2019/09/19 by Tzuf Paz-Argaman, Reut Tsarfaty, Paz-Argaman, Tzuf +1 · 1 citation
Social Sciences · Computer Science · #Geographic Information Systems Studies #Human Mobility and Location-Based Analysis #Data Management and Algorithms
paper · pdf · doi:10.48550/arxiv.1909.08970
Following navigation instructions in natural language requires a composition\nof language, action, and knowledge of the environment. Knowledge of the\nenvironment may be provided via visual sensors or as a symbolic world\nrepresentation referred to as a map. Here we introduce the Realistic Urban\nNavigation (RUN) task, aimed at interpreting navigation instructions based on a\nreal, dense, urban map. Using Amazon Mechanical Turk, we collected a dataset of\n2515 instructions aligned with actual routes over three regions of Manhattan.\nWe propose a strong baseline for the task and empirically investigate which\naspects of the neural architecture are important for the RUN success. Our\nresults empirically show that entity abstraction, attention over words and\nworlds, and a constantly updating world-state, significantly contribute to task\naccuracy.\n