Dongyeop KANG (강동엽)


dongyeok@cs.cmu.edu

I am a Ph.D. student in the Language Technologies Institute of the School of Computer Science at Carnegie Mellon University, working with my fantastic advisor, Eduard Hovy. Formerly, I also worked with Eric Xing. During my Ph.D. study, I interned at Facebook AI, Allen Institute for Artificial Intelligence (AI2), and Microsoft Research. In the middle of the study, I finished my alternative military service in South Korea at Naver Labs and KAIST Institute as a researcher. Before joining CMU, I obtained my BS and MS in Computer Science Engineering at KAIST, Korea, focusing on large scale data mining.





Ph.D. Thesis Proposal: Research Statement (short version):
Natural language generation (NLG) is a key part of the pipeline in many NLP applications such as dialogues, summarization, and more. One might think that the only information an NLG system needs is that contained in the utterance. However, usually many different sentences can say the same thing, and most of these have slightly different connotations not explicitly mentioned in the text. To select more appropriate or human-like output, the system needs to be guided by additional information such as the knowledge, the speaker’s persona, the relationship with the hearer, the structures, and many other parameters.
One of M. Halliday's linguistic theory called Systemic Functional Linguistics (SFL) (1978) suggests that such information could be categorized into three metafunctions; ideational, textual, and interpersonal, where each contains a lot of separate types of information called facet. My thesis repackages some of them, particularly focusing on the three facets; knowledge, structures, and styles, to make the NLG system more (1) knowledge-guided, (2) coherently structured, and (3) stylistically appropriate. By combining different facets and dynamically modeling their interactions, I believe one can build a more human-like NLG system (e.g., PAULINE (Hovy, 1987)). For each facet and its combination, I describe my research questions and approaches as follows: Research Interests:

News



Publication

2019


2018


2017


Before



Services



Experience



Awards



Teaching


Last updated in November 2019