongsa: XR Co-op Dance Game

Led the user-perception study for an XR co-op dance game, analyzing 12,000 news articles about dancing to surface how public perception differs and to inform the game’s direction. (HCI Korea 2024)

Text MiningHCIGames

Spiritfarer & Coping with Loss

Studied how the death-themed game Spiritfarer helps players cope with loss, using in-depth interviews to trace how grief oscillates between loss and restoration. (CHI EA 2021)

HCIQualitative ResearchGames
참여연구사업

Projects & Grants

Funded research projects

  1. 2023.07 – 2025.12

    Project Manager (Participating Org.) · Korea Creative Content Agency (KOCCA)

    Broadcast monitoring technology for royalty settlement of literary and artistic works

  2. 2023.06 – 2026.05

    Researcher · National Research Foundation (NRF)

    Network-science analysis of legal information for understanding law and development in the post-AI era

  3. 2022.03 – 2025.02

    Researcher · Korea Institute for Advancement of Technology (KIAT)

    VR/AR specialist training program (RCMS)

  4. 2021.08 – 2023.12

    Project Manager, Lead Org. (2023) · KAIST

    Imagining post-AI culture technology

  5. 2023.07 – 2023.12

    Researcher · National Information Society Agency (NIA)

    Korean text–video–sound multimodal dataset

  6. 2022.02 – 2022.12

    Researcher · National Cyber Security Lab (NCSL)

    AI-based video summarization and captioning

  7. 2021.05 – 2021.12

    Researcher · National Information Society Agency (NIA)

    Video and image editing technology dataset (No. 26)

  8. 2020.09 – 2020.12

    Researcher · National Information Society Agency (NIA)

    Emotion-recognition and video-summarization AI dataset (No. 35): public DB, training pipeline, and applications

강의 · 강연

Teaching & Talks

  • 2025.09 –
    Lecturer

    Applied Digital Humanities (regular course)

    Korea University, College of Liberal Arts

  • 2024.01
    Teaching Assistant

    2nd Digital Humanities Winter School (special course)

    Korea University, College of Liberal Arts

  • 2023
    Invited Talk

    AI and Literature — Scientist’s Garden program

    Daejeon Science Festival

  • 2023.01
    Teaching Assistant

    1st Digital Humanities Winter School (special course)

    KAIST School of Digital Humanities & Social Sciences

  • 2022 – 2023
    Teaching Assistant

    Understanding Korean Literature · Rereading Korean Fiction · Logical Writing

    KAIST School of Digital Humanities & Social Sciences

  • 2022.05
    Invited Talk

    Korean AI & Data Science (special lecture)

    Sungkyunkwan University, College of Liberal Arts

  • 2020 – 2022
    Teaching Assistant

    GSCT website design and development

    KAIST Graduate School of Culture Technology

학력

Education

  1. 2022 – 2027

    Ph.D. in Science (coursework completed Aug 2024 / expected Feb 2027)

    KAIST Graduate School of Culture Technology

    Thesis Constructing Word-Color Associations from Web-Scale Image Data: Framework and Applications

  2. 2020 – 2022

    M.S. in Engineering

    KAIST Graduate School of Culture Technology

    Thesis Characterising the Evolution of Seo Jeong-Ju's Poetic World Through Computational Analysis on Keywords, Topics, and Sentiments

  3. 2015 – 2020

    B.E. · B.S. in Culture & Arts Convergence Software · B.A.

    Dongguk University — Industrial & Systems Engineering / Culture & Arts Software (linked) / Creative Writing (double major)

연락처

Get in touch

Open to collaborations, talks, and advisory work.