private void DoOneSentence <Token>(IList <TOKEN> words)
where Token : CoreLabel
{
QuantifiableEntityNormalizer.AddNormalizedQuantitiesToEntities(words, collapse);
}
public override IList <CoreLabel> ClassifyWithGlobalInformation(IList <CoreLabel> tokens, ICoreMap document, ICoreMap sentence)
{
IList <CoreLabel> output = base.Classify(tokens);
if (applyNumericClassifiers)
{
try
{
// recognizes additional MONEY, TIME, DATE, and NUMBER using a set of deterministic rules
// note: some DATE and TIME entities are recognized by our statistical NER based on MUC
// note: this includes SUTime
// note: requires TextAnnotation, PartOfSpeechTagAnnotation, and AnswerAnnotation
// note: this sets AnswerAnnotation!
RecognizeNumberSequences(output, document, sentence);
}
catch (RuntimeInterruptedException e)
{
throw;
}
catch (Exception e)
{
log.Info("Ignored an exception in NumberSequenceClassifier: (result is that some numbers were not classified)");
log.Info("Tokens: " + StringUtils.JoinWords(tokens, " "));
Sharpen.Runtime.PrintStackTrace(e, System.Console.Error);
}
// AnswerAnnotation -> NERAnnotation
CopyAnswerFieldsToNERField(output);
try
{
// normalizes numeric entities such as MONEY, TIME, DATE, or PERCENT
// note: this uses and sets NamedEntityTagAnnotation!
if (nerLanguage == NERClassifierCombiner.Language.Chinese)
{
// For chinese there is no support for SUTime by default
// We need to hand in document and sentence for Chinese to handle DocDate; however, since English normalization
// is handled by SUTime, and the information is passed in recognizeNumberSequences(), English only need output.
ChineseQuantifiableEntityNormalizer.AddNormalizedQuantitiesToEntities(output, document, sentence);
}
else
{
QuantifiableEntityNormalizer.AddNormalizedQuantitiesToEntities(output, false, useSUTime);
}
}
catch (Exception e)
{
log.Info("Ignored an exception in QuantifiableEntityNormalizer: (result is that entities were not normalized)");
log.Info("Tokens: " + StringUtils.JoinWords(tokens, " "));
Sharpen.Runtime.PrintStackTrace(e, System.Console.Error);
}
catch (AssertionError e)
{
log.Info("Ignored an assertion in QuantifiableEntityNormalizer: (result is that entities were not normalized)");
log.Info("Tokens: " + StringUtils.JoinWords(tokens, " "));
Sharpen.Runtime.PrintStackTrace(e, System.Console.Error);
}
}
else
{
// AnswerAnnotation -> NERAnnotation
CopyAnswerFieldsToNERField(output);
}
// Apply RegexNER annotations
// cdm 2016: Used to say and do "// skip first token" but I couldn't understand why, so I removed that.
foreach (CoreLabel token in tokens)
{
// System.out.println(token.toShorterString());
if ((token.Tag() == null || token.Tag()[0] == 'N') && "O".Equals(token.Ner()) || "MISC".Equals(token.Ner()))
{
string target = gazetteMapping[token.OriginalText()];
if (target != null)
{
token.SetNER(target);
}
}
}
// Return
return(output);
}