Mathematical equations in LaTeX are composed with tags that express formatting as opposed to structure. For conversion from LaTeX to other word-processing systems, the structure of each equation must be inferred. We show how a form of least cost parsing used with a very general and ambiguous grammar may be used to select an appropriate structure for a LaTeX equation. MathML provides another application for the same technology; it has two alternative tagging schemes - presentation tags to specify formatting and content tags to specify structure. While conversion from content tagging to presentation tagging is straightforward, the converse is not. Our implementation of least cost parsing is based on Earley’s algorithm.