Automatic Approach to Understanding Mathematical Expressions Using MathML Parallel Markup Corpora

Abstract

This paper explores the use of MathML Pallel Markup Corpora for mathematical expression understanding, the task of which is formulated as a translation from Presentation to Content MathML Markups in our approach. In contrast to existing researches that mainly relied on manually encoded transformation rules, we adopt a Statistical-Machine-Translation-based method to automatically extract translation rules from parallel markup corpora. Our study shows that the structural features embedded in the MathML tree can be effectively exploited in the sub-tree alignment and the translation rules extracted from the alignment give boost to the translation system. Experimental results on the Wolfram Function Site show that our approach achieves an improvement against a rule-based system.