Obed Junias

Researcher at HUBBS, University of Colorado Boulder, M.S. Computer Science

I am a researcher at the HUBBS Lab at the University of Colorado Boulder, working with Dr. Theodora Chaspari. I am also advised by Dr. Maria L. Pacheco in the BLAST Lab. My research is driven by a fundamental question: why do language models behave the way they do? I am interested in how models acquire capabilities through training, how they represent and use information internally, and why these capabilities sometimes deteriorate. I currently investigate these questions through behavioral evaluation and interpretability.

Before grad school, I worked at Oracle as a Senior Member of Technical Staff. I completed my M.S. in Computer Science at CU Boulder in 2026.

I’m applying to PhD programs for Fall 2027.

My Work includes

LOGICAL-COMMONSENSEQA (ACL 2026), a benchmark investigating how language models compose plausibility judgments with logical constraint.
A follow-up on structured compositional reasoning, where I look at how to help models handle the logical combinations they struggled with in the benchmark.
Model collapse, where I study how recursive training on synthetic data changes a model’s behavior and its internal representations.
Assessing demographic bias in language-based depression detection (IEEE-BHI 2025).

A lot of this sits close in the intersection of NLP and cognitive science, and I enjoy working across fields. Feel free to reach out if you’d like to discuss research.

news

Sep 16, 2026	New preprint! From Atomic Evidence to Logical Composition: Structured Compositional Reasoning over Compound Answer Options is now on arXiv!
Jun 24, 2026	LOGICAL-COMMONSENSEQA is now officially published in the ACL Anthology. See you in San Diego! Read the full paper here!
Apr 07, 2026	Thrilled to announce that our work, “LOGICAL-COMMONSENSEQA: A Benchmark for Logical Commonsense Reasoning,” has been accepted at ACL 2026 (Short Papers). Paper

latest posts

Apr 28, 2026	Bayesian Updating in Machine Learning: Making Sense of Training Data
Apr 26, 2026	Conditional Probability, Priors, Likelihood, and Bayes' Rule: The Foundations

selected publications

Assessing Algorithmic Bias in Language-Based Depression Detection: A Comparison of DNN and LLM Approaches

Obed Junias, Prajakta Kini, and Theodora Chaspari

In 2025 IEEE EMBS International Conference on Biomedical and Health Informatics (BHI), 2025

DOI arXiv Bib HTML PDF

@inproceedings{junias2025assessing,
  title = {Assessing Algorithmic Bias in Language-Based Depression Detection: A Comparison of DNN and LLM Approaches},
  author = {Junias, Obed and Kini, Prajakta and Chaspari, Theodora},
  booktitle = {2025 IEEE EMBS International Conference on Biomedical and Health Informatics (BHI)},
  pages = {1--7},
  year = {2025},
  organization = {IEEE},
  doi = {10.1109/BHI.2025.11269509},
  url = {https://ieeexplore.ieee.org/abstract/document/11269509},
}

From Atomic Evidence to Logical Composition: Structured Compositional Reasoning over Compound Answer Options

Obed Junias and Maria Leonor Pacheco

2026

arXiv Bib HTML PDF

@misc{junias2026atomicevidencelogicalcomposition,
  title = {From Atomic Evidence to Logical Composition: Structured Compositional Reasoning over Compound Answer Options},
  author = {Junias, Obed and Pacheco, Maria Leonor},
  year = {2026},
  eprint = {2608.12836},
  archiveprefix = {arXiv},
  primaryclass = {cs.CL},
  url = {https://arxiv.org/abs/2608.12836},
}

LOGICAL-COMMONSENSEQA: A Benchmark for Logical Commonsense Reasoning

Obed Junias and Maria Leonor Pacheco

In Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 2: Short Papers), Jul 2026

Abs arXiv Bib HTML PDF

Commonsense reasoning often involves evaluating multiple plausible interpretations rather than selecting a single atomic answer, yet most benchmarks rely on single-label evaluation, obscuring whether statements are jointly plausible, mutually exclusive, or jointly implausible. We introduce LOGICAL-COMMONSENSEQA, a benchmark that reframes commonsense reasoning as logical composition over pairs of atomic statements using plausibility-level operators (AND, OR and NEITHER/NOR). Evaluating instruction-tuned, reasoning-specialized, and fine-tuned models under zero-shot, few-shot, and chain-of-thought prompting, we find that while models perform reasonably on conjunctive and moderately on disjunctive reasoning, performance degrades sharply on negation-based questions. LOGICAL-COMMONSENSEQA exposes fundamental reasoning limitations and provides a controlled framework for advancing compositional commonsense reasoning.
@inproceedings{junias-pacheco-2026-logical, title = {{LOGICAL}-{COMMONSENSEQA}: A Benchmark for Logical Commonsense Reasoning}, author = {Junias, Obed and Pacheco, Maria Leonor}, booktitle = {Proceedings of the 64th Annual Meeting of the {A}ssociation for {C}omputational {L}inguistics (Volume 2: Short Papers)}, month = jul, year = {2026}, address = {San Diego, California, United States}, publisher = {Association for Computational Linguistics}, pages = {746--758}, url = {https://aclanthology.org/2026.acl-short.61/}, }