Measuring Soft Line Breaks in Page Description Languages

L. Chen and M. Alvarez

Abstract

We measure how often visual line wraps are written to the clipboard as hard breaks, and show that a geometric pass recovers paragraph text on untagged office exports.

Page description languages record ink, not discourse. Tagged PDF exists, yet printer drivers still flatten structure.

Each glyph is clustered by baseline, then columns are split at the widest vertical gutter. Headers that repeat across pages are dropped.
