net.sf.okapi.steps.gcaligner

Class GaleAndChurch<T>

java.lang.Object

net.sf.okapi.steps.gcaligner.GaleAndChurch<T>

All Implemented Interfaces:

AlignmentScorer<T>

public class GaleAndChurch<T>
extends Object
implements AlignmentScorer<T>

Constructor Summary

Constructors

Constructor and Description
GaleAndChurch()

Method Summary

Modifier and Type	Method and Description
int	contractionScore(T p_sourceTuv1, T p_sourceTuv2, T p_targetTuv) Calculate the cost of contracting two source segments to one target segment.
int	deletionScore(T p_sourceTuv) Calculate the cost of deletion of source segment.
int	expansionScore(T p_sourceTuv, T p_targetTuv1, T p_targetTuv2) Calculate the cost of expanding one source segment to two target segments.
int	insertionScore(T p_targetTuv) Calculate the cost of insertion of target segment.
int	match(int len1, int len2) Return -100 * log probability that an source sentence of length len1 is a translation of a foreign sentence of length len2.
int	meldingScore(T p_sourceTuv1, T p_sourceTuv2, T p_targetTuv1, T p_targetTuv2) Calculate the cost of melding of two source segments to two target segments.
double	prob(int len1, int len2) Return the probability that an source sentence of length len1 is a translation of a foreign sentence of length len2.
void	setLocales(LocaleId p_sourceLocale, LocaleId p_targetLocale) Set source and target locales.
int	substitutionScore(T p_sourceTuv, T p_targetTuv) Calculate the cost of substitution of source segment by target segment.

Methods inherited from class java.lang.Object

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

Constructor Detail

GaleAndChurch

public GaleAndChurch()

Method Detail

setLocales

public void setLocales(LocaleId p_sourceLocale,
                       LocaleId p_targetLocale)

Set source and target locales.

Specified by:

setLocales in interface AlignmentScorer<T>

Parameters:

p_sourceLocale - Source locale

p_targetLocale - Target locale

substitutionScore

public int substitutionScore(T p_sourceTuv,
                             T p_targetTuv)

Calculate the cost of substitution of source segment by target segment.

Specified by:

substitutionScore in interface AlignmentScorer<T>

Parameters:

p_sourceTuv - Source TUV. Source is in X sequence in the DP map.

p_targetTuv - Target TUV. Target is in Y sequence in the DP map.

Returns:

cost of the substitution

deletionScore

public int deletionScore(T p_sourceTuv)

Calculate the cost of deletion of source segment.

Specified by:

deletionScore in interface AlignmentScorer<T>

Parameters:

p_sourceTuv - Source TUV. Source is in X sequence in the DP map.

Returns:

cost of the deletion

insertionScore

public int insertionScore(T p_targetTuv)

Calculate the cost of insertion of target segment.

Specified by:

insertionScore in interface AlignmentScorer<T>

Parameters:

p_targetTuv - Target TUV. Target is in Y sequence in the DP map.

Returns:

cost of the insertion

contractionScore

public int contractionScore(T p_sourceTuv1,
                            T p_sourceTuv2,
                            T p_targetTuv)

Calculate the cost of contracting two source segments to one target segment.

Specified by:

contractionScore in interface AlignmentScorer<T>

Parameters:

p_sourceTuv1 - Source TUV1. Source is in X sequence in the DP map.

p_sourceTuv2 - Source TUV2. Source is in X sequence in the DP map.

p_targetTuv - Target TUV. Target is in Y sequence in the DP map.

Returns:

cost of the contraction

expansionScore

public int expansionScore(T p_sourceTuv,
                          T p_targetTuv1,
                          T p_targetTuv2)

Calculate the cost of expanding one source segment to two target segments.

Specified by:

expansionScore in interface AlignmentScorer<T>

Parameters:

p_sourceTuv - Source TUV. Source is in X sequence in the DP map.

p_targetTuv1 - Target TUV1. Target is in Y sequence in the DP map.

p_targetTuv2 - Target TUV2. Target is in Y sequence in the DP map.

Returns:

cost of the expansion

meldingScore

public int meldingScore(T p_sourceTuv1,
                        T p_sourceTuv2,
                        T p_targetTuv1,
                        T p_targetTuv2)

Calculate the cost of melding of two source segments to two target segments.

Specified by:

meldingScore in interface AlignmentScorer<T>

Parameters:

p_sourceTuv1 - Source TUV1. Source is in X sequence in the DP map.

p_sourceTuv2 - Source TUV2. Source is in X sequence in the DP map.

p_targetTuv1 - Target TUV1. Target is in Y sequence in the DP map.

p_targetTuv2 - Target TUV2. Target is in Y sequence in the DP map.

Returns:

cost of the melding

match

public int match(int len1,
                 int len2)

Return -100 * log probability that an source sentence of length len1 is a translation of a foreign sentence of length len2. The probability is based on two parameters, the mean and variance of number of foreign characters per source character. Gale and Church hardcoded foreign_chars_per_eng_char as 1. It apparently works OK for European language alignment. We take the coefficient as a parameter so that non European languages can be aligned as well.

prob

public double prob(int len1,
                   int len2)

Return the probability that an source sentence of length len1 is a translation of a foreign sentence of length len2. The probability is based on two parameters, the mean and variance of number of foreign characters per source character. Gale and Church hardcoded foreign_chars_per_eng_char as 1. It apparently works OK for European language alignment. We take the coefficient as a parameter so that non European languages can be aligned as well.