Acquisition and Assessment of Semantic Content for the Generation of Elaborateness and Indirectness in Spoken Dialogue Systems

Proceedings of the The 8th International Joint Conference on Natural Language Processing, pages 915–925, Taipei, Taiwan, November 27 – December 1, 2017 c⃝2017 AFNLP Acquisition and Assessment of Semantic Content for the Generation of Elaborateness and Indirectness in Spoken Dialogue Systems Louisa Pragst1, Koichiro Yoshino2, Wolfgang Minker1, Satoshi Nakamura2 and Stefan Ultes3 1Ulm University, Ulm, Germany 2 Nara Institute of Science and Technology (NAIST), Nara, Japan 3 University of Cambridge, Cambridge, UK {louisa.pragst,wolfgang.minker}@uni-ulm.de {koichiro,s-nakamura}@is.naist.jp, su259@cam.ac.uk Abstract In a dialogue system, the dialogue man- ager selects one of several system actions and thereby determines the system’s be- haviour. Defining all possible system ac- tions in a dialogue system by hand is a te- dious work. While efforts have been made to automatically generate such system ac- tions, those approaches are mostly focused on providing functional system behaviour. Adapting the system behaviour to the user becomes a difficult task due to the limited amount of system actions available. We aim to increase the adaptability of a dia- logue system by automatically generating variants of system actions. In this work, we introduce an approach to automatically generate action variants for elaborateness and indirectness. Our proposed algorithm extracts RDF triplets from a knowledge base and rates their relevance to the origi- nal system action to find suitable content. We show that the results of our algorithm are mostly perceived similarly to human generated elaborateness and indirectness and can be used to adapt a conversation to the current user and situation. We also discuss where the results of our algorithm are still lacking and how this could be im- proved: Taking into account the conver- sation topic as well as the culture of the user is likely to have beneficial effect on the user’s perception. 1 Introduction In a dialogue system (DS), the dialogue man- ager (DM) is responsible for choosing the sys- tem’s contribution to a conversation. Several stud- ies (e.g. (Ultes et al., 2015; Bertrand et al., 2011; Jaksic et al., 2006; Partala and Surakka, 2004)) show that adjusting the system’s behaviour to the user can improve the user experience. To enable such adaptivity, the system needs several possi- ble dialogue actions from which to choose. Of- ten, those system actions are predefined manu- ally. Hence, the amount of variants that the DM can choose from to adapt the system behaviour is limited by conversational skills, creativity and time of the person responsible for creating those actions. Foreseeing every possible situation the DS could find itself in and coming up with mul- tiple viable system actions while considering pos- sible types of users and their preferences in a con- versation is demanding work. Approaches to the automatic generation of system actions, such as (Kadlec et al., 2015), have been presented to fa- cilitate that process. However, those approaches often consider only system actions that are nec- essary from a functional point of view. There is no variety of system actions produced that would enable the DM to adapt to specific users charac- teristics …

Content truncated to 500 words