Easy measures for evaluating non-English corpora for language engineering. Some lessons from Arabic and Bengali

Sarkar, A.; De Roeck, A. and Garthwaite, P. (2004). Easy measures for evaluating non-English corpora for language engineering. Some lessons from Arabic and Bengali. Technical Report 2004/05; Department of Computing, The Open University.

DOI: https://doi.org/10.21954/ou.ro.00016027

Abstract

In this paper we develop a framework for fast profiling and quality verification of datasets for language engineering and information retrieval research. The profiling steps consist of an initial tokenization of the corpus to produce a frequency list from which some basic statistics are derived. Manual sampling is carried out to detect obvious discrepancies. Two diagnostic tests are performed to check for sparseness related measures. The behaviour of the function words is traced to gauge homogeneity of their distribution in documents.

Viewing alternatives

Metrics

Item Actions

Export

You can export this page using these formats Digital Object Identifier - DOI