Making Arabic PDF books accessible using gamification
Hend AlRouqi, Hend S. Al-Khalifa · 2014 · Proceedings of the 11th Web for All Conference (W4A) · doi:10.1145/2596695.2596712
Summary
This paper addresses the inaccessibility of Arabic PDF books for people with visual impairments, blindness, and dyslexia. Most online Arabic books are scanned images of printed originals, making them unreadable by screen readers and text-to-speech software. Arabic OCR technology remains far less mature than its counterparts for Latin scripts, suffering from the complexities of Arabic script — including right-to-left directionality, connected letter forms, diacritical marks, and ligatures that change letter shapes depending on font. The authors propose a novel approach that bypasses OCR entirely by combining crowdsourcing with gamification. Their prototype system takes the form of a mobile word-recall game: scanned book pages are segmented into individual word images on the server side, then presented briefly to players who must type the word they saw before time runs out. The system verifies answers using shape context feature extraction and matching — comparing the player's typed text (converted to an image) against the original word image, similar to techniques used in word spotting. Quality assurance mechanisms include player testing (inserting known words to detect unreliable players) and repetition (requiring multiple players to agree on a word before accepting it).
Key findings
The authors implemented two core components: word segmentation using projection profile techniques, and feature extraction and matching for answer verification. Testing the matching component across 188 word images revealed a key challenge: the average similarity measures for correct answers (0.0606) and answers with one wrong letter (0.0630) were very close, making it difficult to set a reliable threshold for accepting or rejecting player input. This proximity is attributed to Arabic script characteristics — diacritical signs and ligatures can make visually similar letters produce nearly identical shape features. Completely wrong answers (0.0766) were more distinguishable. The system represents an early-stage prototype; the game itself had not yet been tested with actual players, so engagement and feasibility of the gamification approach remain unvalidated. The authors note that the word segmentation step, while imperfect, can be manually verified offline before words enter the game database.
Relevance
This paper highlights a significant gap in digital accessibility: the vast majority of Arabic-language books remain inaccessible to people with print disabilities because the technical infrastructure for digitisation — particularly OCR — lags behind what exists for Latin scripts. The gamification approach is creative, drawing on the success of human computation games like reCAPTCHA and PHETCH to harness human recognition skills at scale. While the prototype is early-stage, the underlying problem remains pressing — Arabic is spoken by over 400 million people, yet accessible digital content in Arabic is scarce. The work also demonstrates how accessibility challenges intersect with language and script diversity, reminding practitioners that solutions developed for English-language contexts often do not transfer to other writing systems without significant adaptation.
Tags: document accessibility · gamification · crowdsourcing · OCR · Arabic accessibility · print disability · multilingual accessibility