PropBank

The PropBank dataset (or, Proposition Bank) is annotated with Semantic Role Label.

The dataset is available here [1] by request.

Sentences of the FrameNet come from the British National Corpus, which provides the largest corpus of English with a balanced mixture of text genres. The corpus are annotated with automatically generated part-of-speech tags for each word, but it does not have full syntactic parses.

The FrameNet roles are defined based on each semantic frame. A frame is a representation of situations regarding various participants, props, and other conceptual roles. For example, the frame CONVERSATION has related verbs like argue, banter, debate, converse, and gossip, and the roles defined in this frame are PROTAGONISTS, MEDIUM, and TOPIC.