1  Introduction

This website collects and briefly introduces publications and other results that came out of the "InPro Project",1 which was funded by the German Research Council in the Emmy Noether Programme and ran from 2007 to 2011, initially at the University of Potsdam, Germany, and from 2010 on at Bielefeld University, Germany. PI of the project was David Schlangen, with members of the group were Timo Baumann, Michaela Atterer, Gabriel Skantze and Okko Buß.

The goal of the project, as specified in the 2006 grant proposal, was ...

... to build a formal, computational model of two key aspects of human dialogue interaction management: turn taking and feedback giving [...]. The model will be derived from an empirical study of corpora of human-human interaction, and will be implemented within a framework for dialogue management in general (including dialogue content management). To achieve this, the project will develop novel techniques for incremental processing in dialogue systems, including incremental parsing, dialogue act recognition, and dialogue content management.

The plan for reaching this goal was ...

... to first construct quantitative models of the phenomena by collecting and analysing data, and then use these to formulate generative models which will be implemented in a computer system. To guarantee an appropriate level of concreteness, the model of these interaction management phenomena will be embedded in a general model of dialogue management.
An essential part of the project will be the development of incremental dialogue processing techniques which are crucial for enabling the projection required in dynamic behaviour. In the first phase of the project, the model will first be evaluated in a dialogue-"simulation" setting, where two copies of the system `converse' with each other (and hence act both as speaker and hearer). [...] This represents a novel development paradigm, in which time-critical behaviour can be modelled in abstraction from engineering constraints [...].
The setting will allow us to move incrementally from full simulation (first half of phase-I) towards full real-time behaviour (phase-II), by allowing us to test ever more components with direct user-studies, and to incrementally replace simulated modules with automatic speech recognition techniques (phase-II).

As is often the case, especially with longer-term projects, the goals and the adopted approach evolved during the lifetime of the project. Empirical study of human-human interaction became somewhat less central than initially planned - although we did collect various corpora, using them more for training models than for exploration -, and the computational modeling / system engineering aspect somewhat more. This was mostly due to our early realisation that real-time processing using ASR was already within reach, and hence we (mostly) skipped the simulation setting and directly worked on real-time, user-facing systems. While we kept turn-taking and feedback-giving as our main showcase phenomena that are enabled by incremental processing, we also extended our interest to modelling non-linguistic "overlapping" actions.
Something that we didn't anticipate fully in the original proposal was the extent to which incremental processing requires a rethinking of modularity in language processing. We spent a lot of effort not only on developing incremental processing modules (described here in Section 4), but also on develeping a principled architecture for combining such modules in an end-to-end system. This architecture is described in Section 3, and the systems in Section 5.
In the time since we have started the project, incremental processing has become a more active area of research in spoken dialogue systems, as evidenced by the increasing number of publications on this topic, and the formation of research projects that include this topic. We are not free of the somewhat immodest hope that our work has contributed to this development, and that our work can provide a basis for some of the future work that we will see in this exciting area.

Footnotes:

1 The full title of the project was: "Incrementality and projection in dialogue processing:
interfacing interaction management and content management in dialogue".

About...

This website reports on some results of the research project "InPro", which was led by David Schlangen and ran from 2006 to 2011. The information contained in this website is for general information purposes only; we make no representations or warranties of any kind, express or implied, about the completeness, accuracy, reliability, suitability or availability with respect to the website or the information, products, services, or related graphics

... Continued

contained on the website for any purpose. Any reliance you place on such information is therefore strictly at your own risk.
Through this website you are able to link to other websites which are not under the control of us or Bielefeld University. We have no control over the nature, content and availability of those sites. The inclusion of any links does not necessarily imply a recommendation or endorse the views expressed within them.