While current Q&A systems are competent and useful with respect to the
information they process, they are very limited when compared to a conversation an
analyst could have with a human who has read the same information. Solomon, a
radically new Q&A system that will transcend the limitations of existing
systems by approaching real conversation with real humans.
Solomon is capable of producing rational, justified answers for conceptual,
hypothetical, and even open-ended questions related to knowledge bases derived from
reading documents. The theoretical approach underlying the system - which, in
short, is to model Q&A on a more sophisticated form of human-machine
interaction: one in which the machine has the power of cutting-edge machine
Six distinguishing attributes of Solomon are:
- Knowledge Acquisition via Reading
Solomon acquires knowledge through a process akin to how humans learn by reading,
not by shallow text extraction technology. The knowledge Solomon acquires by
reading far exceeds the knowledge acquired by current Q&A systems, which cannot
extract arbitrarily complex knowledge from text.
- Human - Computer Collaboration via Conversation
Both Solomon and its users are active participants in the question answering
process, with each asking and answering questions of the other. Their
collaboration is in the form of a dynamic conversation in English wherein the
answer to a question depends in part on the prior conversation (the questions
previously asked and answered). Solomon's conversational Q&A is not
reducible to the decomposition of a single complex query.
- Natural Suppositional Reasoning
Solomon is not limited to simply answering questions of fact. Solomon supports
conversation-based suppositional reasoning, i.e., what if... questions
that introduce logical and linguistic contexts wherein further conversation is
interpreted and evaluated.
- Defensible Answers, Rational Justifications, & Intuitive Explanations
Solomon incorporates sophisticated automated reasoning and model finding. Answers
and justifications relate to either counter-examples or defensible arguments
(proofs and arguments by deductive, inductive, abductive, or probabilistic means).
These answers and justifications are explained in an intuitive fashion, in English,
as part of the normal course of conversation.
- Unified Reasoning over Visual and Symbolic Knowledge
Solomon is able to answer questions requiring comprehension of visual as well as
symbolic information. It is not that Solomon reduces the visual to the symbolic,
for diagrams, pictures, satellite images, movies, maps, etc., all the things that
are at the heart of human-level Q&A, are most certainly not symbolic entities.
Solomon uses a new family of visual logics, known simply as Vivid, to represent
and reason directly over any computable image.
- Seamless Integration of Existing Q&A Systems
Solomon can be integrated on top of other existing databases and Q&A systems as
a meta-Q&A system. Solomon extends Attributes 2-5 across various disparate
domains and specialized systems through the sound decomposition of proof-theoretic
operations into direct model inspections that are then submitted to subordinate
systems as yes/no questions of fact.
May 2007 AQUAINT Meeting Materials
RTE DEV 2007 Submission