Qualitative Analysis of Low-Level Logical Structures

This paper presents a qualitative approach to logical structure recognition of library references. The system is driven by a generic model of a reference class and by an OCR flow, given in SGML format, that include ASCII code of the characters and information about the typographic style and the lexical affiliation of words. The approach used is based on hypotheses production and verification about the existence of sub-field limits in the reference area. At each step of the analysis, the generated hypotheses are sorted on the basis of their confidence scores and the most likely hypothesis is analyzed. The result is a structured flow containing, in UNIMARC format, the list of different sub-fields recognized, accompanied with their confidence score.