Semantic Role Labeling
Semantic role labeling is a task of detecting the semantic arguments of a sentence. Typical semantic arguments are usually about roles related with the predicate or verb of a sentence such as agent, patient, and instrument. Recognizing and labeling semantic arguments is relatively domain-independent, and could be important in all NLP tasks such like Information extraction, Question Answering, Machine Translation, and Document Summarization.
Semantic role labeling has been studied in three shared tasks at Senseval 3, CoNLL-2004 and CoNLL-2005.
FrameNet was used for Senseval 3, and PropBank was used for both CoNLL-2004 and CoNLL-2005.
The following sentence from the PropBank corpus shows the example of semantic role labeling.
[A0 He ] [AM-MOD would ] [AM-NEG n't ] [V accept ] [A1 anything of value ] from [A2 those he was writing about ] .
Semantic Role Labeling problem is to determine a label on phrases of a sentence s, given a predicate p.
Subtasks of Semantic Role Labeling are like below.
- Identification: detect argument phrases.
- Classification: decide semantic labels for argument phrases
Accuracy, Precision, Recall, and F-Measure are usually used for evaluation.
The unit of processing of Semantic Role Labeling is a sentence. Depth of semantics can be said shallow. Semantic Role Labeling covers broad domains, and mostly not connected to applications directly.
The Gildea and Jurafsky Computational Linguistics 2002 is the first statistical model on FrameNet solving the semantic role labeling problem.
- Wikipedia article on Semantic Role Labeling - 
- CCG - Illinois Semantic Role Labeling Demo - 
- CoNLL-2004 and CoNLL-2005 Shared Tasks - 
- FrameNet Website 
- PropBank Website