Artificial Conversational Companions

This document describes the problem statement, the methodological framework, the current state of the work and the expected contribution of my doctoral dissertation. The main focus of my dissertation is long-term interaction with an Artificial Conversational Companion in the context of conversation training for second language acquisition. I use a data-driven approach and conversation analysis methods to build computational models for long-term interaction as a meaningful activity. I work on the concept of interaction profiles for human-agent interaction. The resulting models will be integrated in an AIML-based chatbot that helps to practice conversation in a foreign language. Introduction and Problem Statement The term Artificial Companion (AC) has been introduced in (Wilks 2005). The most important characteristics of an AC are a sustained discourse over a long time period, a capability to serve interests of the user, and a lot of personal knowledge about the main user. Similar definitions can be found in (Pulman et al. 2010; Benyon and Mival 2008; 2010). An AC is seen as a personalised, helpful and persistent conversational agent that knows its owner and interacts with the user over a long period of time. The form of an AC influences all the issues of interaction and possibilities for companionship (see also (Benyon and Mival 2010)). We therefore use the term Artificial Conversational Companion (ACC) for companions that are aimed to simulate interaction with the user in a natural language. Recent contributions in the domain of ACC are the EU-funded Companions project with the “How Was Your Day” Companion (Pulman et al. 2010), the Senior Companion (Wilks et al. 2011), and the Health and Fitness Companion (Turunen et al. 2011), and the ALIZ-E project focusing on robot companions for children in a hospital environment (Baxter et al. 2011). The idea to use conversational agents in second language acquisition domain (SLA) was investigated using modified chatbots for conversation training (Jia 2009). We consider the application scenario where advanced learners of a foreign language practice conversation in dialogues with a language expert the ACC. We focus on Copyright c © 2013, Association for the Advancement of Artificial Intelligence (www.aaai.org). All rights reserved. interaction via instant messenger (IM) because it combines the advantages of spoken and written communication being conceptually oral and medially written (Koch and Oesterreicher 1985). In earlier work, we identified the minimum requirements that an artificial agent must satisfy in order to be mentioned as an ACC (Danilava, Busemann, and Schommer 2012). We refined the requirements for the application scenario of conversation training in SLA (Danilava et al. 2013). Language acquisition “requires meaningful interaction in the target language [...] in which speakers are concerned not with the form of their utterances but with the messages they are conveying and understanding” (Krashen 1981). However, the design of the agents is still focused on the content of responses, but not on language as a meaningful activity co-constructed according to rules of social interaction. This research was inspired by the work on interaction profiles by (Spranz-Fogasy 2002). Interaction profiles incorporate the entire interactional phenomena of a talk and the connections among them related to each single participant of an interaction. Our investigations on interaction profiles for ACC rely on an analysis of an empirical data set of IM interactions and focus in particular on the following questions: 1. How the participants of an IM interaction make the meaningful activity, social interaction and emotions explicit by means of an IM chat? How can these phenomena be implemented in an AIML-based ACC? 2. Pattern based language understanding of learner language. What strategies language experts apply in an interaction if learners produce errors? How these strategies can be implemented in an AIML-based ACC? 3. Which strategies the users are likely to use to indicate non-understanding? Which strategies can the ACC apply for meaning negotiation? How can these strategies be implemented in an AIML-based ACC? 4. How can learner’s responsiveness values be use for recognition of particular types of turns, e.g. self-repairs?