Mapping Natural Language Instructions to Mobile UI Action Sequences

TL;DR

This paper introduces a new problem of translating natural language instructions into mobile UI actions, creates datasets including PIXELHELP, and proposes a Transformer-based model that achieves over 70% accuracy in predicting action sequences.

Contribution

It defines the task of grounding natural language to mobile UI actions, creates new datasets, and develops a Transformer model for accurate action sequence prediction.

Findings

Achieved 70.59% accuracy on PIXELHELP dataset.

Created three new datasets for natural language to UI action mapping.

Proposed a Transformer-based approach for grounding instructions to UI actions.

Abstract

We present a new problem: grounding natural language instructions to mobile user interface actions, and create three new datasets for it. For full task evaluation, we create PIXELHELP, a corpus that pairs English instructions with actions performed by people on a mobile UI emulator. To scale training, we decouple the language and action data by (a) annotating action phrase spans in HowTo instructions and (b) synthesizing grounded descriptions of actions for mobile user interfaces. We use a Transformer to extract action phrase tuples from long-range natural language instructions. A grounding Transformer then contextually represents UI objects using both their content and screen position and connects them to object descriptions. Given a starting screen and instruction, our model achieves 70.59% accuracy on predicting complete ground-truth action sequences in PIXELHELP.