Kevin Krahn
PhD Student, Computer Science and Engineering
The Ohio State University, Columbus
krahn.6 [at] osu.edu · GitHub
I am a Computer Science and Engineering PhD student at The Ohio State University. My primary research interests are in natural language processing and speech processing, with a focus on speech technologies for under-resourced languages. In particular, I am interested in training models to learn languages from only speech, without text, in order to support languages are largely unwritten or lack a standardized orthography.
Before starting my PhD at OSU, I worked on NLP for historical and ancient languages, driven by an interest in computational linguistics and the digital humanities.
Outside of research, I like building things, from games to apps to small tools. See my side projects →
News
- Nov 2026 Our paper "What Do Multilingual Self-Supervised Speech Encoders Know About Sentence Meaning?" was accepted to EACL 2027 (main conference). Preprint on arXiv.
- Jul 2026 Presented our poster for the HydraQE speech translation quality estimation system at IWSLT 2026 in San Diego. Paper · Code
Research Interests
- Unwritten languages
- Building speech technology for languages that lack a standard writing system (translation quality estimation, speech translation).
- Learning from raw audio
- Self-supervised speech representations and what they capture about meaning, without relying on text supervision.
- Earlier work
- NLP for historical and ancient languages; computational linguistics and digital humanities.
Publications
-
2026
What Do Multilingual Self-Supervised Speech Encoders Know About Sentence Meaning?
EACL 2027, to appear
-
2026
HydraQE: OSU's Submission for the IWSLT 2026 Speech Translation Metrics Shared Task
Proceedings of the 23rd International Conference on Spoken Language Translation (IWSLT 2026), pp. 323–331
-
2024
Heidelberg-Boston @ SIGTYP 2024 Shared Task: Enhancing Low-Resource Language Analysis With Character-Aware Hierarchical Transformers
Proceedings of the 6th Workshop on Research in Computational Linguistic Typology and Multilingual NLP (SIGTYP), pp. 131–141
*Equal contribution. First place in the constrained subtask.
-
2023
Sentence Embedding Models for Ancient Greek Using Multilingual Knowledge Distillation
Proceedings of the Ancient Language Processing Workshop, pp. 13–22