Talk the Walk: Navigating New York City through Grounded Dialogue

TL;DR

This paper presents 'Talk The Walk,' a large-scale grounded dialogue dataset where two agents communicate via natural language to navigate New York City, introducing a novel grounding mechanism and establishing baseline results.

Contribution

It introduces a new dataset and task for grounded dialogue navigation, along with a novel grounding mechanism called MASC and baseline results.

Findings

MASC improves grounding accuracy in navigation tasks

The dataset enables research on natural language grounded in real-world environments

Baseline models demonstrate the task's complexity and potential for future work

Abstract

We introduce "Talk The Walk", the first large-scale dialogue dataset grounded in action and perception. The task involves two agents (a "guide" and a "tourist") that communicate via natural language in order to achieve a common goal: having the tourist navigate to a given target location. The task and dataset, which are described in detail, are challenging and their full solution is an open problem that we pose to the community. We (i) focus on the task of tourist localization and develop the novel Masked Attention for Spatial Convolutions (MASC) mechanism that allows for grounding tourist utterances into the guide's map, (ii) show it yields significant improvements for both emergent and natural language communication, and (iii) using this method, we establish non-trivial baselines on the full task.