Locating places described in natural language using human-machine dialogue.

Abstract

While humans communicate, several situations that require the description of the location of some landmark may arise. For instance, when reporting some occurrence to the authorities over the phone. The existence of models that are capable of interpreting these descriptions, locating an object in space, has the potential to allow the development of many applications that make use of this type of information such as robots that can assist emergency dispatch services, or even new interfaces for existing products, such as services based on autonomous vehicles and web mapping applications. However, natural language location descriptions are not always in a clean and simple format such as street name and number. In daily conversation, people often tend to reference points of interest, nearby landmarks and their relations to the location being described; Moreover, the usage of natural language introduces the possibility of error due to uncertainty, ambiguity and vagueness, natural aspects of communication that makes use of the human natural languages. The correct interpretation of natural language expressions, of spatial relations and landmarks, as well as the projection of the regions described by these relationships, represent great challenges to the development of geographic aware systems. This work presents a conceptual model that seeks to represent conversations that convey information about location. Capturing the crucial concepts of the mentioned domain, such as landmarks and spatial relations, the conceptual model may guide the construction of computational dialogue systems by working as a set of development guidelines and alerting for common pitfalls that can befall system designers. Being spatial relations some of the most important concepts in spatial communication, a study that soughtto better comprehend the usage of such relations in language by people was conducted. As a result of this investigation a set of algorithms that project the spatial extents described by the spatial relations that are most often used by people is proposed. These procedures take as input an identifier for a landmark used as reference and produce a polygon that represents the region described by the relation. Finally, a case study is presented where the conceptual model was used to support the development of a chatbot. Having as scope the city of Campina Grande at Paraíba, this chatbot makes use of the proposed algorithms to try and locatethrough dialogue, points being described in an urban scenario.