Extracting Definitions of Mathematical Expressions in Scientific Papers

Abstract

Natural language definitions of mathematical expressions often play an essential role to understand mathematical content of scientific papers. However, there exists no available annotated data that contains relations between mathematical expressions and their definitions. Such annotated data enables us to apply and compare different approaches for this relation extraction task. In this paper, we first introduce our guideline for annotating definitions of mathematical expressions in scientific papers. This guideline describes specific types of mathematical expression description that should be annotated as definitions and how to annotate them. Based on the guideline, we manually annotated 14 papers from International Journal on Document Analysis and Recognition (IJDAR) from Springer. We also developed and investigated the performance of pattern matching and machine learning based methods in comparison with an existing naive practice that is based on the head noun of preceding text.