Arabic Entity Linking Surface-Form Coverage: benchmark, anchor dictionary, surface-form pool and analysis code

Resources accompanying the study How Far Can String Matching Take Arabic Entity Linking? Layered Surface-Form Coverage in Wikidata and Arabic Wikipedia. Contents: a 200,000-mention Arabic entity-linking evaluation set built from Arabic Wikipedia anchors with a held-out split; an anchor dictionary of over two million surface-form/entity pairs; a combined surface-form pool with per-row provenance covering Wikidata labels and aliases, ParaNames, Arabic Wikipedia titles, redirects and anchor text; snapshots of the underlying Wikidata and Wikipedia tables; and the dialect-label filtering data and pipeline. See README.md for file schemas, licensing and known limitations. PROVENANCE.json records the snapshot dates without which the figures cannot be reproduced. Version 1.1.0 adds the analysis code (code.zip, MIT licence) and the snapshot dates in PROVENANCE.json; the data files are unchanged.