Process-Level Representation of Scientific Protocols with a Text-Based Game Annotation Interface

Abstract

We develop Process Execution Graphs (PEG), an executable document-level representation of real-world wet lab biochemistry protocols, addressing challenges such as cross-sentence relations, long-range coreference, grounding, and implicit arguments. We built a corpus of complex lab protocols with a novel interactive simulator built upon a text-based game engine that keeps track of entity traits and semantic constraints during annotation, yielding high quality annotated PEGs. Our framework presents several directions for future work, including the modelling of challenging long range dependencies, application of text-based games for real-world procedural text understanding, and extending simulation-based annotation to new domains.

Publication
WordPlay Workshop @ NeurIPS2020