Learning (about) Information Structure
Oana Postolache

Existing theories of Information Structure (IS) state principles about the possible realization of IS, its linguistic meaning and its interpretation within the discourse context using carefully selected illustrative examples. In this talk, I will address what we can learn about IS and how we can learn IS looking at real texts. For investigations, we use the Penn Treebank corpus which we started to augment with information about informativity using a descriptive annotation scheme, un-biased to any of the existing IS theories. We restrict the annotation to discourse entities (NPs and NP-like structures) to which values for various IS-related attributes (familiarity status, informativity status, determination, denotation, quantification, discourse relations as coreference and
bridging relation) are assigned. The goal is to investigate: (1) models for learning the informativity status, and (2) to which extent we can derive diferrent IS theories using our attributes.