On multi-subjectivity in linguistic summarization of relational databases

Abstract

We focus on one of the most powerful computing methods for natural-language-driven representation of data, i.e. on Yager’s concept of a linguistic summary of a relational database (1982). In particular, we introduce an original extension of that concept: new forms of linguistic summaries. The new forms are named Multi-Subject linguistic summaries, because they are constructed to handle more than one set of subjects, represented by related sets of records/objects collected in a database, like ”cars, bicycles and motorbikes” (within vehicles), ”male and female” (within people), e.g. More boys than girls play football well. Thanks to that, the generated linguistic summaries – quasi-natural language sentences – are more interesting and human-oriented. Moreover, they can be applied together with the classic forms od summaries, to enrich naturality of comments/ descriptions generated. Apart from traditional interpretions linguistic summaries in termsof fuzzy logic, we also introduce some higher-order fuzzy logic methods, to extend possibilities of representing too complex or too ill-defined linguistic terms used in generated messages. The new methods are applied to a computer system that generates natural language description of numeric data, that makes them possible to be clearly presented to an end-user.