Dealing with Trouble: A Data-Driven Model of a Repair Type for a Conversational Agent

Troubles in hearing, comprehension or speech production are common in human conversations, especially if participants of the conversation communicate in a foreign language that they have not yet fully mastered. Here I describe a data-driven modeling approach for simulation of dialogue sequences where the learner user does not understand the talk of a conversational agent and asks for clarification. Conversational agents for educational purposes, specifically for Second Language Acquisition (SLA) (Stewart and File 2007) use different approaches to support language learning through conversation. CSIEC chatbot (Jia 2009) can correct spelling errors. CLIVE (Zakos and Capper 2008) understands input in the native language of the learner. The language and culture training system (Sagae, Johnson, and Valente 2011) supports learning in the form of task based dialogues with agents in a serious game environment. The systems are supposed to simulate the native speaker (NS) in conversations with...

• Type: Journal
• Year: 2015
• Where: AAAI
• Authors: Sviatlana Höhn
• Keywords: AAAI 2015, Intelligent Agents