Mariet        


Projects (past and ongoing)
The FACT project aims at automatically enriching a corpus of Dutch folktales with metadata such as names, keywords, and a summary. We also classify folktales according to genre and international folktale type, and investigate if useful relations between folktales can be found by means of clustering. FACT is supported by the CATCH (Continuous Access to Cultural Heritage) programme of NWO. CATCH logo
As part of the COMMIT project "Interaction for Universal Access" we investigate the use of interactive storytelling techniques in serious games used for training purposes. Our focus is on providing artificial agents (“non-player characters” in the game) with social intelligence to support training. For this purpose we build on our results from the Virtual Storyteller project, see below. COMMIT logo
The Virtual Storyteller project is aimed at automatic story generation using a multi-agent architecture. The project focuses on the development of intelligent agents, capable of both rational and emotional behaviour. It also involves research on language generation, embodied information presentation and speech synthesis. An article (in Dutch) explaining in simple terms what the Virtual Storyteller is about appeared in the Dutch storytelling magazine Vertel Eens, 1(3), December 2006, p. 29.
Go to the Virtual Storyteller homepage
In the GATE project (Game research for training and entertainment) we try to model the cognitive processes involved in natural communication. We focus on the production of verbal and non-verbal behavior by virtual characters. This behaviour needs to be appropriate in context, i.e. given the character’s personality and mental state, its social and emotional relationship towards other agents, etc.
Go to the GATE homepage
In the ICIS project (Interactive Collaborative Information Systems) we worked on the multimodal presentation of information in high-load presentation scenario's, aiming at the reduction of stress and cognitive load, and the avoidance of framing bias and confirmation bias in decision making. As an application domain we focused on crisis management.
Go to the ICIS page at HMI
IMOGEN was a subproject of the IMIX programme (NWO). In IMOGEN we worked on multimodal output generation in the context of a question-answering system. The following description of our work (in Dutch) appeared in a Dutch magazine on applied language and speech technology: Theune, M., E. Krahmer, W. Bosma, E. Marsi en C. van Hooijdonk. Mag het ietsje meer zijn? Antwoordpresentatie in een QA-systeem. DIXIT, tijdschrift over toegepaste taal- en spraaktechnologie, 4(1), June 2006, pp. 12-15.
Go to the IMOGEN homepage
A past research project is ANGELICA (A Natural-language Generator for Embodied, Lifelike Conversational Agents), funded by NWO. In this project an embodied Virtual Guide was developed that can give route directions in a virtual environment using a combination of modalities, including gestures and text/speech. For Dutch readers, here's an article from the Dutch magazine Computable (2003) in which I talk about the  ANGELICA project.
Go to the Virtual Guide homepage
Go to the GoalGetter homepageAs part of my PhD research, I developed the language generation component of a spoken dialogue system that provided train timetable information. Another system I worked on was GoalGetter, a system that generated spoken soccer reports on the basis of tabular data. To read an interview with me about GoalGetter (in Dutch) click here. My PhD research was carried out at IPO, the Institute for Perception Technology / Center for User-System Interaction of the Eindhoven University of Technology.
One of my specific interests in the area of natural language generation is the generation of referring expressions (GRE). With the GRAPH team (using a graph-based GRE algorithm proposed by Krahmer et al. 2003) we have been participating in the 2007 ASGRE Challenge, the 2008 REG Challenge, and the 2009 TUNA Challenge.


BACK TO HOMEPAGE