The automatic selection of concordance lines.
This thesis presents the results of an experiment into the automatic selection of
concordance lines from very large corpora.
Corpora now exist which are in excess of 100 million words in size, but the increase in
size of corpora brings with it certain problems. These problems are discussed in the light
of information obtained from professional corpus users and the continuing centrality of
the concordance as the main means of interpreting the contents of the corpus is highlighted.
A possible means of overcoming the problems associated with the use of large corpora is
presented. This solution is based upon software which was designed for the purposes of
textual abridgement, this being carried out via an automatic analysis of lexico-cohesive
bonds within the text. An analogy is drawn between conventional text and concordances;
this analogy is then further explored by processing sets of concordance lines with the
modified abridgement software.
In order to determine the success of the approach in identifying concordance lines which
illustrate key features of the node word, an evaluation exercise is carried out, involving
expert corpus users as respondents.