Description

Span break on possible superscript requires char bbox to have no overlap with previous span. The logic is too strict for documents where char bboxes have slight overlap.

# Character is likely a superscript
if all([
char["bbox"][1] < (span["bbox"][1] - span["bbox"].height * line_distance_threshold), # char top is above span
char["bbox"][3] < (span["bbox"].height * superscript_height_threshold) + span["bbox"][1], # char bottom is not full line height
char["bbox"][0] > span["bbox"][2], # char is to the right of the span

If a chars left edge[0] overlaps with the previous spans right edge[2], it fails the third condition.

Reproduction

Test file: https://doi.org/10.1177/1748006X17699145
Current behavior: detects 4 superscript spans.
Expected behavior: Should detect 35 superscript spans that fit existing filtering logic.

Results

Using the superscript chars bbox width as reference:
30 of the missed superscript chars have <5% of overlap with the last char of the previous span. 1 of the missed superscript have around 7% overlap.
Average overlap: 2.33% of character width

Suggestion

Allow overlap in

char["bbox"][0] > span["bbox"][2], # char is to the right of the span