cc.mallet.fst.confidence
Class ConstrainedViterbiTransducerCorrector

java.lang.Object
  cc.mallet.fst.confidence.ConstrainedViterbiTransducerCorrector

All Implemented Interfaces:

TransducerCorrector

public class ConstrainedViterbiTransducerCorrector
extends java.lang.Object
implements TransducerCorrector

Corrects a subset of the Segments produced by a Transducer. It's most useful to find the Segments that the Transducer is least confident in and correct those using the true Labeling (correctLeastConfidenceSegments). The corrected segment then propagates to other labelings in the sequence using "constrained viterbi" -- a viterbi calculation that requires the path to pass through the corrected segment states.

Constructor Summary

ConstrainedViterbiTransducerCorrector(Transducer model)

ConstrainedViterbiTransducerCorrector(TransducerConfidenceEstimator confidenceEstimator, Transducer model)

Method Summary

java.util.ArrayList	correctLeastConfidentSegments(InstanceList ilist, java.lang.Object[] startTags, java.lang.Object[] continueTags)
java.util.ArrayList	correctLeastConfidentSegments(InstanceList ilist, java.lang.Object[] startTags, java.lang.Object[] continueTags, boolean findIncorrect) Returns an ArrayList of corrected Sequences.
java.util.ArrayList	getLeastConfidentSegments() Returns the least confident segments from each sequence in the previous call to correctLeastConfidentSegments
java.util.ArrayList	getLeastConfidentSegments(InstanceList ilist, java.lang.Object[] startTags, java.lang.Object[] continueTags) Returns the least confident segments in ilist
java.util.Vector	getSegmentConfidences()

Methods inherited from class java.lang.Object

clone, equals, finalize, getClass, hashCode, notify, notifyAll, toString, wait, wait, wait

Constructor Detail

ConstrainedViterbiTransducerCorrector

public ConstrainedViterbiTransducerCorrector(TransducerConfidenceEstimator confidenceEstimator,
                                             Transducer model)

ConstrainedViterbiTransducerCorrector

public ConstrainedViterbiTransducerCorrector(Transducer model)

Method Detail

getSegmentConfidences

public java.util.Vector getSegmentConfidences()

getLeastConfidentSegments

public java.util.ArrayList getLeastConfidentSegments()

Returns the least confident segments from each sequence in the previous call to correctLeastConfidentSegments

getLeastConfidentSegments

public java.util.ArrayList getLeastConfidentSegments(InstanceList ilist,
                                                     java.lang.Object[] startTags,
                                                     java.lang.Object[] continueTags)

Returns the least confident segments in ilist

Parameters:

ilist - test instances

startTags - indicate the beginning of segments

continueTages - indicate "inside" of segments

Returns:

list of Segments, one for each instance, that is least confident

correctLeastConfidentSegments

public java.util.ArrayList correctLeastConfidentSegments(InstanceList ilist,
                                                         java.lang.Object[] startTags,
                                                         java.lang.Object[] continueTags)

Specified by:

correctLeastConfidentSegments in interface TransducerCorrector

correctLeastConfidentSegments

public java.util.ArrayList correctLeastConfidentSegments(InstanceList ilist,
                                                         java.lang.Object[] startTags,
                                                         java.lang.Object[] continueTags,
                                                         boolean findIncorrect)

Returns an ArrayList of corrected Sequences. Also stores leastConfidentSegments, an ArrayList of the segments to correct, where null entries mean no segment was corrected for that sequence.

Parameters:

ilist - test instances

startTags - indicate the beginning of segments

continueTages - indicate "inside" of segments

findIncorrect - true if we should cycle through least confident segments until find an incorrect one

Returns:

list of Sequences corresponding to the corrected tagging of each instance in ilist