Andrew S. Gordon

The Commonsense Psychology Resource Page

Studying the way people think they think

1. Overview

I am particularly interested in the commonsense models that people have of their own reasoning processes, i.e. commonsense psychology. Commonsense psychology has been studied by many researchers across a number of fields in the past, sometimes under the heading of Naive Psychology, Theory of Mind, and Social Cognition.

Our efforts at USC have focused in four areas. First, we have defined the representational scope of commonsense psychology through a large scale analysis of planning strategies. Second, we have formalized commonsense psychology concepts into axiomatic, predicate calculus theories. Third, we have identified all of the English words and phrases that are used to refer to concepts in commonsense psychology, and encoded them as local grammars. Fourth, we have specified a cognitive architecture based on commonsense psychology that integrates heterogenous cognitive models under a common framework.

These first of these efforts has been described in a number of published reports, particularly the following:

2. Logic

In the tradition of formal knowledge representation, Jerry R. Hobbs and I have begun the authoring of a large collection of integrated theories that each describe components of our commonsense psychological models. To aid in reusability, these theories are being authored as standard predicate calculus theories. We have written a very large book that contains all of these theories, along with a description of our project, and are looking for a suitable publisher.

3. Language

In an effort to better understand the relationship between language and representation, we have authored a lexical-semantic resource for identifying reference to commonsense psychology concepts in English text. This resource was authored as a set of local grammars, encoded as finite-state transducers, that can be applied to large text corpora for concept-level tagging & markup. This resource was authored using the Intex NLP development environment authored by Max Silberztein, originally at the LADL, University of Paris 7, and now at the GRELIS/LASELDI, Universite de Franche-Comte, France. These local grammars are constructed on top of a large-coverage dictionary of English words compiled by Blandine Courtois.

Our group completed finite-state transducers for the 30 representational areas of Commonsense Psychology that were identified through strategy representation. A total of 529 concept tags are used in these transducers. Provided here are the graphs associated with each of these representational areas, in the original Intex graph format.

These resources have been applied in a number of different language analysis projects, and have been evaluated to determine their precision and recall performance. The following reports are available:

4. Cognitive models

One of the specific aims of this research program is to support the devleopment of integrative cognitive models of human cognition. We believe that our commonsense psychological models can serve as an important integrator of existing computational models of different reasoning modules, including planning, scheduling, memory retrieval, similarity judgements, belief revision, explanation, prediction, among others. The following papers are available:

5. Project team

The principle investigators for this research effort are Andrew S. Gordon (gordon @ ict.usc.edu) of the USC Institute for Creative Technologies and Jerry R. Hobbs of the USC Information Sciences Institute (hobbs@isi.edu). Current and past team members include USC students, namely Abe Kazemzadeh, Milena Petrova, Anish Nair, Jeen Jeong, Reid Swanson, Ryan Kashfian, Ashish Mestry, and Rachana Pandya.