Automatic Acquisition of Paradigmatic Relations using Iterated Co-occurrences
Chris Biemann, Stefan Bordag, Uwe Quasthoff
Leipzig University Computer Science Institute, NLP Dept.
We introduce the notion of iterated co-occurrences, which can be obtained by performing the calculation of statistically significant co-occurrences not on sentence level, but on co-occurrence sets of previous calculations. The underlying mechanisms are explained in detail and we give reasons, why this iteration results in sets of semantically homogeneous words. These can be used for the automatic acquisition of paradigmatic relations in order to semi-automatically extend lexical-semantic word nets or thesauri, widening the acquisition bottleneck. A small evaluation for synset expansion for German language and some discussion conclude the work.
automatic lexical acquisition, paradigmatic relations, syntagmatic relations, co-occurrences, iterated co-occurrences, collocations