Incorporating Task-specific Concept Knowledge into Script Learning

In this paper, we present Tetris, a new task of Goal-Oriented Script Completion. Unlike previous work, it considers a more realistic and more general setting, where the input includes not only the goal but also additional user context, including preferences and history. To address the problem using a knowledge-based approach, we introduce Task Concept Graph, an automatically constructed knowledge base from instructional websites. Different from Commonsense Knowledge Base like ConceptNet, the task concept graph schema introduces various types of noun phrases-based nodes specifically for accomplishing a task. To integrate such graphs into script learning, we devise two methods that acquire concepts from the knowledge base as prompts to downstream script completion. On our WikiHow-based dataset, we find that incorporating concepts from the Task Concept Graph consistently improves performance, demonstrating the benefit of Task Concept Graph for this task. Furthermore, the model with gold-standard concepts as prompt outperforms the baseline significantly, further confirming the need for task-specific knowledge in goal-oriented script completion. The dataset, repository, models, and demo will be publicly available to facilitate further research on this new task.