Natural Language Landmark Navigation Instructions (NLLNI)

Turn-by-Turn navigation instructions are based on distances and named streets, whereas navigation instructions naturally used by humans are centered around landmarks like stores, restaurants, churches, parks and other visible structures.

We consider the task of creating such natural language landmark navigation instructions automatically by an end-to-end model that is trained on thousands of human written instruction examples. Our model gets a representation of the route and its surrounding points of interest and automatically generates navigation instructions for it. It does so without any predefined rules for selecting important waypoints or landmark salient scores.


Please cite the following paper when using the map2seq data:
@inproceedings {schumann-riezler-2021-map2seq,
 title = "Generating Landmark Navigation Instructions from Maps as a Graph-to-Text Problem",
 author = "Raphael Schumann and Stefan Riezler",
 year = "2021",
 publisher = "Association for Computational Linguistics"