While at Microsoft Dustin’s research focused on improving natural language understanding for speech.  He developed systems that eased spoken interactions by allowing users to use natural speech, rather than just keywords.


While at Yahoo! Dustin’s research focused on improving ad relevance and revenue for the production sponsored search system.  He collaborated with a great team to develop new approaches for modeling ad relevance, matching, and clicks that led to significant increases in revenue for a system that generated billions of dollars per year for Yahoo!  Dustin was a frequent owner of online “bucket” tests to evaluate new algortithms and models live on millions of users.


While pursuing his PhD, Dustin’s research focused on automatic prediction of sentence structure for speech.  His work developed approaches for predicting sentence boundaries and commas and then improving downstream systems such as named entity tagging, parsing, and machine translation.  He also developed new approaches for improving word confidence estimates and combining the output of multiple speech recognition systems.  His work was funded by DARPA, on the GALE and EARS projects.  In 2006 he also received the MIT Lincoln Labs Fellowship and a NSF DDEP Award.  He was a guest researcher at i6, RWTH in Aachen, Germany (March-August 2006) in Hermann Ney’s lab. In addition, Dustin was a Visiting Fellow at the Program on Networked Governance, Kennedy School of Government, Harvard (2006-2007), where he applied machine learning approaches to the analysis of Congressional activity.


As an undergrad, Dustin worked on a NSF project to understand meetings.   His research developed machine learning approaches to automatically detect regions of agreement and disaggrement based on lexical and prosodic cues.  He recieved the UW Electrical Engineering Outstanding Undergraduate Research Assistant Award.


A list of publications resulting from the above research.