The VISL system: Constraint Grammar based NLP-products

The paper presents the VISL system of Natural Language Parsing (NLP), an integrated research and teaching initiative at the University of Southern Denmark (http://beta.visl.sdu.dk). The project has a strong cross-language perspective, and involves 6 research and 22 teaching languages. For reasons of robustness, performance and linguistic maintenance, the system's NLP tools are based on the Constraint Grammar formalism (Karlsson, 1995), but hybrid techniques allow users to choose between different descriptive paradigms, such as shallow word based parsing, dependency grammar and constituent trees. The original kernel of programs was built around a multi-level parser for Portuguese, PALAVRAS (Bick 2000), and modules for other languages, such as the DanGram parser for Danish, employ a similar architechture, with mature systems achieving a tagging accuracy of 99% for part of speech, and 95% for syntactic function. The most developed applicational front end is Computer Aided Language Learning (CALL), with Internet based grammar games and teaching tools for schools and universities, while the main current research focus is on the construction of NLP resources, both parsers and grammatically annotated corpora, such as treebanks (http://corp.hum.sdu.dk).