Comprehensiveness and interpretability of linguistic data summaries: A natural language focused perspective

Abstract

We consider the important problem of comprehensiveness of linguistic data summaries equated with linguistically quantified propositions in Zadeh's sense. Motivated by Michalski's [29] seminal approach to the comprehensiveness of data mining and machine learning results, with a clear emphasis on natural language, we advocate the use of linguistic summaries which provide a new quality and an exceptional human consistency and comprehensiveness. Extending our previous works, we first relate our approach to some related results on the interpretability and comprehensiveness of fuzzy rule bases, both with respect to structural and semantical complexity. We show the use of a fuzzy querying interface as not only an approach that is effective and efficient but which provides an exceptional comprehensiveness through its highly human consistent HCI (human computer interface). We emphasize a psychological and cognitive aspect of comprehensibility and interpretability analyses. We advocate the use of human consistent methods based on natural language. We indicate a possibility of using quantitative evaluations. We illustrate our analysis by two examples related to the linguistic summarization of both static and dynamic data in the area of innovation and Web log analyses, and justify the results obtained by domain experts positive assessments. In general, we propose a synergistic combination of formal and natural language based methods to solve the inherently human specific problem of comprehensiveness and interpretability.

Abstraction and Summarization of Meaning in Natural Language Processing

“…As remarked in [4], the interpretability of semantic meaning of sentences has not been much studied by fuzzy rule-based systems. The understanding of a semantic meaning of a sentence in the text depends on its linguistic expression, and the linguistic variables attached to the attributes and to the fuzzy quantifier.…”

Oppositions in Fuzzy Linguistic Summaries

“…Fuzzy Linguistic Summaries (FLS) [1], [2] can be defined as texts made of several sentences describing distinct characteristics of a given dataset. Their interpretability [3], [4] is a central issue, related more specifically to the one of consistency, i.e. absence of contradiction.…”

Interpretability of fuzzy linguistic summaries

“…As remarked in [8], the interpretability of fuzzy linguistic summaries has not been much studied as compared to fuzzy rule-based systems, probably for two reasons. First of all, less attention has been brought to them than fuzzy rule-based though they have been introduced in 1982 [9] and mainly investigated since 2001 [10].…”