New Applications of Natural Language Processing: Measuring Ideas and Machine-in-the-Loop Creativity
Presented by Dr. Noah Smith, University of Washington

Abstract:

Natural language processing has many recent success stories, but we should be even more excited about new applications on the horizon. In this two-part talk, Smith will discuss new applications of automated natural language analysis and natural language generation. The first takes an idea expressed by a human in English, and estimates the extent to which the idea is expressed in a given text. Aggregating over a corpus, or making comparisons between different subcorpora, leads to a statistical problem that's distinct from familiar problems of retrieval. Smith will present some initial findings on measuring ideas in corpora and on the aggregation problem. The second part of the talk turns to generation. While current work tends to focus on transforming content (translation into another language, or summarizing a document) or on conversations between humans and machines, he'll suggest that natural language generation models have a role to play in assisting creative language production by humans. Smith will introduce creative writing with a machine in the loop and present findings from our initial user studies along with the latest models for generating language in this setting.

His collaborators on the work are PhD students Lucy Lin, Dallas Card, Elizabeth Clark, and Anne Ross, postdoc Yangfeng Ji, former postdoc Chenhao Tan (now on the faculty at UC Boulder), and Scott Miles of UW's Department of Human Centered Design & Engineering.

Noah Smith's bio can be found here: https://homes.cs.washington.edu/~nasmith/bio.html