WikiTexVC: MediaWiki's native LaTeX to MathML converter for Wikipedia (2401.16786v1)

Abstract: MediaWiki and Wikipedia authors usually use LaTeX to define mathematical formulas in the wiki text markup. In the Wikimedia ecosystem, these formulas were processed by a long cascade of web services and finally delivered to users' browsers in rendered form for visually readable representation as SVG. With the latest developments of supporting MathML Core in Chromium-based browsers, MathML continues its path to be a de facto standard markup language for mathematical notation in the web. Conveying formulas in MathML enables semantic annotation and machine readability for extended interpretation of mathematical content, in example for accessibility technologies. With this work, we present WikiTexVC, a novel method for validating LaTeX formulas from wiki texts and converting them to MathML, which is directly integrated into MediaWiki. This mitigates the shortcomings of previously used rendering methods in MediaWiki in terms of robustness, maintainability and performance. In addition, there is no need for a multitude of web services running in the background, but processing takes place directly within MediaWiki instances. We validated this method with an extended dataset of over 300k formulas which have been incorporated as automated tests to the MediaWiki continuous integration instances. Furthermore, we conducted an evaluation with 423 formulas, comparing the tree edit distance for produced parse trees to other MathML renderers. Our method has been made available Open Source and can be used on German Wikipedia and is delivered with recent MediaWiki versions. As a practical example of enabling semantic annotations within our method, we present a new macro that adds content to formula disambiguation to facilitate accessibility for visually impaired people.