Alignment-based expansion of textual database fields

Abstract

Our study describes the induction of a secondary metadata layer from textual databases in the cultural heritage domain. Metadata concept candidates are detected and extracted from complex fields of a database so that content can be linked to new, finer-grained labels. Candidate labels are mined drawing on the output of Alignment-Based Learning, an unsupervised grammatical inference algorithm, by identifying head - modifier dependency relations in the constituent hypothesis space. The extracted metadata explicitly represent hidden semantic properties, derived from syntactic properties. Candidates validated by a domain expert constitute a seed list for acquiring a partial ontology. © 2008 Springer-Verlag Berlin Heidelberg.