English Recipe Flow Graph Corpus

Proceedings of the 12th Conference on Language Resources and Evaluation (LREC 2020), pages 5187–5194 Marseille, 11–16 May 2020 c⃝ European Language Resources Association (ELRA), licensed under CC-BY-NC 5187 English Recipe Flow Graph Corpus Yoko Yamakata∗, Shinsuke Mori†, John Carroll‡ ∗The University of Tokyo, 7-3-1 Hongo, Bunkyo-ku, Tokyo, Japan yamakata@mi.u-tokyo.ac.jp †Kyoto University, Yoshida-Honmachi, Sakyo-ku, Kyoto, Japan forest@i.kyoto-u.ac.jp ‡University of Sussex, Falmer, Brighton, UK J.A.Carroll@sussex.ac.uk Abstract We present an annotated corpus of English cooking recipe procedures, and describe and evaluate computational methods for learning these annotations. The corpus consists of 300 recipes written by members of the public, which we have annotated with domain-specific linguistic and semantic structure. Each recipe is annotated with (1) ‘recipe named entities’ (r-NEs) specific to the recipe domain, and (2) a flow graph representing in detail the sequencing of steps, and interactions between cooking tools, food ingredients and the products of intermediate steps. For these two kinds of annotations, inter-annotator agreement ranges from 82.3 to 90.5 F1, indicating that our annotation scheme is appropriate and consistent. We experiment with producing these annotations automatically. For r-NE tagging we train a deep neural network NER tool; to compute flow graphs we train a dependency-style parsing procedure which we apply to the entire sequence of r-NEs in a recipe. In evaluations, our systems achieve 71.1 to 87.5 F1, demonstrating that our annotation scheme is learnable. Keywords: cooking recipe corpus, English recipes, recipe named entity, recipe flow graph, procedural text annotation 1. Introduction Procedural text describes in an objective way how to carry out some task. Although this kind of text is widespread, it is found particularly in instruction manuals and cookbooks. The information in procedural text could be used to sup- port a wide range of computer-based applications involv- ing intelligent search, reasoning, and non-textual forms of human-computer interaction; but first the text would need to be transformed into a more computationally tractable representation. At minimum, this representation would rep- resent domain entities, actions, and relationships between them. A flow graph is one kind of computationally tractable rep- resentation; it contains steps linking actions with entities, and specifies the obligatory sequencing relationships be- tween steps. A flow graph abstracts away from differences in surface linguistic expression that are irrelevant to cor- rectly performing a procedure. For example, when cooking a dish, the instructions add diced carrots to the pot and dice carrots and add to the pot are different textually, but would correspond to the same flow graph since they involve the same actions carried out in the same order. Our research focuses on the cooking domain, and specifi- cally the procedural part of cooking recipes. Our general approach and some of our computational techniques are in- spired by recent research on recipes written in Japanese (Mori et al., 2014). We extend that work by adapting the annotation to English, annotating more data, investigating the consistency of annotation, and improving accuracy of processing. In this paper, we describe how we annotated a corpus of 300 English recipes, which were contributed …

Content truncated to 500 words