• Continue building the component library. We have a good start, but much remains to be done. In September, we plan to write spec's and components for (the clusters rooted at) MAKE, JOIN, and REMOVE. Pursue the possibility of collaborating with the VerbNet project.
• Composition operators: building knowledge bases by composing components is key to our research, but our range of composition operators is limited. We have a good start:
  • composition involves unification, and KM implements it well
  • (multiple) inheritance causes composition
  • embedded units, such as:

           (every Foo has (slot ((a Bar with ...))))
                                 ^^^^^^^^^^^^^^^^
                                 an embedded unit
    

    also cause composition
  However, we need more ways to specify composition. Examples include:
  • cliches: Recently Ken proposed "cliches" as a way of packaging and reusing KM axioms. For example, we might specify Locomotion as Move composed with the cliche "reflexive" (i.e. agent=object). (See Ken Barker's working note on cliches.)
  • viewpoints: In some ways, the human body is a container (of blood and guts), but it seems wrong to assert a taxonomic relation. Rather, we would like to say something like:

          (every HumanBody has 
                 (viewpoint ((a Container with 
                                (contents ("Blood-and-Guts"))))))
    

  Both are cases of composition. So, the task involves:
  • defining the space of composition operators
  • designing representations for them
  • implementing them
• develop a general framework for representing the values of properties. This includes several significant sub-tasks:
  • for properties such as size and texture, develop a representation for scalar quantities (e.g. "large", "rough")
  • for properties such as age and area, develop a representation for cardinal quantities (e.g. positive numbers) and units of measurement (e.g. "years", "square feet")
  • for properties such as color and sex, develop a representation for categorical values.
  • design a way of comparing property values, e.g. so they can be used in inference rules
  Use the framework to encode values for the properties in our slot dictionary, starting with the ones crucial for microbiology.
• develop a general framework for representing modifiers, such as "very" and "almost", and reasoning about their affect on the values they modify.
• investigate modal logic and how it can be incorporated into KM. (Modal logic includes ways of expressing "possibly", "usually", "never", and "eventually".) A good starting point is:

            E. A. Emerson, 
            "Temporal and Modal Logic",
            Handbook of Theoretical Computer Science: 
            Volume B, Formal Models and Semantics, 
            J. van Leeuwen, ed., 
            pp.  995--1072, 
            North-Holland Pub. Co./MIT Press, 1990. 
            postscript
  

• continue our preliminary work on representing teleology. Using the results from our work on modal logic, develop a theory of teleology, implemented in KM, that includes the inferences warranted by teleological statements.
• continue our preliminary work on representing modes of existence. The next step is to develop a theory of modes of existence, implemented in KM, that includes the inferences warranted (or blocked) by such statements.
• develop a way of viewing components in their expanded forms. Our components are terse, by design. Many of their axioms come through composition, and currently we rely on mental gymnastics to envision a component in its expanded form. KM is capable of expanding a component, of course, but our browser for the component library does not run KM. Perhaps we can use KM to expand components, cache the results, then display them through the browser.
• coordinate with Pat Hayes and Vinay Chaudhri on representations of time and space. Incorporate their representations into our components.
• design the "component evaluation" for our work. Its purpose is to assess the degree to which:
  • our components represent consensus reality
  • our components can be composed and otherwise reused
  Coordinate with Bob Schrag on conducting the evaluation.

• compile a list of actions, objects, and modifiers that we want to encode in our first release of the component library. Several sources of information contribute to this list:
  • Roget's thesaurus. See Ken Barker's analysis
  • Levine's semantic classes.
  • Art Souther's analysis of the chapters 1-6 of the Alberts text.
• develop standard vocabularies for describing:
  • actions. Our initial vocabulary of "case roles" is based on Ken Barker's dissertation, chapter 3.
  • objects. Our initial vocabulary of "object <-> object properties" is based on Ken Barker's dissertation, chapter 4.
• continue populating our component library. For this summer, the major classes of actions include: Move, Assemble, Make, Change-Possession, Eject, Remove, and Convert.
• develop a method for cross-indexing each of our components to the major linguistic resources used today: Roget's thesaurus, Wordnet, Framenet, Longman's Defining Dictionary, New Oxford Dictionary, and IKB. This information is part of our specification for each component. See a current examples.
• explore ways that we might benefit from (connect with) the VerbNet project, which is just underway at UPenn (see proceedings of AAAI'2000, pages 691-696).
• develop a way of publishing our component library on the internet, primarily to make it available to all team members. Each component will be thoroughly documented and cross-indexed with the other major linguistic resources. See our current component library.
• extend the KM language with new features required by our work on the component library. Peter Clark has the lead on this.
• consult with Peter Clark on question-answering methods and with Jerome Thomere on explanation generation.
• consult with Pat Hayes and Mabry Tyson on connecting our components to the CMap interface.
• consult with IET on evaluating the end-to-end systems.

Back to RKF Group Home Page Created by Bruce Porter Maintained by Dan Tecuci
Last modified August 22, 2000