Dota - Tianai Dong

Hi! I am a third-year PhD student, affiliated with Multimodal Language Department at the Max Planck Institute for Psycholinguistics, and Predictive Brain Lab at the Donders Institute (Centre for Cognitive Neuroimaging). I am co-advised by Floris de Lange , Lea-Maria Schmitt , Stefan Frank, and Paula Rubio-Fernández . I also work closely with Mariya Toneva at the Max Planck Institute for Software Systems. I am funded by an IMPRS fellowship.

I study how humans acquire, mentally represent, and generate predictions about language through our rich multimodal experiences. My approach combines computational methods with insights from neuroscience, linguistics, and psychology, with the dual goals of understanding the human mind and advancing artificial intelligence.

If you want to discuss any academia-related topics, please feel free to reach out to me :)

Email // Google Scholar // BlueSky

We evaluated seven MLMs against humans on three word classes: vocabulary words, possessive pronouns, and demonstrative pronouns. We observed a consistent difficulty hierarchy shared by both humans and models, but a clear performance gap remains: while MLMs approach human-level performance on vocabulary tasks, they show substantial deficits with possessive and demonstrative pronouns.

We propose to probe a pre-trained multimodal video transformer model, guided by insights from neuroscientific evidence on multimodal information processing in the human brain.

We present DiscoGeM, a crowdsourced corpus of 6,505 implicit discourse relations from three genres: political speech, literature, and encyclopedic text.

We assess the performance on explicit connective identification of four parse methods (PDTB e2e, Lin et al., 2014; the winner of CONLL2015, Wang and Lan, 2015; DisSent, Nie et al., 2019; and Discopy, Knaebel and Stede, 2020), along with a simple heuristic.

We define and evaluate a methodology for extracting history-dependent spatial questions from visual dialogues.

Template from here