CraftAssist Instruction Parsing: Semantic Parsing for a Minecraft Assistant
This work addresses the problem of robust instruction parsing for Minecraft assistants, but it is incremental as it focuses on dataset creation and baseline evaluation.
The authors tackled the problem of instruction-driven communication with an agent in Minecraft by creating a large-scale semantic parsing dataset of 35K human-generated instructions with annotations, and found that while this dataset enables training a usable parser, it presents generalization challenges.
We propose a large scale semantic parsing dataset focused on instruction-driven communication with an agent in Minecraft. We describe the data collection process which yields additional 35K human generated instructions with their semantic annotations. We report the performance of three baseline models and find that while a dataset of this size helps us train a usable instruction parser, it still poses interesting generalization challenges which we hope will help develop better and more robust models.