Multi-threaded Interaction Management for Dynamic Spatial Applications

Proceedings of the of the EACL 2014 Workshop on Dialogue in Motion (DM), pages 48–52, Gothenburg, Sweden, April 26-30 2014. c⃝2014 Association for Computational Linguistics Multi-threaded Interaction Management for Dynamic Spatial Applications Srinivasan Janarthanam Interaction Lab Heriot-Watt University Edinburgh sc445@hw.ac.uk Oliver Lemon Interaction Lab Heriot-Watt University Edinburgh o.lemon@hw.ac.uk Abstract We present a multi-threaded Interaction Manager (IM) that is used to track differ- ent dimensions of user-system conversa- tions that are required to interleave with each other in a coherent and timely man- ner. This is explained in the context of a spoken dialogue system for pedestrian navigation and city question-answering, with information push about nearby or vis- ible points-of-interest (PoI). 1 Introduction We present a multi-threaded Interaction Manager (IM) that is used to track different dimensions of user-system conversations and interleave the dif- ferent converational threads coherently. The IM that we present interacts with the user in a spa- tial domain and interleaves navigation informa- tion along with historical and cultural information about the entities that users can see around them. In addition, it aims to answer questions that users might have about those entities. This presents a complex conversational situation where several conversational threads have to be interleaved in such a way that the system utterances are pre- sented to the user at the right time but in a pri- oritised order, and with bridging utterances when threads are interrupted and resumed. For instance, a navigation instruction may be important (since the user is walking up to a junction at which they need to turn) and therefore it needs to be spoken before continuing information presentation about an entity or answering other ongoing questions. 2 Related work Previously, multi-threaded interaction was used to handle multiple simultaneous tasks in human- robot interaction (HRI) scenarios (Lemon and Gruenstein, 2004). This idea also turns out to be important for cases where humans are interacting with a variety of different web-services in paral- lel. Human multitasking in dialogue is discussed in (Yang et al., 2008). (Lemon and Gruenstein, 2004) presented a multi-threaded dialogue management approach for managing several concurrent tasks in an HRI scenario. The robot could, for example be flying to a location while simultaneously searching for a vehicle, and utterances about both tasks could be interleaved. Here, conversational threads were managed using a representation called the “Dia- logue Move Tree”, which represented conversa- tional threads as branches of the tree, linked to an “Activity Tree” which represented the states of on- going robot tasks (deliver medical supplies, fly to a waypoint, search for a truck), which could be ac- tive simultaneously. The situation for our pedes- trian navigation and information system is simi- lar - concurrent tasks need to be managed coher- ently via conversation. The approach adopted in this paper is similar to (Lemon and Gruenstein, 2004). However, in this work we separate out a domain-general thread called ‘dialogue control’ which handles generic issues like clarification of reference across all tasks. This increasing modu- larisation of the dialogue threads makes …

Content truncated to 500 words