Representing Knowledge Soup

In Language and Logic

John F. Sowa

Chief Scientist

VivoMind LLC

A talk presented at the Conference on Knowledge and Logic, 15 June 2002, at the Technische Universität Darmstadt.

Knowledge Soup

Aspects of the Soup


The Problem

How can we extract a crystalline theory from a soupy mess?

Some Views of the Problem

Comments by two logicians and a poet.

Point of agreement:  We make order out of disorder.

Models and Reality

George Box:  "All models are wrong, but some are useful."

Truth is an Indexical

Induction, Deduction, and Abduction

Peirce's three categories of reasoning:

  1. Induction or learning:
    • Start with observations.
    • Look for commonalities.
    • Derive a theory that summarizes the data.

  2. Deduction or inference:
    • Start with a theory.
    • Observe some new data.
    • Use the theory to deduce the implications.

  3. Abduction or guessing:
    • Start with disconnected observations.
    • Guess (hypothesize) a theory that relates them.
    • Test the hypothesis by induction and deduction.

The Knowledge Cycle

A summary of Peirce's theory of pragmatism

Problems of Communication

People with different background, views, theories, and purposes

Natural Language


The talk presented on 15 June 2002 contained a brief overview of a method called Task-Oriented Semantic Interpretation, which implements a technique for dealing with the knowledge soup.

The slides on the TOSI method have been deleted from this section. For more detail on TOSI, see


The framework of knowledge soup can span the gap between natural language semantics and formal systems of logic.

Natural languages must do everything:

  • Express the full range of everything that people do, think, want, need, hope, and fear.

  • Serve as a medium of communication between people with totally different views of what to do, think, want, need, hope, and fear.

  • Enable people to negotiate their differences and reach compromises on specific tasks where they have common interests.

Formal logics have been designed to do one thing at a time:

  • Express clearly delimited propositions about one aspect of one problem for one purpose.

  • Serve as a medium of communication between people or computers that are in complete agreement about the topic, the vocabulary, and the purpose.

  • Support deep reasoning about that narrow topic.

Bridging the gap requires a framework that accommodates both:

  • Support the full richness of natural languages.

  • Represent the precise, narrow theories of formal logic.

  • Enable the enormous scope of natural language to be mapped to appropriate logical theories as needed.

Copyright ©2002, John F. Sowa