Learning Biological Processes with Global Constraints

Proceedings of the 2013 Conference on Empirical Methods in Natural Language Processing, pages 1710–1720, Seattle, Washington, USA, 18-21 October 2013. c⃝2013 Association for Computational Linguistics Learning Biological Processes with Global Constraints Aju Thalappillil Scaria∗, Jonathan Berant∗, Mengqiu Wang and Christopher D. Manning Stanford University, Stanford Justin Lewis and Brittany Harding University of Washington, Seattle Peter Clark Allen Institute for Artificial Intelligence, Seattle Abstract Biological processes are complex phenom- ena involving a series of events that are re- lated to one another through various relation- ships. Systems that can understand and rea- son over biological processes would dramat- ically improve the performance of semantic applications involving inference such as ques- tion answering (QA) – specifically “How?” and “Why?” questions. In this paper, we present the task of process extraction, in which events within a process and the rela- tions between the events are automatically ex- tracted from text. We represent processes by graphs whose edges describe a set of temporal, causal and co-reference event-event relations, and characterize the structural properties of these graphs (e.g., the graphs are connected). Then, we present a method for extracting rela- tions between the events, which exploits these structural properties by performing joint in- ference over the set of extracted relations. On a novel dataset containing 148 descrip- tions of biological processes (released with this paper), we show significant improvement comparing to baselines that disregard process structure. 1 Introduction A process is defined as a series of inter-related events that involve multiple entities and lead to an end result. Product manufacturing, economical de- velopments, and various phenomena in life and so- cial sciences can all be viewed as types of processes. Processes are complicated objects; consider for ex- ample the biological process of ATP synthesis de- scribed in Figure 1. This process involves 12 en- tities and 8 events. Additionally, it describes rela- tions between events and entities, and the relation- ship between events (e.g., the second occurrence of the event ‘enter’, causes the event ‘changing’). ∗Both authors equally contributed to the paper Automatically extracting the structure of pro- cesses from text is crucial for applications that re- quire reasoning, such as non-factoid QA. For in- stance, answering a question on ATP synthesis, such as “How do H+ ions contribute to the production of ATP?” requires a structure that links H+ ions (Figure 1, sentence 1) to ATP (Figure 1, sentence 4) through a sequence of intermediate events. Such “How?” questions are common on FAQ websites (Surdeanu et al., 2011), which further supports the importance of process extraction. Process extraction is related to two recent lines of work in Information Extraction – event extrac- tion and timeline construction. Traditional event ex- traction focuses on identifying a closed set of events within a single sentence. For example, the BioNLP 2009 and 2011 shared tasks (Kim et al., 2009; Kim et al., 2011) consider nine events types related to proteins. In practice, events are currently almost al- ways extracted from a single sentence. Process ex- traction, on the …

Content truncated to 500 words