Cognitive Program Compiler
Abstract
Cognitive Programs (CP) specify computational tasks for the executive controller of visual attention. They are built on top of the Selective Tuning (ST) model of attention and its extension, STAR. Currently, the common way of specifying CPs is via diagrams, which are neither standardized nor directly machine-readable. This necessitates tedious and time-consuming implementation of CPs by hand, which slows research and prevents rapid experimentation.
This thesis presents the specification and reference implementation of the Cognitive Program Compiler (CPC). The CPC reads tasks written in the Cognitive Program Description (CPD) format, based on a novel controlled natural language called Imperative English (IE). The CPC can then output executable code in a regular programming language. The reference implementation is easily extensible, and several output modules are provided.
The CPC output has been evaluated by specifying several real-world psychophysical experiments and comparing the generated code against known human performance for those experiments.
This thesis presents the specification and reference implementation of the Cognitive Program Compiler (CPC). The CPC reads tasks written in the Cognitive Program Description (CPD) format, based on a novel controlled natural language called Imperative English (IE). The CPC can then output executable code in a regular programming language. The reference implementation is easily extensible, and several output modules are provided.
The CPC output has been evaluated by specifying several real-world psychophysical experiments and comparing the generated code against known human performance for those experiments.