Transcript parsing

[wfdd]

• Use poppler's pdftohtml -xml to convert PDFs into XML documents
• Depending on the horizonal and vertical spacing between arbitrary-length text
  objects, which are arbitrarily strewn on the page, figure out whether they
  are: a continuation of the same word or paragraph; or otherwise part of a
  table
• Intelligently collapse contiguous blank text objects on page breaks
• Construct syntax tree to assign parts of transcript to classes?
• Parse into Akoma Ntoso