Motion, space & grammar
Morphosyntactic and semantic patterns in motion-event descriptions, with particular attention to path, manner, subordination, and case marking in Turkish.
Research programme
Language, minds
& machines
I investigate how meaning is structured, processed, and evaluated—from mental simulations of space and motion to culturally grounded assessments of multilingual language models.
My research sits at the intersection of theoretical linguistics, cognitive science, psycholinguistics, and natural language processing. I combine structural analysis with behavioural and experimental methods to study how Turkish and English speakers encode and mentally represent motion and space.
A parallel line of work applies linguistic expertise to AI evaluation. It focuses on low-resource and culturally specific benchmarks, human quality assessment, LLM-as-a-judge methods, and the social questions surrounding bias, trust, accuracy, and responsibility.
Research themes
Three scales of analysis
Structures within language, representations within minds, and evaluations of language technology.
Language & mental simulation
How fictive-motion language influences real-time processing, visual attention, and mental representations of otherwise static spatial scenes.
Human-centred AI evaluation
How linguistic knowledge, cultural context, and accountable evaluation design can make multilingual model assessment more meaningful.
Selected work
Projects & collaborations
Mental simulation, space and language
A cross-linguistic programme examining how fictive-motion expressions in Turkish and English affect language processing and spatial mental simulation. The project combines controlled reading studies, visual-world evidence, and elicited language production.
TurkBench
A broad benchmark for evaluating Turkish large language models across linguistic capability, knowledge, reasoning, safety, and culturally grounded tasks. The published benchmark contains 8,151 samples across 21 subtasks.
Benchmark quality for low-resource languages
A case study of Turkish benchmark datasets that examines data quality and the reliability of automated evaluation, including the opportunities and limitations of using language models as judges.
Trust, bias and responsibility in AI
A mixed-methods investigation of how experts and non-expert users perceive AI systems, and how experience shapes views of bias, trust, accountability, and appropriate responsibility.
Methods
Evidence at multiple levels
Linguistic analysis
Morphosyntactic, semantic, typological, and discourse analysis of naturally occurring and elicited data.
Behavioural experiments
Self-paced reading, picture elicitation, and visual measures for testing processing and mental simulation.
Evaluation design
Benchmark construction, task and rubric design, annotation protocols, quality assurance, and error analysis.
Mixed methods
Quantitative and qualitative evidence for understanding both system performance and human perceptions of AI.
Academic interests
Questions across language & technology
- Large Language Models and NLP
- Morphosyntax, typology, and semantics
- Cognitive linguistics · space and language
- Psycholinguistics · perception and processing of language
- Discourse analysis and pragmatics
- Applied linguistics
- Turkish Sign Language
Research · collaboration · speaking