Ready for the Next Step? - Investigating the Effect of Incremental Information Presentation in an Object Fetching Task Monika Chromik CITEC - Bielefeld University 33615 Bielefeld, Germany mchromik@techfak.uni- bielefeld.de Birte Carlmeyer CITEC - Bielefeld University 33615 Bielefeld, Germany bcarlmey@techfak.uni- bielefeld.de Britta Wrede CITEC - Bielefeld University 33615 Bielefeld, Germany bwrede@techfak.uni- bielefeld.de ABSTRACT In this paper we present a human-agent interaction study investigating the effect of incremental (just in time) infor- mation presentation on human task performance and the subjective ratings of the agent. On the one hand we show that the task of fetching ingredients and utensils for cooking is better performed in case of incremental information pre- sentation. On the other hand it yields a negative effect on the subjective ratings of the agent. Keywords incremental processing; dialog management; multi-modal systems 1. INTRODUCTION Through their rich sensing and acting capabilities smart homes provide fundamentally new means of interaction that allow for semantically deep interpretations of the user’s be- havior in order to provide support exactly when it is needed. A first step towards this vision are assistive systems which support a user in daily activities such as packing a bag or fetching ingredients for a cake or a menu. While apps that support such activities on mobile devices by reading out lists already exist almost no integration of environmental or on- board sensors is foreseen that allow the system to adapt to the user’s current activity. In [3] it was shown that adaptation of an information providing app to the user’s current activity yields better memory performance by interrupting the information pre- sentation process during difficult phases. Interestingly, this positive memory effect could not be replicated in a smart home environment [2] where the system provided informa- tion and interrupted itself when the user was distracted by events in the environment. [5] investigated this relationship in more depth in a scenario where a robot dictated non- ACM ISBN 978-1-4503-2138-9. DOI: 10.1145/1235 Figure 1: Person interacting with the agent Flobi. native (high difficulty) and native (low difficulty) sentences to a user whose task it was to write them down. In the easy task the users speeded up their activity in the adaptive condition while they used the adaptation capability in the difficult task to slow down between sentences. However, this study did not report any effects on errors. The exact nature of the task thus seems to play an impor- tant role in determining in detail what kind of adaptation can be helpful for the user. In our study we therefore follow a two step approach in order to determine optimal adap- tation strategies in the preparation phase of a cooking sce- nario: (1) we investigated how users would structure their information in a human-human interaction and (2) used the gained insights to model a (wizarded) adaptive system be- havior and compared it to a non-adaptive system behav- ior. Importantly, we not only looked at duration but also at errors as dependent variable. Results indicate that an adaptation strategy based on chunking the information into smaller packages with timing signals derived from the user’s ostensive behavior yields objectively better performance but receives lower subjective evaluations especially wrt the tim- ing behavior. 2. EXPERIMENT Generation of Adaptation Strategy. In order to assess how users would structure the infor- mation in the preparation phase of a cooking scenario we carried out a human-human pilot study. In total 12 sub- jects took part in 6 trials with 2 subjects interacting with