.
FRDCSA | internal codebases | QUAC

QUAC Architecture Diagram: GIF

Jump to: Project Description | Capabilities

Project Description

In initially thinking about how to write a system - these are the known resources. You have AnswerFinder, AnswerPad, PAI, and QANDA. The first one is not a QA system, but a framework, AnswerPad is untested. PAI is not suited to indexing texts, and QANDA is only half released. We will therefore use the half released component of QANDA (phrag). Somewhere along the lines I acquired a question type classifier, we can use that as well. I did acquire Webclopedia but I have no idea how in the world to set it up. It also depends on a named entity system from BBN, which only runs on Solaris, so we have to have an emulator or vm to run it. Realistically, what we can do is put an answer type classifier onto a sentence level type classfier, and index our domain texts with Glimpse. We could also possibly implement a few routines for cross validation of answers where possible, and probably throw on a search engine scraper as well. Beyond this I don't have any ideas. Actually, we could use the information fusion formula from Javelin's website and interface it with clear, etc, to provide some of that background knowledge knowledge. I really don't want to write such a component (an essential component of mush), but looks necessary at this point.

Capabilities

• quac can use the API for answerbus for general purpose questions.
• quac should first analyze the question against certain contextual modules to see if it applies.
• Make quac operate more generally...
• Hook up learner and quac to create a question question-asker/question-answerer feedback loop
• architect should use quac for question answering over capabilities: how can I edit FSMs?
• Use glimpseserver for quac