Learning to Decompose and Organize Complex Tasks

Proceedings of the 2021 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, pages 2726–2735 June 6–11, 2021. ©2021 Association for Computational Linguistics 2726 Learning to Decompose and Organize Complex Tasks Yi Zhang†∗, Sujay Kumar Jauhar§, Julia Kiseleva§, Ryen W. White§ and Dan Roth† †University of Pennsylvania, Philadelphia, PA, USA §Microsoft Research, Redmond, WA, USA †{yizhang5, danroth}@cis.upenn.edu, §{sjauhar, jukisele, ryenw}@microsoft.com Abstract People rely on digital task management tools, such as email or to-do apps, to manage their tasks. Some of these tasks are large and com- plex, leading to action paralysis and feelings of being overwhelmed on the part of the user. The micro-productivity literature has shown that such tasks could benefit from being de- composed and organized, in order to reduce user cognitive load. Thus in this paper, we propose a novel end-to-end pipeline that con- sumes a complex task and induces a depen- dency graph from unstructured text to repre- sent sub-tasks and their relationships. Our so- lution first finds nodes for sub-tasks from mul- tiple ‘how-to’ articles on the web by injecting a neural text generator with three key desiderata – relevance, abstraction, and consensus. Then we resolve and infer edges between these sub- task nodes by learning task dependency rela- tions. We collect a new dataset of complex tasks with their sub-task graph to develop and evaluate our solutions. Both components of our graph induction solution are evaluated in experiments, demonstrating that our models outperform a state-of-the-art text generator sig- nificantly. Our generalizable and scalable end- to-end solution has important implications for boosting user productivity and assisting with digital task management. 1 Introduction People today increasingly rely on digital modalities and applications to organize, track and complete tasks from their work and life. They depend on email to structure their communications as a way of tracking pending tasks (Bellotti et al., 2003), is- sue commands to their digital assistants 1 for timely task reminders (Brewer et al., 2017), and use task management applications (Bellotti et al., 2004) 2 ∗Most of this work was done while the first author was an intern at Microsoft Research. 1Siri, Cortana, Google Assistant etc. 2Google Keep, Todoist, Microsoft To Do etc. Figure 1: The subtask graph of “plan a birthday party”, where nodes are subtasks, and the directed edges repre- sent the dependencies between subtasks. to organize grocery runs, thesis writing and every- thing in between. In this work, we focus on tasks that are com- plex (Hassan Awadallah et al., 2014), and – which research in micro-productivity has shown (Kirsh, 2000; Teevan et al., 2016a) – may benefit from thoughtful organization. For example, consider Figure 1, where we show how the complex task “plan a birthday party”3 can be broken down into more manageable pieces and structured by mutual temporal dependencies, in order to create an action- able plan that is simpler and more effective. In this paper we propose to help automate gener- ating such actionable plans in order to reduce cog- nitive …

Content truncated to 500 words