Chenyi (Skylar) Ding
MA student in Linguistics, University of California, Davis
I am a second-year MA student in Linguistics at UC Davis, where I work with Emily Morgan on computational psycholinguistics. Most of my work uses language models as a stand-in for the statistics of linguistic experience and asks how far those statistics go in explaining what people do.
With Emily Morgan, I am studying whether second-language readers rely on syntactic and semantic cues in the same way native readers do. Rather than treating language model surprisal as a single black-box predictor, we decompose it into a syntactic and a semantic component and ask whether the two components predict reading times differently for native and non-native readers. With Masoud Jasbi, I am asking how much of the order in which children learn words can be predicted from the semantic structure of child-directed speech, using language models trained on child-scale corpora as a proxy for that input. The same question, what language models capture and what they miss, runs through my work with Seth Frey’s Computational Communication Lab, where we study why large language models used to code academic texts fail on some variables but not others, and use pragmatic theory to predict where those failures will land.
Before Davis, I worked with Hong Liu and Darcy Sperlich at Xi’an Jiaotong-Liverpool University on two eye-tracking projects on second-language processing. With Hong Liu, I designed and ran a visual-world experiment on prediction in Chinese–English bilinguals, from stimulus norming through data collection and analysis; the paper is under review at Second Language Research. As a research assistant to Hong Liu and Darcy Sperlich, I built the reading stimuli and ran a sentence-reading eye-tracking study on how sentence predictability affects lexical processing.
I will be applying to PhD programs for Fall 2027, and plan to keep working at the border between computational modeling and experimental psycholinguistics.
News
- Talk with Emily Morgan on decomposing language model surprisal in L1 and L2 reading accepted at LSA 2027 (San Francisco, Jan 2027).
- Paper on age of acquisition in BabyLM representations accepted at the BabyLM Workshop, EMNLP 2026 (Budapest). A poster on the same project accepted at BUCLD 51 (Boston, Nov 2026).
- Presented work on semantic entropy as a reliability diagnostic for LLM-based content analysis at IC²S² 2026 (Burlington, VT).
- Started a summer Graduate Student Researcher appointment with Emily Morgan, working on decomposing language model surprisal in L1 and L2 reading.
Publications
Papers
Reading acquisition order out of BabyLM representations: Child-scale training produces developmentally aligned semantic spacesforthcomingDing, C., Houghton, Z. N., & Jasbi, M. Proceedings of the BabyLM Workshop at EMNLP 2026, Budapest, Hungary.
The joint influence of proficiency and cognitive processing efficiency on L2 predictive sentence processingunder reviewDing, C., & Liu, H. Second Language Research.
Working papers
Why here and not there? A pragmatic account of selective errors in LLM-based construct extractionDing, C., Chen, S., Zhao, Z., Ye, M., & Frey, S.
Mapping construct operationalization practices: An LLM-based extraction approach to detecting jingle-jangle fallacies in social scienceChen, S., Ding, C., Zhao, Z., Ye, M., & Frey, S.
Talks and posters
Decomposing language model surprisal reveals differential cue weighting in L1 and L2 readingDing, C., & Morgan, E. Talk, Annual Meeting of the Linguistic Society of America (LSA 2027), San Francisco, CA, January 2027.
Children’s word-learning order is already in their input: Age of acquisition in the semantic space of a child-scale language modelDing, C., Houghton, Z. N., & Jasbi, M. Poster, 51st Boston University Conference on Language Development (BUCLD 51), Boston, MA, November 2026.
Measuring pragmatic difficulty in LLM-based content analysis: Semantic entropy as a diagnostic for variable-level reliabilityDing, C., Chen, S., Zhao, Z., Ye, M., & Frey, S. Talk, 12th International Conference on Computational Social Science (IC²S²), Burlington, VT, July 2026.
Why here and not there? A pragmatic account of selective failures in LLM-based construct extractionDing, C., Chen, S., Zhao, Z., Ye, M., & Frey, S. Talk, Communication Horizons @ UC Davis 2026, Davis, CA, May 2026.
Pragmatic vulnerability in academic discourse: Using semantic entropy to predict where LLMs failDing, C., Chen, S., Zhao, Z., Ye, M., & Frey, S. Talk, 12th Annual UC Davis Symposium on Language Research (CLR), Davis, CA, May 2026.
Teaching
Teaching assistant, Department of Linguistics, UC Davis
- LIN 177Computational LinguisticsFall 2026
- LIN 106English GrammarSpring 2026
- LIN 1YIntroduction to LinguisticsWinter 2026
- LIN 1Introduction to LinguisticsFall 2025
Grader
- LIN 105Topics in Sign Language LinguisticsFall 2026
CV
A full CV, with education, publications, research and teaching experience, skills, and service, is available as a PDF.