Question

Explain in detail about the Semantic Role Labelling with Conditional Random Fields. Use flowcharts and diagrams to explain your points better.

Answer 1

Semantic role labeling (SRL), generally, is the NLP task of automatically assigning semantic roles to predicate arguments according to labeling schemes like PropBank and FrameNet.

Conditional random fields are discriminative models with an undirected graphical structure belonging to the general class of graphical models.

CRFs are aimed at structured learning problems such as sequence, graph, and tree labeling which makes them apropos for labeling or segmenting natural language data. CRFs have been successfully applied to a wide-range of NLP tasks including SRL

often with state of the art or near state of the art results.CRFs, being discriminative in nature, allow conditioning on arbitrarily large feature sets.Accommodation of a large yet appropriate feature space has been shown to be a critical factor in the construction of SRL systems

A typical linear-chain CRF defines the conditional probability of a label sequence y given an observation sequence x as:

p(y|x,Λ) = 1 Z(x) Y c∈C exp( X k λkfk(yc , xc))

where Λ is a set of parameters, Z(·) is a partition function, fk(·) are feature functions defined on input, and the set C is the set of cliques or transitions defined over the graph.

The system consists of a supervised machine learning pipeline, using automatically tagged, tokenized, and dependency parsed English language data that: (1) extracts features, (2) generates a set of of relations under consideration for each predicate, and (3) trains and tests SRL labeling over the sequence of suggested relations