Networks of Linguistic Annotation: The Linguist's Web

Anette Frank (University of Heidelberg)

Treebanks have had an enormous impact on natural language processing 
over the last ten years. The field has moved from monolingual and 
monostratal annotation to multi-layer annotation, thus adding a 
vertical dimension to annotated corpora. In the future we need to work
towards a third dimension: cross-linguistic annotation. 

I will reflect on different ways of creating and exploiting corpora
along these different dimensions, trying to assess their prospects and
limits. I will contrast these prospects with my own and the community's 
experience in the exploitation of treebanks for grammar induction and 
evaluation, and the induction of language resources in general, 
focusing on the areas of syntactic and semantic processing.
I will illustrate these observations with activities and experiences 
centered around the TIGER corpus, and will conclude with some reflections 
on standardisation efforts within or across grammar frameworks.

Unified linguistic annotation across different frameworks and languages 
is what seems to be required to be able to take full advantage of the 
vision and prospects of multi-layer and multi-lingual annotation. 
However, unified annotation can by no means be achieved by zeroing out
differences --- else nothing of substance will remain.
Drawing on experiences gained in multi-level annotation carried out in 
the Saarbrücken SALSA project, I will argue for a logical and transparent 
formalisation of annotated corpora that allow us to model and bridge
differences between annotations in a clear and transparent way.