Corpus linguistics and interpretable machine learning, surfacing behavioral patterns in learner language, mental health discourse, legal text, and human-AI interaction.
People use language differently.
The differences are measurable.
Language as Data Lab · Youngmeen Kim, Ph.D.
NBC 11Alive
Television and online coverage of AI-based detection of depression-related language in online communities.
GSU Newsroom & Phys.org
University and science-press features: AI can spot signs of depression in Reddit posts.
Keynote · Univ. of North Georgia
Keynote at the Linguistics Symposium on automatic language evaluation using machine learning and LLMs.
AAAL 2025 · Denver
Machine-learning and corpus-analytic comparison of language use across online mental health communities.
Research Methods in Applied Linguistics
The power of the prompt: the linguistic effects of prompt engineering on GenAI and learners.
GenAI Learner Platform
Built a GenAI human-computer interaction system with TTS and STT for recently immigrated English learners.
A scroll through the pipeline that turns raw text into interpretable findings.
Where it all begins
Real-world text: posts from online mental health communities, learner essays, legal documents, chatbot exchanges. Language in the wild is unstructured, contextual, and rich with behavioral signal.
Online community post
"I can't focus on anything lately. The worrynever really switches off."
Legal document
"The parties shall perform all obligationsset forth in this agreement."
Learner essay
"My writingimproves every semester but articles are still hard for me."
Structure beneath the surface
Tokenization, POS tagging, dependency parsing, and collocation analysis. Corpus tools turn text into measurable features: lexical sophistication, syntactic complexity, frequency profiles.
Part-of-speech tagging
Collocational strength
Meaning becomes geometry
BERT and sentence transformers encode words and sentences as high-dimensional vectors. Semantically similar language clusters together, making distance and difference measurable.
PCA projection of all-MiniLM-L6-v2 embeddings (precomputed)
Modeling at corpus scale
Mixed-effects regression and structural equation modeling estimate effects and relationships; classical and deep classifiers learn to predict from linguistic features. Both are evaluated at corpus scale.
Classification accuracy
As reported in Kim & Römer-Barron (2025), Corpus-based Studies across Humanities, 3(2).
Opening the black box
SHAP, LIME, permutation importance, and feature ablation reveal which linguistic features drive each prediction.
Feature attribution
Model behavior becomes evidence
Explained predictions become linguistic findings that can be checked against the corpus and stated in the language of the discipline.
Published finding
Depression-related language in an online community can be detected and analyzed with machine learning, topic modeling, and corpus-linguistic approaches.
Kim & Römer-Barron (2025) · doi:10.1515/csh-2025-0005
Features, not black boxes, carry the explanation
Findings are mapped back to linguistic theory. The result is a claim a linguist can interrogate, not just a score.
From model to theory
Quantitative modeling paired with qualitative linguistic analysis: patterns are measured at scale and interpreted through linguistic theory.
Corpus linguistics and machine learning applied where language carries evidence about people: mental health, language learning, law, and human-AI interaction.
Machine learning, topic modeling, and corpus-linguistic analysis of depression-related language in online communities, the core of my dissertation research.
Corpus-based Studies across Humanities (2025) · PTSD/anxiety study in preparation
Tracing collocation use, readability, and cohesion across proficiency levels with interpretable machine learning on learner corpora.
ELLIPSE Corpus, IJLCR (2023) · Applied Linguistics & Applied Corpus Linguistics (under review)
Advising J.D. candidates on NLP tools and corpus-linguistic approaches to the interpretation of U.S. legal documents.
Georgia State University College of Law · research advisory roles
Building GenAI-based interaction systems, chatbots with TTS and STT, for recently immigrated English learners, and studying their linguistic effects.
Research Methods in Applied Linguistics (2026) · RCALL seed grant ($48,500) · Cambridge English Funded Research Programme (FRP), £13,000
Dixon, D. H., Kim, YJ., & Kim, Y.
Research Methods in Applied Linguistics, 5(2), Article 100327.
doi:10.1016/j.rmal.2026.100327Kim, Y. & Römer-Barron, U.
Corpus-based Studies across Humanities, 3(2), 283–312.
doi:10.1515/csh-2025-0005Crossley, S., Tian, Y., Baffour, P., Franklin, A., Kim, Y., Morris, W., Benner, M., Picou, A., & Boser, U.
International Journal of Learner Corpus Research, 9(2), 248–269.
doi:10.1075/ijlcr.22026.croPlus 2 manuscripts under review and 2 in preparation. All publications
| Course | Institution | Period |
|---|---|---|
| Language in Society | Georgia State University | January 2026– May 2026 |
| Introduction to English Linguistics | Georgia State University | August 2025– December 2025 |
| Introduction to Linguistics | Georgia State University | January 2025– May 2025 |
| Introduction to English Linguistics | Georgia State University | June 2024– August 2024 |
| Introduction to Linguistics | Georgia State University | January 2024– May 2024 |
| Linguistic Interpretation and Investigation of U.S. Legal Documents | Georgia State University | January 2023– May 2024 |
I teach linguistics as a data-driven discipline. Whether the course is Introduction to Linguistics or Language in Society, students work with real language data and learn to turn observation into evidence. My aim is for students to leave the course able to ask precise questions about language and to answer them with data.
University teaching from 2021 to 2026: linguistics courses and legal-linguistics support at Georgia State University, and Korean language instruction.
View all teachingReal linguistic feature extraction you can run yourself. Every tool is computationally correct, honestly labeled, and cited.
Lexical diversity (TTR, MATTR, MTLD, HD-D), lexical density, n-grams, and cohesion indices on any text, computed in your browser.
Transformer sentence embeddings, cosine similarity, a 2-D projection, and a token-level similarity heatmap, all in-browser.
POS tagging and dependency parsing with arc visualizations, plus live dependency-based complexity: tree depth and mean dependency distance.