001/*
002 * Copyright (c) 2011, Regents of the University of Colorado 
003 * All rights reserved.
004 * 
005 * Redistribution and use in source and binary forms, with or without
006 * modification, are permitted provided that the following conditions are met:
007 * 
008 * Redistributions of source code must retain the above copyright notice, this list of conditions and the following disclaimer. 
009 * Redistributions in binary form must reproduce the above copyright notice, this list of conditions and the following disclaimer in the documentation and/or other materials provided with the distribution. 
010 * Neither the name of the University of Colorado at Boulder nor the names of its contributors may be used to endorse or promote products derived from this software without specific prior written permission. 
011 * 
012 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
013 * AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
014 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
015 * ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE
016 * LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
017 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
018 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
019 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
020 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
021 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
022 * POSSIBILITY OF SUCH DAMAGE. 
023 */
024package org.cleartk.timeml.time;
025
026import java.util.ArrayList;
027import java.util.List;
028
029import org.apache.uima.UimaContext;
030import org.apache.uima.analysis_engine.AnalysisEngineDescription;
031import org.apache.uima.analysis_engine.AnalysisEngineProcessException;
032import org.apache.uima.jcas.JCas;
033import org.apache.uima.resource.ResourceInitializationException;
034import org.cleartk.ml.CleartkSequenceAnnotator;
035import org.cleartk.ml.Feature;
036import org.cleartk.ml.Instances;
037import org.cleartk.ml.chunking.BioChunking;
038import org.cleartk.ml.feature.extractor.CleartkExtractor;
039import org.cleartk.ml.feature.extractor.CoveredTextExtractor;
040import org.cleartk.ml.feature.extractor.FeatureExtractor1;
041import org.cleartk.ml.feature.extractor.NamedFeatureExtractor1;
042import org.cleartk.ml.feature.extractor.TypePathExtractor;
043import org.cleartk.ml.feature.extractor.CleartkExtractor.Following;
044import org.cleartk.ml.feature.extractor.CleartkExtractor.Preceding;
045import org.cleartk.ml.feature.function.CharacterCategoryPatternFunction;
046import org.cleartk.ml.liblinear.LibLinearStringOutcomeDataWriter;
047import org.cleartk.timeml.type.Time;
048import org.cleartk.timeml.util.CleartkInternalModelFactory;
049import org.cleartk.timeml.util.TimeWordsExtractor;
050import org.cleartk.token.type.Sentence;
051import org.cleartk.token.type.Token;
052import org.apache.uima.fit.factory.AnalysisEngineFactory;
053import org.apache.uima.fit.util.JCasUtil;
054
055import com.google.common.collect.Lists;
056
057/**
058 * <br>
059 * Copyright (c) 2011, Regents of the University of Colorado <br>
060 * All rights reserved.
061 * 
062 * @author Steven Bethard
063 */
064public class TimeAnnotator extends CleartkSequenceAnnotator<String> {
065
066  public static final CleartkInternalModelFactory FACTORY = new CleartkInternalModelFactory() {
067
068    @Override
069    public Class<?> getAnnotatorClass() {
070      return TimeAnnotator.class;
071    }
072
073    @Override
074    public Class<?> getDataWriterClass() {
075      return LibLinearStringOutcomeDataWriter.class;
076    }
077
078    @Override
079    public AnalysisEngineDescription getBaseDescription() throws ResourceInitializationException {
080      return AnalysisEngineFactory.createEngineDescription(TimeAnnotator.class);
081    }
082  };
083
084  private List<NamedFeatureExtractor1<Token>> tokenFeatureExtractors;
085
086  private List<CleartkExtractor<Token, Token>> contextFeatureExtractors;
087
088  private BioChunking<Token, Time> chunking;
089
090  @Override
091  public void initialize(UimaContext context) throws ResourceInitializationException {
092    super.initialize(context);
093
094    // define chunking type
095    this.chunking = new BioChunking<Token, Time>(Token.class, Time.class);
096
097    // add features: word, character pattern, stem, pos
098    this.tokenFeatureExtractors = Lists.newArrayList();
099    this.tokenFeatureExtractors.add(new CoveredTextExtractor<Token>());
100    NamedFeatureExtractor1<Token> ex = CharacterCategoryPatternFunction.createExtractor();
101    this.tokenFeatureExtractors.add(ex);
102    this.tokenFeatureExtractors.add(new TimeWordsExtractor<Token>());
103    this.tokenFeatureExtractors.add(new TypePathExtractor<Token>(Token.class, "stem"));
104    this.tokenFeatureExtractors.add(new TypePathExtractor<Token>(Token.class, "pos"));
105        
106    // add window of features before and after
107    this.contextFeatureExtractors = Lists.newArrayList();
108    for (FeatureExtractor1<Token> extractor : this.tokenFeatureExtractors) {
109      this.contextFeatureExtractors.add(new CleartkExtractor<Token, Token>(Token.class, extractor, new Preceding(
110          3), new Following(3)));
111    }
112  }
113
114  @Override
115  public void process(JCas jCas) throws AnalysisEngineProcessException {
116
117    // classify tokens within each sentence
118    for (Sentence sentence : JCasUtil.select(jCas, Sentence.class)) {
119      List<Token> tokens = JCasUtil.selectCovered(jCas, Token.class, sentence);
120
121      // extract features for all tokens
122      List<List<Feature>> featureLists = new ArrayList<List<Feature>>();
123      for (Token token : tokens) {
124        List<Feature> features = new ArrayList<Feature>();
125        for (FeatureExtractor1<Token> extractor : this.tokenFeatureExtractors) {
126          features.addAll(extractor.extract(jCas, token));
127        }
128        for (CleartkExtractor<Token, Token> extractor : this.contextFeatureExtractors) {
129          features.addAll(extractor.extractWithin(jCas, token, sentence));
130        }
131        featureLists.add(features);
132      }
133
134      // during training, convert Times to chunk labels and write the training Instances
135      if (this.isTraining()) {
136        List<Time> times = JCasUtil.selectCovered(jCas, Time.class, sentence);
137        List<String> outcomes = this.chunking.createOutcomes(jCas, tokens, times);
138        this.dataWriter.write(Instances.toInstances(outcomes, featureLists));
139      }
140
141      // during prediction, convert chunk labels to Times and add them to the CAS
142      else {
143        List<String> outcomes = this.classifier.classify(featureLists);
144        this.chunking.createChunks(jCas, tokens, outcomes);
145      }
146    }
147
148    // add IDs to all Times
149    int timeIndex = 1;
150    for (Time time : JCasUtil.select(jCas, Time.class)) {
151      if (time.getId() == null) {
152        String id = String.format("t%d", timeIndex);
153        time.setId(id);
154        timeIndex += 1;
155      }
156    }
157  }
158}