Segmenting and Tagging Structured Content

Bilingual dictionaries hold great potential as a source of lexical resources for training automated systems for optical character recognition, machine translation, and cross-language information retrieval. More importantly, they represent a class of document that have a great deal of structure, and this structure is not fundamentally spatial. Structure is provided by the authors (or publishers) who use of different fonts, font style, spacing and special symbols to implicitly convey information on the structure of the content. Our system is divided into three phases – Dictionary Segmentation, Entry Tagging and Generation. In segmentation, pages are divided into logical entries based on structural features learned from selected examples. The extracted entries are associated with functional labels and passed to a tagging module that associates linguistic labels with each word or phrase in the entry. The output of the system is a structure that represents the entries of the dictionary. In this document, we discuss the fundamental image processing approach that lets us extract this structure. Details of how we perform content tagging is mentioned briefly and references to related publications are provided.