::1997-1998::
My first project was the LINGUADUCT project. The research goal of this project was to investigate contextual interpretation of natural language through abductive reasoning and inductively acquired knowledge. In this project, we investigated the role of algorithm bias in the context of part-of-speech tagging. During the course of this project, we also fine-tuned a memory-based version of Rens Bod's Data-Oriented Parsing model. The latter work was included in my PhD thesis.

::1998-2002::
During the next four years, I was a research fellow of the Fund for scientific research (aspirant FWO). In this project, we investigated an agent-based evolutionary computing approach to memory-based syntactic parsing of natural language. Starting off with a fully fledged data-driven parser (cf. the aforementioned memory-based DOP model), we gradually dismantled the parser's data sources to eventually arrive at a society model for the emergence of grammatical concepts. The general framework for this research was dubbed GRAEL (GRAmmar EvoLution). Most research was spent investigating how an agent-based society can be used to optimize the probability mass distribution of an annotated treebank. While the evolutionary aspect of GRAEL made sure that several solutions were considered and optimized over time, the agent-based properties of GRAEL ensured that the optimization process was grounded in the task at hand: parsing natural language. You can download my PhD thesis here.

::2002-2006::
One day after my PhD defense, I started working on the FLaVoR project. This is a joint project between CNTS at the University of Antwerp and ESAT at the University of Leuven. The FLaVoR project tries to develop a new modular architecture for speech recognition which allows for the introduction of relatively intricate language models. My particular job in this project is the development of a module for prosodic annotation and a morpho-syntactic language model. The morphological model developed at the University of Antwerp uses memory-based learning. You should check out the demo. The syntactic part of the language model makes use of shallow parsing to present structural information in a sentence.

::2004::
I also took part in an investigation on speech rate in the Dutch language area. Using a corpus of 160 conversations with teachers of Dutch all over Flanders and the Netherlands, we were able to determine average speech rate for the different areas. The results were surprising in that we indeed found a statistically significant difference in speech rate between Dutch people and Flemish people (the former speaking significantly faster). We also found that men speak faster than women. Less surprisingly, we also found that older people speak slower than young people. The prototypical slow speech of people from Limburg could not be observed.

::2006-...::
African Language Technology