The web consists of large quantities of Natural Language text in the form of documents, reviews, comments, chats, tweets, etc. NLP problems, such as Speech Recognition and Machine Translation make up only a subset of the natural language research taking place at Google. I will give an overview of the traditional NLP work and will also focus on an example of a non-traditional NLP task. Consider the case of training a language model to score user queries, distinguishing similar queries from less-similar queries. There are billions of examples of users selecting documents - potentially fulfilling the information-need expressed by their query - but very few examples of users explicitly stating that one query is similar to another. I will present an approach which learns from induced examples in order to model query similarity.
Keith Hall is a Senior Research Scientist at Google, Zurich where he works on Natural Language Processing techniques to improve web search and sponsored search. His research focuses on structured models of language and discriminative machine learning techniques for task-specific language processing. He was previously an Assistant Research Professor of Computer Science at Johns Hopkins University and affiliated with the Center for Language and Speech Processing (CLSP). He completed his Ph.D. in 2004 at Brown University, where he worked with Mark Johnson and Eugene Charniak; following, he held a postdoctoral research position under Frederick Jelinek at the Johns Hopkins CLSP.