RecipeRef Corpus for Modeling Anaphora Resolution from the Procedural Text of Recipes

Current research on anaphora resolution mostly focuses on declarative text, e.g. journals, newswires, and dialogues. Procedural text, however, as one important type of text, contains rich anaphora phenomena but has only received limited attention. For procedural text comprehension, it is critical to resolving anaphora phenomena - resolution of anaphora in these texts is required to determine what action applies to which entity. To enrich the anaphora research in the procedural text, we investigate the anaphora phenomena in the procedural text of recipes. More specifically, we detailedly investigate the similarities and differences between recipes and chemical patents. We propose a generalized annotation guideline for the procedural text of recipes, adopting from chemical patents with domain adaptation. Based on the proposed guideline, we further create the RecipeRef corpus, in the hope that it will facilitate the research on anaphora resolution in procedural text.