Hyewon Jang
Computational linguist
About Me
I work as a postdoctoral researcher at CLASP at the University of Gothenburg with Sharid Loáiciga, Shalom Lappin, and Jey Han Lau on a project about the role of multimodal context for sentence acceptability judgment. I received a PhD in computational linguistics at the University of Konstanz, advised by Diego Frassinelli and Bettina Braun. My PhD project focused on analyzing how language models process pragmatic information that makes human language sophisticated and nuanced, with sarcasm as an instance of it. My PhD thesis is about the examination of computational sarcasm detection through the lens of psycholinguistics.
Research Interests
In short, I like understanding how complex communicative phenomena are processed by humans and computers. Below are more specific keywords about my research interests.- Psycholinguistic experiments: I design experiments for human participants to test hypotheses about linguistic phenomena and analyze collected data from multiple angles to reveal facts about how humans and language models process language.
- Computational linguistics: I probe the capabilities of language models using ideas and data from psycholinguistics. Though humans and computers have their differences in processing language, comparing the behaviors of models and humans serves as a useful indicator for understanding what models can or cannot do.
- Multimodality: Communication through language happens in multiple modalities. I examine the effects of textual and audiovisual modalities on human communication.
- Pragmatics and utterance interpretation: A lot of subtlety and complexity occur in human communication by the use of pragmatics. I examine ways in which we can pull these complexities out of an utterance and lay them out in a transparent manner to process utterances better.
- Motivation and utterance: An utterance is often the result of underlying emotional triggers and motivations. I use these underlying elements to analyze the capabilities of language models.
- Perspective-aware models: I am interested in investigating whether language models can navigate through different communicator roles (speaker vs. listener vs. observer). Do language models assume the role of a speaker or a third-party observer in solving various linguistic tasks, if at all?