LTH                      Naturvetenskaplig Fakultet
Institutionen för datavetenskap
UTBILDNING:   LTH    NatFak    Forskarutbildning                          FORSKNING

 

Research Outline

^Presentation

My research area is within natural language processing and I am currently mainly concerned with semantic processing and applications of it. Semantic parsing is a recent achievement. Together with Richard Johansson, we participated in the CoNLL 2008 shared task on the joint parsing of syntactic and semantic dependencies and the SemEval-2007 task on Frame-semantic Structure Extraction. In both evaluations, our systems obtained the best results. We have a home page here that lists our activities and links to software we have developed.

I am the author of textbook on natural language processing. It was published by Springer and it is now on its second printing. I maintain a page that provides free chapters, slides, and quantities of links on many aspects of the field. See here.

I am also involved in the ROSETTA project, where I participate in the construction of a knowledge integration framework to represent knowledge and skills in robotized production. Before I was with the SMErobot to build ontologies to generate user interfaces. More generally, I am interested in the use of NLP techniques in advanced interfaces. This includes the design and the implementation of conversational agents within a multimodal framework and text visualization. The Carsim system is a former project that exemplifies it.

Carsim's objective is to analyze car accident reports and to recreate them visually in a three-dimensional space. Carsim tries to reconstruct -- to imagine -- the written text through a symbolic animation. We started to work with the results of a deep semantic analysis of the events contained in the report. Here are some examples of what we obtained: collision and overtaking [1], [2], [3]. We then used information extraction techniques that enable us to visualize nearly 35% of the reports in a corpus of French texts. The Swedish version of Carsim is available online. You can try it here.

Direkt Profil was another project. Its purpose was to evaluate text and profile learners of French. It was done done in cooperation with the Institute of Romance languages at Lund university.

Before Carsim, with a small group of students, we designed and implemented a conversational agent -- Ulysse -- that enables a user to navigate in a virtual reality environment using language. Ulysse parses the word stream and builds a logical form from it. It features a chart parser and a semantic module. Ulysse has a reference solver to map words to entities of the virtual world and a geometric reasoner. Ulysse is embedded into the virtual representation of the user in the world and it animates using planning rules. It then navigates this user into the world. Ulysse has been designed with the DIVE environment from the SICS. It can be interfaced with a speech recognition system such as IBM ViaVoice for instance. In the beginning, the whole environment was intended to be a teleconferencing tool and the agent, a supporting device.

Finally, more generally I am also interested in cognitive links between language, visualization, and psychology. I took part in two European community projects, VREPAR-II and VEPSY, to use virtual reality in clinical psychology where I led the French group on social phobias. VESPY won an honorable mention from the eEurope award program .

^Documents and Source Code

Some documents in this section are in French. If you can't read it, try Google, Voilà, or Babelfish.

Semantic parsing

Direkt Profil

Carsim

Carsim Old Stuff

Ulysse

^Two Three-Dimensional Models

Click on the images to get the VRML worlds.

The Ithaque world of the Ulysse project

The car model of the Carsim project

Ithaque

Carsim

^PhD and Master's Theses

Current

Magister

Alumni

PhDs

Magister

Institutionen för Datavetenskap 2008. Ansvarig: Pierre Nugues
Last update: Tuesday, 21-Sep-2010 10:28:13 CEST