Syntactic parsing as a knowledge acquisition problem

Corpus linguistics involves the construction and annotation of large databases of text from spoken and written language. These have applications in NLP and taught grammar. This annotation represents the problem of the KA "bottleneck" in a new application area. This paper introduces parse checking as a KA problem, and compares it to other tree-oriented KA methodologies such as laddering and clustering. It argues that corpus linguistics represents a significant application area for KA. The laddering tools discussed here have been used to process thousands of tree structures. The paper compares two tools in use on the ICE-GB corpus. One tool, ICE Tree II, exploits the structure of grammatical trees more fully than the other. Timing results show a main learning effect which dominates any difference comparison. However, the more integrated tool reduces the scope for error.