مارسا12

. From the Text

Mining perspective, they fall short to cover a significant part of the domain knowledge,

i.e. they are still sparsely populated, and they do not incorporate morphological and

syntactical variability (again, not the purpose of an ontological resource). On the

other hand biological researchers put significant effort into the development of more

and more complete ontological resources.

This topic was in the center of repeated and fruitful discussions throughout the whole

meeting. Only to pick a few examples, the following presentations in the seminar

showcased ongoing work spanning from Text Mining to ontologies. Paul Bruitelaar

(DFKI, Saarbrücken, D) focused on the ontology life cycle, highlighting on the need to

constantly adapt ontologies to the changing needs of the community. This topic was

also covered by Jong C. Park (KAIST, Korea) who demonstrated his analyses on

tracking changes in the gene ontology. Robert Stephens (University of Manchester,

U.K.), Christopher Brewster (University of Sheffield, U.K.) and David Shotton (University

of Oxford, U.K.) reported on an ongoing project for bootstrapping an ontology

for animal behavior using Text Mining; initial results from similar projects for phenotype

data and lipid metabolism were presented by Ulf Leser (Humboldt University,

Berlin, D) and Thomas W?chter (University of Dresden, D), respectively. All three

presentations highlighted the problem that despite the many papers on ontology

learning, actually very few methods are readily available. Stephan Schultz (University

of Freiburg, D) explained the latest developments in the BioTOP ontology which intends

to bridge from top-level ontologies like BOF to domain specific ontologies such

as the Gene Ontology. Studies are underway to use BioTop for word sense disambiguation,

an essential step in Text Mining.

Robert Stephens pointed out in his summary of the day, that there is a certain danger

that “ontologists” are disappointed by lack of perfection when using results from Text

Mining for ontology development, and that Text Mining researchers are disappointed

by deficiencies of existing ontologies, such as incompleteness and inadequate modeling

of lexical variation in the terminology used to express the labels of the concepts.

However, the seminar was successful in showing up the borderlines and the crossovers

between both research domains giving inspiration to novel approaches using

the best of both breeds.