When dialogue systems, through the use of incremental processing, are not bounded anymore by strict, nonoverlapping turn-taking, a whole range of additional interactional devices becomes available. We explore the use of one such device, trial intonation. We elaborate our approach to dialogue management in incremental systems, based on the Information-State-Update approach, and discuss an implementation in a microdomain that lends itself to the use of immediate feedback, trial intonations and expansions. In an overhearer evaluation, the incremental system was judged as significantly more human-like and reactive than a non-incremental version. 1

The occurrence of split utterances (SUs) in dialogue raises many puzzles for grammar formalisms, from formal to pragmatic and even philosophical issues. This paper presents an account of some of the formal details that grammars need to incorporate in order to accommodate them. Using Dynamic Syntax (DS), we illustrate how by incorporating essential interaction with the context into the grammar itself, we can deal with speaker change in SUs: not only its effects on indexicals I and you, but also the multiple illocutionary forces that can arise. We also introduce a Split Turn Taking Puzzle (STTP) showing that the current speaker and the agent of the resulting speech act are not necessarily the same. 1

interpretation: the dynamics of ellipsis construal

We present the novel task of predicting temporal features of continuations of user input, while that input is still ongoing. We show that the remaining duration of an ongoing word, as well as the duration of the next can be predicted reasonably well, and we put this information to use in a system that synchronously completes a user’s speech. While we focus on collaborative completions, the techniques presented here may also be useful for the alignment of back-channels and immediate turn-taking in an incremental SDS, or to synchronously monitor the user’s speech fluency for other reasons. 1

This chapter introduces the phenomenon of compound contributions in dialogue and discusses their implications for NLG in interactive systems. Contributions to dialogue are often split across multiple utterances – possibly by different speakers – with each adding to an incrementally emerging representation of meaning. A suitable NLG module must therefore not only behave incrementally – able to use and extend existing partial representations – but must also be tightly integrated with an equally incremental interpretation module. We examine the different aspects of incrementality required, and see that no previous approach to incremental NLG provides them all; we outline one possible approach that does. 1.1

2.1 Dynamic Syntax...................................... 3

to incomplete contributions in dialogue

turn is it anyway?