Writing was shaped by the human visual system. Over five thousand years, scripts and then page layouts kept the shapes and arrangements that human eyes pick out easily and dropped the rest. A PDF is the latest form of that. Software that reads documents has to read the shapes and arrangements, not just the characters.
Writing fits the eye
The visual system is hundreds of millions of years old; writing is recent. Evolution had no time to build a reading organ, so reading had to use what was already there: the parts of the visual cortex that recognise objects by their edges and the junctions between edges. Stanislas Dehaene calls this neuronal recycling.
Mark Changizi measured this. Across about a hundred writing systems, the kinds of line junction in letters (L, T, X, Y etc.) occur in roughly the proportions they occur in photographs of natural scenes. Letters look like pieces of objects because pieces of objects are what the brain is built to see. The shapes were not chosen because they are easy to draw; they were chosen because they are easy to see.
The pressure is still running. Olivier Morin compared over a hundred scripts and found horizontal and vertical strokes far more common than chance, and that scripts become more so over time. A new script invented in Liberia in the 1830s went the same way in a century.
The page went the same way
A page is made of the same lines at a larger scale. Columns, margins, rules, indents, a heading in a bigger font size. Each one tells the eye where a group starts and ends, what order to read in, and what matters. Nobody worked this out from a theory. Scribes and printers kept what readers could follow and dropped what they could not. Grouping, reading order and emphasis are carried by position and shape, not by the characters.
A PDF fixes the layout because the layout carries that information. Convert a PDF to a text stream and the characters survive, but the grouping, the reading order and the emphasis are gone. That is why PDF beats the formats that reflow text. It keeps the page in the form human readers need.
What this means for reading by machine
Software that strips a page down to a stream of character codes discards the spatial structure built for the eye. To read a page the way a person does, software must measure the lines, the gaps, the alignments, and the type sizes first, and recover the groups and reading order from the geometry.
Layout is geometry on a page: coordinates, widths, alignments, type sizes. It can be measured exactly, on your own machine, and the measurement can be checked. So the first step in reading a PDF should be measurement, not a question put to a cloud model. Interpretation comes after, and starts from numbers that are known to be right.
What we are doing about it
- The method: a local-first architecture that keeps your documents on your own machines. The Local-First Appliance Model
References
- Changizi, M. A., Zhang, Q., Ye, H. & Shimojo, S. (2006). The structures of letters and symbols throughout human history are selected to match those found in objects in natural scenes. American Naturalist 167(5), E117–E139.
- Dehaene, S. & Cohen, L. (2007). Cultural recycling of cortical maps. Neuron 56, 384–398.
- Dehaene, S. (2009). Reading in the Brain. Viking.
- Morin, O. (2018). Spontaneous emergence of legibility in writing systems: the case of orientation anisotropy. Cognitive Science 42, 664–677.
- Kelly, P., Winters, J., Miton, H. & Morin, O. (2021). The predictable evolution of letter shapes: an emergent script of West Africa recapitulates historical change in writing systems. Current Anthropology 62(6).
peter@docs2info.com
