Reading Order
A key problem in document parsing that concerns arranging text blocks in the correct order according to human reading habits. It is critical to the input quality of downstream NLP tasks.
Core Facts
Timeline (last 90 days)
Reading order errors are difficult to correct automatically via language models and need to be accurately modeled at the parsing stage.
The text stream in document formats such as PDF is not always stored in visual order; multi-column layouts, mixed text and image arrangements, and interleaved footnotes and headers can cause the underlying text stream to be misaligned with visual order.
Reading order and table parsing are considered core metrics for measuring document comprehension capability.
All Facts (3)
The text stream in document formats such as PDF is not always stored in visual order; multi-column layouts, mixed text and image arrangements, and interleaved footnotes and headers can cause the underlying text stream to be misaligned with visual order.
75%UnverifiedReading order errors are difficult to correct automatically via language models and need to be accurately modeled at the parsing stage.
50%UnverifiedReading order and table parsing are considered core metrics for measuring document comprehension capability.
50%