AI Lunch and Seminar

Ramanathan V. Guha - Communicating Semantics

Abstract: Messages often refer to entities such as people, places and events. Correct identification of the intended reference is an essential part of communication. Lack of shared unique names often complicates entity reference. Humans do not and cannot have a shared unique names for everything. Yet, we communicate in our daily lives about things that do not have a unique name (like John McCarthy) or lack a name (like his first car). Our long term goal is to enable programs to achieve communication just as effectively.

Humans overcome the lack of shared unique names by using descriptions. We use some shared language and shared domain knowledge to construct uniquely identifying descriptions for such entities. We call this `Reference by Description' and argue that it could form the basis for a programatic use of descriptions.

We introduce a mathematical model of `Reference by Description'. We use an information theoretic approach to address questions such as: What is the minimum that needs to be shared for two communicating parties to understand each other? When can we bootstrap from no shared language? What is the computational cost of using descriptions instead of unique names?

In addition to their inherent interestingness, these questions also have practical applications that range from database/application integration to privacy preserving information sharing.

Bio: Guha is the creator of widely used web standards such as RSS, RDF and Schema.org and products such as Google Custom Search. He has made substantial contributions to Netscape’s browser, Google’s Adwords, Google Now, etc. He was a co-founder of Epinions.com and Alpiri. Until recently, he was a Google Fellow. He has a Ph.D. in computer science from Stanford University and B.Tech in mechanical engineering from IIT Chennai.