Place: CS Conference Room in MUDD
Abstract:
The Aquaint project at Columbia addresses questions that solicit open-ended answers, or in other words, questions that inherently can be answered in more than one way. Like many NLP applications, this requires the researcher to define the application goals, or in other words to conduct feasibility studies. I will introduce what I mean by Guided Information Alignment, an approach to information extraction informed by text structure. Then I will discuss two aspects of the current feasibility phase. The first consists of defining an annotation task and the data it will yield. The second involves examining the language features of texts that can potentially inform the implementation, e.g., lexical features.