latexml is a program, written in Perl, that attempts to faithfully
mimic TeX's behavior, but produces XML instead of dvi. The document
model of the target XML makes explicit the model implied by LaTeX.
The processing and model are both extensible; you can define the
mapping between TeX constructs and the XML fragments to be created.
A postprocessor, latexmlpost converts this XML into other formats
such as HTML or XHTML, with options to convert the math into MathML
(currently only presentation) or images.
