Skimming for Context

- is the name of a master's thesis in computer science by Ole Torp Lassen from Dept. of Computer Science, University of Copenhagen, (DIKU).

The defence was on:
wednesday dec. 7th 2005 - 13.00 - 15.00,
Auditorium 2, August Krogh Instituttet,
Universitetsparken 13, 2100 København Ø.

Supervisors were :
Professor Neil D. Jones (DIKU).
Assistant Professor Peter Rossen Skadhauge (CBS).

Censor was :
Professor Henning Christiansen (RUC).

Abstract:

The dissertation represents a research proposal that involves an experimental method - skimming, for the partial recognition of the semantic context in natural language texts.

Lexical ambiguity is among the biggest obstacles for the development of generally applicable natural language processing systems. I.e., systems that can be applied in more than the most restricted of circumstances. The degree of lexical ambiguity in a text can in genral be reduced the more the semantic context of the text can be confined/defined. The semantic context of a natural language text is proposed to be represented by those concepts and conceptual relationships, that are central to the situation described in the text.

The experimental algorithm for "skimming" exploits semantic relationships between concepts in order to narrow in on the semantic context at hand. The experimental prototype developed as a part of the project may serve as a solid vantage-point for further research and development of the general method.

Skimming for Context.pdf, (12.5 MB) or zipped, (8.6 MB)
The dissertation in full (English).
Advisors: Neil D. Jones, Peter R. Skadhauge.

Skimming for Paragraphs.pdf, (565 KB)
A spin-off paper, applying the proposed method to the problem of finding semantically well founded partitons in natural language text (English).
Advisors: Neil D. Jones, Peter R. Skadhauge.

Skimming in Comparison.pdf, (388 KB)
Another spin-off paper comparing the thesis to similar work in the field (English).
Advisor: Neil D. Jones.