S-GIVE: A Corpus of Spoken Instructions in Virtual Environments

Introduction

The S-GIVE corpus is a corpus of human-human interactions in a virtual environment. It uses the treasure hunting task from the GIVE Challenge. This task involves a pair of human participants. The instruction follower (IF) can move around and manipulate buttons in a virtual environment, but has no knowledge of the map of the world or the specific behavior of objects within that world (such as, which buttons to press to open doors or the safe). The instruction giver (IG) has complete knowledge of the world, but cannot move around in it or interact with it directly, and therefore has to give instructions to the IF to guide him/her to accomplish the task.

The setup of our data collection study was similar to the one used to collect the GIVE-2 corpus of typed instructions (Gargett et al., 2010, pdf). But instead of typing their instructions, the IGs gave spoken instructions in our setup.

We have collected a small pilot corpus of German data, which is described in (Striegnitz et al., 2012, pdf) and are now preparing a larger English corpus.

We will make the corpus available here, once it is completely prepared and annotated. We have finished collecting data, and are currently transcribing them.

Publications

Striegnitz, K., H. Buschmeier, and S. Kopp (2012, May). Referring in Installments: A Corpus Study of Spoken Object References in an Interactive Virtual Environment. In Proceedings of the 7th International Natural Language Generation Conference (INLG). [pdf]

Contact

Kristina Striegnitz: striegnk@union.edu

Hendrik Buschmeier: hbuschme@uni-bielefeld.de

Links