SpaceRef: A corpus of street-level geographic descriptions

SPACEREF: A Corpus of Street-Level Geographic Descriptions Jana G¨otze and Johan Boye School of Computer Science and Communication KTH Royal Institute of Technology Stockholm, Sweden {jagoetze,jboye}@kth.se Abstract This article describes SPACEREF, a corpus of street-level geographic descriptions. Pedestrians are walking a route in a real urban environment, describing their actions. Their position is automatically logged, their speech is manually transcribed, and their references to objects are manually annotated with respect to a crowdsourced geographic database. We describe how the data was collected and annotated, and how it has been used in the context of creating resources for an automatic pedestrian navigation system. Keywords: reference resolution, pedestrian wayfinding, geographic data 1. Introduction We are introducing SPACEREF, a small corpus of spoken geographic and spatial descriptions given by pedestrians while moving in an urban environment. The corpus con- tains transcribed utterances, along with the pedestrians’ GPS coordinates and information about which objects in their geographical surroundings they are referring to. We believe this corpus will be a useful resource for researchers studying reference resolution, landmark salience in the con- text of route instructions (Richter, 2013), geographical di- alogue systems (Boye et al., 2014), and qualitative spatial reasoning (Freksa, 1991). These research problems are tightly interconnected, but of- ten studied in isolation and on the basis of different data sources. References to objects in the surrounding physical environment, like they are often found in situated speech, are typically studied in small-scale environments such as objects aligned on a table (Matuszek et al., 2014). Mech- anisms that study aspects of human-like route instructions, including references to landmarks, are often based on data collections where participants give written instructions, in- structions for prospective routes (i.e. instructions that need to be remembered by the route follower), or instructions in virtual environments (Lovelace et al., 1999; Goschler et al., 2008; Hirtle et al., 2010, for example). One reason for this is that collecting data in real environments is generally a time-consuming undertaking that offers limited possibility to control the experimental conditions. In this paper, we describe our efforts of collecting data of pedestrians walking in an urban environment and describ- ing their actions as if talking to a route follower. The data has been collected with two specific applications in mind, namely as basis for developing algorithms that resolves ref- erences in the real physical environment while pedestrians are moving in it, and as basis for deriving models of land- mark salience from mentions of landmarks in specific rout- ing situations. Both applications are necessary in a system that can automatically give route instructions to pedestrians (Boye et al., 2014). The aim was therefore to put the pedes- trians into a situation that resembles as closely as possible the situation of a potential user of such a system. In the remainder of this paper, we describe the two studies L2 L1 ‘I continue in a southwesterly direction down the steps [L1] towards the arch at the bottom [L2]’ Excerpt from its representation in SPACEREF (the identifier numbers …

Content truncated to 500 words