Speaker "Chandra Khatri" Details Back
-
Name
Chandra Khatri
-
Company
Got It AI
-
Designation
Chief Scientist
Topic
Detecting Sensitive Content and Understanding Language in Open-Domain Dialog Systems
Abstract
In this talk, I will go over the problem of detecting Sensitive Content (such as Profanity, Sexism, and Racism) within the language, an extremely challenging problem even for humans. I will describe the Two-stage Large-scale Semi-supervision technique for detecting the sensitive content and showcase how this approach generalizes well across a variety of sensitive classes. Furthermore, task-based Dialog Systems (e.g. most Voice Assistant such as Alexa, Google Assistant) have a well-defined metric associated with the understanding and completion of each task (e.g. "play music", "inform weather", etc.), which is not the case of Open-Domain Dialog Systems due to subjective nature of the problem. In this talk, I will also go over the techniques for Understanding Language in such Dialog Systems.