ALFADEL: A lexicon-aware environment for correcting Arabic OCR/HTR and editing historical orthography
Year: 2026
Type: software
Abstract: ALFADEL is a local, lexicon-aware environment for assisted correction of Arabic OCR/HTR and the editorial treatment of historical Arabic orthography. It distinguishes probable recognition errors from legitimate historical spellings and standard forms, and provides separate standardization and preserve-historical-forms workflows with explicit human editorial control. Named after the eleventh-century Arabic Christian scholar and translator **ʿAbd Allāh ibn al-Faḍl (عبد الله ابن الفضل)**, ALFADEL was developed to distinguish between three phenomena that conventional spelling correction can easily conflate: - OCR/HTR recognition errors;- legitimate historical spellings;- standard Arabic forms. The application combines lexical lookup, Arabic morphological analysis, clitic and pronominal-suffix processing, historical-orthographic rules, controlled textual evidence, and OCR/HTR-sensitive fuzzy matching. All proposed changes remain under explicit human editorial control. ALFADEL provides two distinct workflows: Correct and standardize the text Historical spellings and probable OCR/HTR errors are identified and canonical standard readings are proposed. For example: نبتدي → نبتدئ The proposed change is never applied automatically. Correct while preserving historical forms Historical spellings already present in the text are recognized and preserved. Standard counterparts may be displayed for reference without requiring normalization. Standard forms for which ALFADEL knows a licensed historical spelling are presented as explicit editorial choices. For example: صناعة → صناعه In this mode, ALFADEL distinguishes visually between: - historical — an already historical spelling that is preserved;- standard → historical option — a standard form for which a historical alternative is available;- correction — a probable OCR/HTR error requiring editorial attention;- unresolved — a form for which no sufficiently reliable analysis is available. Development ALFADEL began as an attempt to reuse existing Arabic lexical resources as a validation layer for OCR/HTR correction. Testing on historical Arabic texts demonstrated that dictionary lookup alone was insufficient. The correction engine was therefore progressively expanded to distinguish historical orthography from recognition error and to account for Arabic morphology, clitics, attached pronouns, weak-final forms, hamza-seat alternations, and ambiguous segmentations. Development proceeded iteratively through real-text error analysis. Recurrent false positives were converted whenever possible into general linguistic constraints. Examples that informed this process include: - نفسه = نفس + ه, rather than incorrectly proposing نفسة;- وتلميذه = و + تلميذ + ه, rather than وتلميذة;- سوالهم → سؤالهم, preserving the attached pronoun while restoring the hamza;- ببقايه → ببقائه, accounting for hamza-seat behavior before a suffix;- فانية, generated morphologically from the weak-final adjective فانٍ;- العطايا, recognized from attested evidence despite the absence of the plural form in the corresponding dictionary entry. More generally, missing dictionary forms were supplemented only when supported by attested evidence, rather than inferred automatically. A guiding principle that emerged from this development is that morphological analysis should precede historical normalization, and historical normalization should precede fuzzy OCR/HTR correction. Resource provenance and copyright The lexical resources used during development and testing were developed within the AAW project. These dictionaries and lexica are copyright Don Davide Righi and are therefore not included in this public release. The controlled corpus used during development derives from the e-Cheikho project, developed between CEDRAC and GREgORI, and is likewise not distributed. The repository preserves the expected resource structure using empty placeholders only.
Author: Ibrahim, Habib
Institution: Goethe University Frankfurt
Language: English
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W7204029499
