QSRL : A Semantic Role-Labeling Schema for Quantitative Facts

Financial text is replete with quantitative information about company, industry, and economy-level performance. Until now however, work on financial narrative processing has overlooked this information in favor of softer forms of meaning like textual sentiment. In this paper, we examine such language from two sources—newswire and publicly available quarterly reports—to define an annotation schema for quantitative facts in text to be used in future information extraction (IE) work. The Quantitative Semantic Role Labels (QSRL) representation takes a situationist perspective on quantitative facts, describing quantities not only in terms of hard numerical values, but also the context in which they take on those values. Unlike other semantic role-labeling frameworks however, it is specifically designed with quantitative language in mind, and hence is a much simpler representation. We conclude with a description of some of the challenges we face in quantitative information extraction, as highlighted by the data we consider throughout the paper.