Formalizations of Commonsense Psychology

1
Formalizations of Commonsense Psychology

• Authored By Andrew Gordon Jerry Hobbs
• Presented By G. Ryan Anderson

2
Introduction to Commonsense Psychology

• Concerns all of the aspects of the way people
  think they think
• Plans, goals, threats, emotions, memories, and
  all other mental states that can exist
• Central in our ability to reason and make deep
  inferences
• More informed by the field of cognitive
  psychology than by AI

3
How could this be useful to AI?

• Systems that can successfully reason about people
  are likely to be much more useful than systems
  grounded in the natural sciences

4
How can this be used in AI?

• Formal axioms for commonsense knowledge
  representation
• Theories with adequate competence
• Theories with adequate coverage

5
Representational Requirements

• Group aspects and characteristics of various
  domains into a manageable set of representational
  areas
• 48 total representational areas

6
Representational Requirements

• Sample set of representational areas below
• Representation somewhat incomplete, and more
  elaboration is necessary

7
Using Natural Language for Commonsense
Representation

• Natural language is very effective in making
  conceptual distinctions
• Language-based methodology for elaborating on the
  previously mentioned representational areas

8
Step One Expression Elicitation

• Acquire an initial set of words, expressions, and
  sentences used to relate to a given
  representational area

9
Step Two Lexical Expansion

• Take our set of expressions from step one
• Search for related words and expressions, using
  linguistic resources
• Builds up the quantity of expressions for a given
  area, thus giving a deeper degree of coverage

10
Step Three Corpus Analysis

• Collection of a large database of examples of
  language use in the representational area

11
Step Four Model Building

• Review the results of step three to understand
  the distinctions made in real language use
• Clustering of sentences, words, and other
  expressions into sets where they are synonymously
  used

12
Building a Commonsense Theory of Memory

• Having built a set of representational
  constructs, we can now axiomize our data into
  formal theories
• Memory Retrieval is one of the 48 areas mentioned
  earlier

13
Concepts in Memory
14
Accessibility of Concepts
15
Association of Memories
16
Association of Memories (cont.)

• Example of formal axiom for concept association

17
Other Aspects of Memory

• Remembering
• Forgetting
• Repressing

18
Conclusions

• Development of theories with adequate coverage
  and competency
• Sorted into manageable amount of domains
• Elaborate on domains using natural language
  tactics
• Results can be moulded into more formal axioms

19
Conclusions (cont.)

• Memory is one of 48 representational areas
• Challenge lies in integrating all 48 areas
• Overall goal is to construct AI systems that have
  a solid representation of human commonsense
  models to more effectively reason the way humans
  do

20
Questions?

• Paper can be found at http//people.ict.usc.edu/g
  ordon/AIMAG04.PDF