Package com.acumenvelocity.ath.mt.confidence

Class HeuristicQualityEstimator

java.lang.Object

com.acumenvelocity.ath.mt.confidence.HeuristicQualityEstimator

All Implemented Interfaces:

AutoCloseable

public class HeuristicQualityEstimator
extends Object
implements AutoCloseable

Heuristic Quality Estimator - Fallback for unsupported language pairs Uses reference-free metrics that work for any language

Constructor Summary

Constructors

Constructor	Description
HeuristicQualityEstimator(String projectId, String projectLocation)

Method Summary

Modifier and Type	Method	Description
double	calculateBackTranslationScore(String source, String translation, String sourceLang, String targetLang)	Back-translation: translate back to source and compare (private helper for legacy method only) The implementation is now redundant as it's part of the inline logic above, but we keep the method signature for compatibility if it was called externally.
static double	calculateLengthRatio(String source, String translation)	Calculate length ratio score (closer to 1.0 is better)
static double	calculateTextSimilarity(String text1, String text2)	Calculate text similarity using character n-gram overlap (Jaccard)
void	close()
List<HeuristicDataStructures.HeuristicResult>	evaluateBatch(List<CommonDataStructures.SegmentTranslations> translations, String sourceLang, String targetLang)	Evaluate translations using heuristic methods (OPTIMIZED - batched back-translation)
HeuristicDataStructures.HeuristicScore	evaluateSingle(String sourceText, net.sf.okapi.common.resource.TextFragment sourceTf, String targetText, String srcLang, String tgtLang)
HeuristicDataStructures.HeuristicScore	evaluateSingleTranslation(String source, String translation, String modelId, String sourceLang, String targetLang)	Evaluate single translation (legacy method, kept for single-translation use cases)
static Set<String>	extractNGrams(String text, int n)	Extract character n-grams

Methods inherited from class java.lang.Object

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

Constructor Detail

HeuristicQualityEstimator

public HeuristicQualityEstimator(String projectId,
                                 String projectLocation)
                          throws IOException

Throws:

IOException

Method Detail

evaluateBatch

public List<HeuristicDataStructures.HeuristicResult> evaluateBatch(List<CommonDataStructures.SegmentTranslations> translations,
                                                                   String sourceLang,
                                                                   String targetLang)
                                                            throws IOException

Evaluate translations using heuristic methods (OPTIMIZED - batched back-translation)

Throws:

IOException

evaluateSingleTranslation

public HeuristicDataStructures.HeuristicScore evaluateSingleTranslation(String source,
                                                                        String translation,
                                                                        String modelId,
                                                                        String sourceLang,
                                                                        String targetLang)
                                                                 throws IOException

Evaluate single translation (legacy method, kept for single-translation use cases)

Throws:

IOException

calculateBackTranslationScore

public double calculateBackTranslationScore(String source,
                                            String translation,
                                            String sourceLang,
                                            String targetLang)
                                     throws IOException

Back-translation: translate back to source and compare (private helper for legacy method only) The implementation is now redundant as it's part of the inline logic above, but we keep the method signature for compatibility if it was called externally.

Throws:

IOException

calculateTextSimilarity

public static double calculateTextSimilarity(String text1,
                                             String text2)

Calculate text similarity using character n-gram overlap (Jaccard)

extractNGrams

public static Set<String> extractNGrams(String text,
                                        int n)

Extract character n-grams

calculateLengthRatio

public static double calculateLengthRatio(String source,
                                          String translation)

Calculate length ratio score (closer to 1.0 is better)

close

public void close()
           throws Exception

Specified by:

close in interface AutoCloseable

Throws:

Exception

evaluateSingle

public HeuristicDataStructures.HeuristicScore evaluateSingle(String sourceText,
                                                             net.sf.okapi.common.resource.TextFragment sourceTf,
                                                             String targetText,
                                                             String srcLang,
                                                             String tgtLang)
                                                      throws IOException

Throws:

IOException