A corpus with semantic role tags for an NLP application

So, I've constructed a NLP program that learns to extract a semantic event description from a sentence, but right now my training set is limited to sentences I've parsed into semantic event components my hand.

While this method does get the job done, its hardly a proper substitute for a large pre-parsed corpus of text. Unfortunately, all of my attempts at finding such a corpus have proven futile.

What I need specifically is a corpus that has tagged the semantic开发者_JAVA百科 roles of each word (or group of words) in a sentence. Examples of roles I had in mind are things like:

• agent
• action
• patient
• instrument
• co-agent
• co-patient
• location
• adverb

If any more specifics are needed, feel free to ask, or refer to this paper that uses a toy corpa with the same constraints as mine.

The CoNLL Shared Task in 2005 was 'Semantic Role Labelling'. This page describes their corpus and what roles they labelled.