Several Natural Language Processing (NLP) tasks, such as Information Extraction, Question Answering or Dialogue, require semantic interpretation of arbitrary texts. Paradigms proposed as meta-models of semantic structures include frame semantics (Fillmore, 1985). According to this theory, frames are linguistic predicates providing a semantic description of real world situations. A frame is evoked in a sentence through specific lexical units (lu), i.e. words (such as nouns or verbs). A frame characterizes the set of semantic roles, i.e. semantic arguments called Frame Elements (FE), describing all participants to the event. For example, the following sentence evokes the KILLING frame, through the lu “kill”, while four FEs, i.e. TIME (yesterday), KILLER (a robber), VICTIM (a guardian) and INSTRUMENT (with a knife), are emphasized in the underlying frame:
[Yesterday]TIME, [a robber]KILLER killed [a guardian]VICTIM [with a knife]INSTRUMENT .
The task of automatic recognizing predicates and FEs in sentences is called Semantic Role Labeling (SRL)(Gildea and Jurafsky, 2002) (Palmer et al, 2010), recently pushed by the FrameNet project that made available a large set of annotated sentences from the British National Corpus (BNC).
In our group we investigate two main Machine Learning approaches to SRL based on Structured Learning paradigms:
- one approach is based on a novel formulation of Convolution Kernels, namely the Smoothed Partial Tree Kernel that models semantic roles by implicitly combining syntactic and lexical information of annotated examples.
- one approach realizes a flexible approach based on the Markovian formulation of the SVM learning algorithm, i.e. SVM-HMM by modeling the SRL problem a sequential tagging problem. A demo system called BABEL has been implemented on this latter approach and it is described in this page.
Results in line with the state-of-the-art have been achieved with both approaches both in English and Italian languages.
Fillmore, Charles J., Frames and the semantics of understanding. Quaderni di Semantica. 6: 222-254 (1985)
Gildea, D., Jurafsky, D.: Automatic Labeling of Semantic Roles. Computational Linguistics 28(3), 245–288 (2002)
Palmer, M., Gildea, D., Xue, N.: Semantic Role Labeling. Synthesis Lectures on Human Language Technologies, Morgan & Claypool Publishers (2010)