Intelligent interactive systems, such as conversational agents or social robots are expected to use natural language and synthetic speech to communicate with their human interaction partners. Usually, the utterances such a system is able to produce are authored by its programmers or professional writers. This limits a system's expressivity in unexpected situations. Natural language generation (automatic planning of what to say, and how to say it) is an approach to utterance production that can overcome this issue.
The seminar will provide a state-of-the art overview of several aspects of natural language generation in interactive systems. Starting with the basic question of communicative intention, we will discuss ‘classic’ NLG topics (e.g., reference, uncertainty, style) from an interactive point of view, inherently dialogue related NLG topics (e.g., how to achieve understanding, language as joint action), and multimodal behaviour generation (e.g., speech-accompanying gestures).
Teilnahmevoraussetzungen, notwendige Vorkenntnisse
Natural language generation is an interdisciplinary research area comprising of (computational) linguistics and artificial intelligence. Students are expected to have prior knowledge in at least one of these two disciplines.
Our primary text for this seminar will be a recent (2014) edited collection:
Stent, A. & Bangalore, S. (2014). Natural Language Generation in Interactive Systems. Cambridge University Press, Cambridge, MA, USA. doi:10.1017/CBO9780511844492
The university library has licensed the eBook version (PDF) of this edited collection. It is freely accessible from within the university network (via LAN/WLAN or VPN): http://dx.doi.org/10.1017/CBO9780511844492