Building Hearts and Minds for Robots
Our system uses both human-authored and machine-authored approaches for actor content creation. Some examples are provided below. Others are available on request.When a computer and a human talk with one another there are discrete steps that are taken that allow the computer to understand and manage the communication on the multiple levels that languages demand. Ontologies (knowledge and category management), semantics (the comprehension of symbols, specifically words), and semiotics (the understanding of the relationships of symbols and words).
While only a small percentage of our available tools are shown (most run on a server, via a shell) other demos are available. Please email for more information.
Table of Contents
Video - E&S Demonstration
Video - HeadCase Demonstrations
Demo Actor (Web Implementation)
Demo Actor (Facebook Implementation)
Demo Actor (Second Life Implementation)
Conceptualizer
Scrapers
NLPFacade
Segmentation Tool
Similarity Tool
Video Demonstrations
Video - E&S Demonstration (2008)
Video - HeadCase Demonstrations of Foundational work (2007)
Product Demonstrations
These chat-bot demonstrations each produce human-authored results. These three implementations show the same hybrid approach in three different systems; web, facebook, and Second Life. Other systemsWeb-based Implementations of Actors [Human Authored]
• - Demo Actor #1
• Function: FAQ system / Language Generation (human authored)Facebook-based Implementations of an Actor [Human Authored]
• Version: v 0.1
• Description: This simple conversational system demonstrates traditional "chat-bot" technology. It may be implemented in other systems (as seen below) and allows for a writer to insert answers to anticipated questions using simple rulesets in XML-like format.
• Demo Actor (Facebook Implementation) [Human Authored]• Function: FAQ system / Language Generation (human authored)Second Life Implementation of Actors [Human Authored]
• Version: v 0.1
• Description: This places traditional "chat-bot" technology into the Facebook social media system.
• BabyNo (Personality #1)
• Pighed (Personality #2)
• Function: FAQ system / Language Generation (human authored)
• Version: v 0.1
• Description: This places traditional "chat-bot" technology into the Second Life virtual world system.Technology Demonstration (Snapshots)
Our natural language tools are under development. Below are selected pieces that show work in progress. In the coming months these tools (each one a link in the natural language processing, understanding, and generation chain) will continue to undergo development and improvement with the eventual goal of replacing all human-authored efforts.Conceptualizer [Computer Authored Language Generation Tool]
• Function: Ontology generation / visualization tool.Content Scrapers [Computer Authored Language Generation Tool]
• Version: v 0.2
• Description: The E&S Conceptualizer transforms arbitrary sites into a rich ontology framework described by an RDF-like schema. Thus, E&S can extract Subject-Verb-Object logic (or triples) out of ordinary text. Our basic workflow allows us to build extraction tools for any structured text in a few hours. (Note: at this point only wikipedia is supported).
• Function: Content generation / authoring tool.
• Version: v 0.1
• Description: We use 'scrapers' to collect content from existing sources, such as screenplays. In this example words that have been taken from the Blade Runner screenplay have been broken into an n-gram analysis. Using this data we are able to then predict the liklihood of that same string recurring later in the text, as well as provide a liklihood that the system would need to generate that word, when replying to a human. Text output are provided.Technology Demonstration (To Come)
NLPFacade [Computer Authored Language Generation Tool]
• Function: Examines sentences and creates tags that identify parts of speechSegmentation Tool [Computer Authored Language Generation Tool]
• Version: v 0.1
• Description: Identifies chunks of language that are important and provides a compliment to tokenization so that meaning can be generated from words. "Heteronyms" are words that have the same spelling but different meaning.
• Function: Creates "tokens" to facilitate linguistic processingSimilarity Tool [Computer Authored Language Generation Tool]
• Version: v 0.1
• Description: This categorizes blocks of text into what are called "tokens." Think of a token as a linguistic symbol; it might represent a letter, an image, a gesture or a sound. This step is needed before the parts of speech and the semantics of that speech may be processed. This formatting is necessary for high-speed processing of large quantities of text. It is also what allows us to work outside of language, allowing a concept to be used instead of a particular word.
• Function: Measures the semantic distance, or meaning, between words
• Version: v 0.1
• Description: This important device allows semantical analysis, or understand the meaning, of a sentence. By entering a test word, a target word, and a threshold number, we are able to compute their similarity in meaning. Anything over the threshold is bold-faced, which indicates a great distance between words, such as "felt" and "sewer." This is currently very hard on the RAM of a given machine and is designed to be deployed on larger servers that can accomodate more memory.
Other Info
We know alot about this stuff. So if you'd like more information, please let us know.
