Timothy Miller

Research Fellow

I have worked in the field of computational linguistics for several years, and in the field of biomedical/clinical informatics for just over a year. My contributions in the former field largely involve my thesis work on incremental syntactic models for processing spoken language. These models are unique in that they work the way people work – one word at a time which is incorporated into the current hypothesis of the syntax of the current sentence, with the capability of modeling and incorporating disfluent speech. Work in this area was published in a series of articles at top computational linguistics conferences (Association for Computational Linguistics annual meeting [ACL] and Empirical Methods in Natural Language Processing [EMNLP]).

In the field of biomedical and clinical informatics I have worked on the problem of clinical coreference resolution, and have been working in the areas of relation extraction and negation detection as well. I have also worked on clinical speech recognition -- in a study that evaluated the performance of a domain-adapted speech recognition interface as a front end for question answering. These results show that domain-specific speech recognition in the biomedical domain is a difficult enough problem that domain-specific methods are probably required to match open domain performance (while open-domain document retrieval performance is unaffected by speech recognition errors). In general, my focus is on bringing state of the art NLP methods from the general domain and using them to build models in the clinical domain. As I learn about the clinical point of view on the interesting problems to be solved I hope to contribute to the field of medicine as well as informatics.

My thesis work and recent work are superficially different, though there are deep principles that have guided my research to date and future research plans. Specifically, I have always been interested in making intelligent systems designed to improve human-computer interaction through computer understanding of human language. This manifested in my thesis work in modeling the way that humans actually use language, with occasional mistakes. More recent work has focused on making medical systems that have the advanced capabilities of computers with interfaces (such as speech) that are natural and optimal for human use.