The Cognitive Aspect of Place Properties

Abstract

GIScience 2016 Short Paper Proceedings The Cognitive Aspect of Place Properties Maria Vasardani, Martin Tomko, Stephan Winter The University of Melbourne, Parkville, 3010, VIC, Australia Email: {mvasardani, tomkom, winter}@unimelb.edu.au Abstract The need to computationally handle the cognitively grounded concept of place is fundamental for spatial human-computer interaction. However, there is thus far no consensus about a formal definition of place. In this paper, we explore the feasibility of a constructor of an abstract data type place by exploring a cognitively supported set of properties of place. We study the applicability of Alexander’s 15 structural properties of a whole to inform a place property parser of natural language place descriptions. 1. Introduction The concept of place is grounded in common sense and evoked in a variety of contexts. It is fundamental to spatial human communication and increasingly to spatial human-computer interaction. The concept of place becomes central to Geographic Information Science when location information about emergencies needs to be extracted from witnesses’ descriptions near real-time. Understanding the cognitive aspects behind natural language (NL) place descriptions is an essential first step for formalising the concept of place, so that that it can be used in spatial reasoning and decision making. Past attempts towards a broadly accepted definition of place have not been successful (see Cresswell, 2014). Vasardani and Winter (2015) argued that rather than providing precise but variable definitions of place according to application domain and context, it is possible to identify places through a set of properties encoding the concept. As a starting point, they suggested Alexander’s (2002) 15 structural properties that characterise a whole and examined how they correspond to properties of the various applications of place as studied throughout GIScience. In this work, we set out to explore whether a sub- or superset of these properties is cognitively supported, with the hope that this set can then be used by a place constructor—a generator of computational representations of place instances, operating as a function of place properties (i.e., attributes). We use textual place descriptions as a source of these properties. 2. Relevant work Salient locations that stand out from the ground become places, thus instances of objects, when applying Kuhn's (2012) terminology of core concepts of spatial information. As such, place instances have identity, exist in space, and exhibit spatial, temporal and thematic properties. Arguably, places exhibit a subset or superset of the properties described in (Vasardani and Winter, 2015). Here, we examine which of these 15 place wholeness properties should be part of a cognitively grounded place constructor. At a basic level, place constructors from text can take the form of parsers. For example, when parsing placenames from geotagged social media text, spatial clusters of placenames form candidate footprints, indicating a consensus about the existence and extent of places. These places with no exact boundaries, but rather a fuzzy membership function (Hollenstein and Purves, 2010; Pasley, 2008). Similarly, when extracting triplets using NL parsing methods (Khan et al., 2013), the parser constructs a single place