* News


A Hybrid Inference Approach to Natural Language Understanding

Prof. Nicholas L Cassimatis
Rensselaer Polytechnic Institute

November 10, 2011
JEC 3117 - 4:00 p.m. to 5:00 p.m.


In creating systems that process human language, there has been a tradeoff between the depth of understanding of methods based on reasoning over complex knowledge structures versus the robustness and flexibility of systems based on statistical approaches. In this talk, I demonstrate an approach to addressing this tradeoff using an inference system developed using the Polyscheme computational cognitive architecture. A key step is to decompose complex knowledge structures can be decomposed into more granular probabilistic constraints and how many aspects of language understanding can be cast as hybrid logical-probabilistic inference problems. Such problems involve several challenging computational complexity issues that are addressed by integrating multiple data structures and algorithms using Polyscheme. I demonstrate how this approach addresses some difficult problems in understanding ungrammatical and non-literal language as well as some advances enabled in human-robot interaction.

Hosted by: Prof. Elliot Anshelevich (x6491)

Last updated: November 4, 2011