I am a postdoc jointly working at UMD (with Yi Ting Huang) and MIT (with Roger Levy) on sentence processing and child language development. My research program is centered on how language experience (frequency, statistical regularities) and language structure (grammatical or thematic roles) shape sentence processing and comprehension (see further below). My work is heavily experimental yet complemented by computational and developmental approaches. Empirical endeavors aim at benchmarking sentence processing difficulties (magnitude, distribution, and time course) and comprehension accuracy under different experimental conditions (syntactic complexity, semantic anomaly, etc). With children, we examine both ****input (experience) and output (processing and comprehension), and how they are linked. Finally, computational modeling allows quantitative evaluations of theories against benchmark datasets. The resulting interdisciplinary goals and tools are shown in this figure on the right.
Example questions I’ve addressed and their tl;drs:
-How well does next-word prediction alone explain structural effects on reading times? (Huang et al., 2024, JML) → great underestimation of processing difficulty by language models’ surprisal
-What can/can’t language models with developmentally plausible training data predict about child processing data? (Huang, Levy, & Huang, 2025 upcoming BUCLD talk; manuscript in prep) → human-like world knowledge and thematic roles are not learned and/or not utilized as humans in text-based language models
-Do predictive processing and structural processing dissociate in reading? (Timkey, Huang, et al., manuscript in prep; Huang & Dillon, 2023 HSP talk) → structural effects render both small, transient and large, lingering difficulties, the latter not ready captured by language models’ surprisal.
-Why and how do readers misread? (Huang & Staub, 2021, Cognition; Huang & Staub, 2023, PBR; Huang & Staub, under review) → prior linguistic knowledge rapidly accessed to override literal input and recover intended message
-What multiword units can be stored and how are they utilized in real-time? (Huang, 2024, dissertation) → form-meaning mapping more directly accessible for frequent noun-noun phrases than for frequent verb-object phrases, despite co-occurrence frequency and contextual predictability matched
Email: kjhuang at umd dot edu
kjhuang0 at mit dot edu
Address: 46-3027 MIT, 77 Massachusetts Ave, Cambridge MA 02139, USA