Generalization in Instruction Following Systems

Proceedings of the 2021 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, pages 976–981 June 6–11, 2021. ©2021 Association for Computational Linguistics 976 Generalization in Instruction Following Systems Soham Dan and Michael Zhou and Dan Roth {sohamdan,mizho,danroth}@seas.upenn.edu University of Pennsylvania Abstract Understanding and executing natural language instructions in a grounded domain is one of the hallmarks of artificial intelligence. In this pa- per, we focus on instruction understanding in the blocks world domain and investigate the language understanding abilities of two top- performing systems for the task. We aim to un- derstand if the test performance of these mod- els indicates an understanding of the spatial do- main and of the natural language instructions relative to it, or whether they merely over-fit spurious signals in the dataset. We formulate a set of expectations one might have from an in- struction following model and concretely char- acterize the different dimensions of robustness such a model should possess. Despite decent test performance, we find that state-of-the-art models fall short of these expectations and are extremely brittle. We then propose a learning strategy that involves data augmentation and show through extensive experiments that the proposed learning strategy yields models that are competitive on the original test set while satisfying our expectations much better.1. 1 Introduction Building agents that can understand and execute natural language instructions in a grounded en- vironment is a hallmark of artificial intelligence (Winograd, 1972). There is wide applicability of this technology in navigation (Chen et al., 2019; Tellex et al., 2011; Chen and Mooney, 2011), col- laborative building (Narayan-Chen et al., 2019), and several others areas (Li et al., 2020b; Brana- van et al., 2009). The key challenge underlying these and many other applications is the need to understand the natural language instruction (to the extent that it is possible) and ground relevant parts of it in the environment. While the use of deep networks has led to significant progress on several 1Our code is publicly available at: http://cogcomp.org/page/publication_view/936 Figure 1: Task: Given a configuration of blocks and an instruction, predict the source and target location. benchmarks (Abiodun et al., 2018) an investigation into the instruction understanding capabilities of such systems remains lacking. We do not know the extent to which these models truly understand the spatial relations in the environment, nor their robustness to variability in the environment or in the instructions. This understanding is also impor- tant from the viewpoint of safety critical applica- tions , where robustness to variability is essential. While robustness to input perturbations at test-time has been studied in computer vision (Goodfellow et al., 2014) and in certain natural language tasks (Alzantot et al., 2018; Wallace et al., 2019; Shah et al., 2020), it remains relatively elusive in the in- struction following task in a grounded environment. This can be attributed to the difficulty in charac- terizing the different expectations of robustness in this setting, due to the multiple channels of input involved, which semantically constrain …

Content truncated to 500 words