The Serialization of Heterogeneous Documents

Tasks involving the analysis of natural language are typically conducted on a corpus or corpora of plain text. However, it is rare that a document is unstructured and freeform in its entirety. Documents such as corporate disclosures, medical journals and other knowledge rich archive contain structured and loosely-structured information that can be used in a variety of important text mining tasks. In this paper we propose a syntactical preprocessing architecture to serialize presentationoriented documents to a machine readable format that aspires to preserve the document structure, contents and metadata. We introduce a hybrid pipeline architecture, discussing the various processes and the future research direction that could potentially lead to a holistic representation of heterogeneous documents.