Technical reports

Memory and context mechanisms for automatic text processing

Hiyan Alshawi

192 pages

This technical report is based on a dissertation submitted December 1983 by the author for the degree of Doctor of Philosophy to the University of Cambridge, Trinity Hall.

DOI: 10.48456/tr-60

Abstract

The thesis describes memory and context mechanisms for natural language text processing. The mechanisms were implemented as part of a computer system that successfully processed a number of short descriptive English texts producing output that can be used to create a relational database. The memory mechanism is concerned with representing and retrieving various kinds of knowledge, while the context mechanism is concerned with accumulating and applying information specifying which fragments of knowledge are currently more salient.

The mechanisms are used in the implemented system by an interpretation component dealing with common language interpretation problems that cannot be handled by simple sentence-level analysis. These problems include resolving references, disambiguating word senses, and discovering implicit relationships. The mechanisms are also used by a task-specific component which carries out the database capture application using database descriptions stored in memory. The choice and handling of the particular application task, interpretation operations, and types of context information, were designed to check that the computational techniques developed for memory and context provide appropriate apparatus for non-trivial text processing involving a wide range of phenomena of language interpretation in context.

The memory representation formalism is based on hierarchies for classifying entities and the associations between them. It has the advantage of simplicity and a well designed semantics. Retrieval from memory is performed by marker processing on a network structure. The context mechanism represents instances of various types of context information as “context factors” which can be combined to derive activation values for memory entities. Context activation is used to choose the results of memory operations and to restrict memory searches. Context factors are created and modified as a result of text processing operations, leading to a gradual alteration of the context representation. Both the memory and context mechanisms utilize an indexing scheme that uses semantic clustering criteria. This increases the efficiency of retrieval from memory and allows efficient access to entities with high activations derived from several factors while individual factors can be managed independently.

Full text

PDF (11.9 MB)

BibTeX record

@TechReport{UCAM-CL-TR-60,
  author =	 {Alshawi, Hiyan},
  title = 	 {{Memory and context mechanisms for automatic text
         	   processing}},
  url = 	 {https://www.cl.cam.ac.uk/techreports/UCAM-CL-TR-60.pdf},
  institution =  {University of Cambridge, Computer Laboratory},
  doi = 	 {10.48456/tr-60},
  number = 	 {UCAM-CL-TR-60}
}