Niranjan Balasubramanian

Affiliation: 
Stony Brook University
Talk Title: 
Knowledge Extraction and Reasoning
Event Type: 
Brown Bag Talk
Semester: 
Spring 2016
Date: 
Wednesday, March 30, 2016, 1:00 pm - 2:00 pm
Location: 
SBS S-207 (Linguistics Seminar Room)
Abstract:
 
Knowledge acquisition and reasoning are essential for many Artificial Intelligence (AI) applications. NLP has made excellent progress in extracting factual knowledge. For AI systems to go beyond factual lookup, we need to acquire inference supporting knowledge, and robust systems that can reason with text-derived knowledge. 
 
While vast amounts of knowledge is available in online texts, extracting them from natural language in itself is a hard task. Most knowledge extraction techniques depend on having i) a pre-specified relation vocabulary for the knowledge, and ii) access to large amounts of labeled data, both of which limit scalability. My recent work explores how regularities in language can be harnessed to address these scalability challenges. I will present three examples. First, I will present work on inducing shallow schemas that capture how events are typically conveyed in text. Next, I will present work on extracting inference supporting knowledge from a 4th grade science text and our troubles with using the extracted knowledge for efficient reasoning. Last, I will talk about work on extracting medical relations from scientific biomedical texts and the challenges involved.