Quantifier Selection for Linguistic Data Summarization

Fuzzy quantifiers like "about sixty percent" are useful tools for expressing linguistic summaries. But, how can we determine the quantifier which best describes the given data? The quality indicators proposed for quantifier selection still make a rather heuristic impression. The paper therefore investigates a more principled way of controlling quantifier selection: a quantifier should be selected for summarization only when it is used in its prototypical sense. We capture this pragmatic issue of appropriate use by defining an associated pragma quantifier which expresses the paradigmatic cases best described by the considered quantifier. The quantifier selection will be based on an appropriateness score of the summary given by the degree of truth of the pragma quantifier. We further show that pragma quantifiers are typically neither absolute nor proportional, and thus demand generalized models of fuzzy quantification and new implementation techniques.