opennlp.tools.sentdetect.SentenceDetectorME Maven / Gradle / Ivy
/*
* Licensed to the Apache Software Foundation (ASF) under one or more
* contributor license agreements. See the NOTICE file distributed with
* this work for additional information regarding copyright ownership.
* The ASF licenses this file to You under the Apache License, Version 2.0
* (the "License"); you may not use this file except in compliance with
* the License. You may obtain a copy of the License at
*
* http://www.apache.org/licenses/LICENSE-2.0
*
* Unless required by applicable law or agreed to in writing, software
* distributed under the License is distributed on an "AS IS" BASIS,
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
* See the License for the specific language governing permissions and
* limitations under the License.
*/
package opennlp.tools.sentdetect;
import java.io.File;
import java.io.FileInputStream;
import java.io.FileOutputStream;
import java.io.IOException;
import java.io.InputStreamReader;
import java.util.ArrayList;
import java.util.HashMap;
import java.util.List;
import java.util.Map;
import opennlp.maxent.GIS;
import opennlp.maxent.GISModel;
import opennlp.model.EventStream;
import opennlp.model.MaxentModel;
import opennlp.tools.dictionary.Dictionary;
import opennlp.tools.sentdetect.lang.Factory;
import opennlp.tools.util.HashSumEventStream;
import opennlp.tools.util.ObjectStream;
import opennlp.tools.util.PlainTextByLineStream;
import opennlp.tools.util.Span;
import opennlp.tools.util.StringUtil;
import opennlp.tools.util.model.BaseModel;
import opennlp.tools.util.model.ModelUtil;
/**
* A sentence detector for splitting up raw text into sentences.
*
* A maximum entropy model is used to evaluate the characters ".", "!", and "?" in a
* string to determine if they signify the end of a sentence.
*/
public class SentenceDetectorME implements SentenceDetector {
/**
* Constant indicates a sentence split.
*/
public static final String SPLIT ="s";
/**
* Constant indicates no sentence split.
*/
public static final String NO_SPLIT ="n";
private static final Double ONE = new Double(1);
/**
* The maximum entropy model to use to evaluate contexts.
*/
private MaxentModel model;
/**
* The feature context generator.
*/
private final SDContextGenerator cgen;
/**
* The {@link EndOfSentenceScanner} to use when scanning for end of sentence offsets.
*/
private final EndOfSentenceScanner scanner;
/**
* The list of probabilities associated with each decision.
*/
private List sentProbs = new ArrayList();
protected boolean useTokenEnd;
/**
* Initializes the current instance.
*
* @param model the {@link SentenceModel}
*/
public SentenceDetectorME(SentenceModel model) {
this(model, new Factory());
}
public SentenceDetectorME(SentenceModel model, Factory factory) {
this.model = model.getMaxentModel();
cgen = factory.createSentenceContextGenerator(model.getLanguage());
scanner = factory.createEndOfSentenceScanner(model.getLanguage());
useTokenEnd = model.useTokenEnd();
}
/**
* Detect sentences in a String.
*
* @param s The string to be processed.
*
* @return A string array containing individual sentences as elements.
*/
public String[] sentDetect(String s) {
Span[] spans = sentPosDetect(s);
String sentences[];
if (spans.length != 0) {
sentences = new String[spans.length];
for (int si = 0; si < spans.length; si++) {
sentences[si] = spans[si].getCoveredText(s).toString();
}
}
else {
sentences = new String[] {};
}
return sentences;
}
private int getFirstWS(String s, int pos) {
while (pos < s.length() && !StringUtil.isWhitespace(s.charAt(pos)))
pos++;
return pos;
}
private int getFirstNonWS(String s, int pos) {
while (pos < s.length() && StringUtil.isWhitespace(s.charAt(pos)))
pos++;
return pos;
}
/**
* Detect the position of the first words of sentences in a String.
*
* @param s The string to be processed.
* @return A integer array containing the positions of the end index of
* every sentence
*
*/
public Span[] sentPosDetect(String s) {
double sentProb = 1;
sentProbs.clear();
StringBuffer sb = new StringBuffer(s);
List enders = scanner.getPositions(s);
List positions = new ArrayList(enders.size());
for (int i = 0, end = enders.size(), index = 0; i < end; i++) {
Integer candidate = enders.get(i);
int cint = candidate;
// skip over the leading parts of non-token final delimiters
int fws = getFirstWS(s,cint + 1);
if (i + 1 < end && enders.get(i + 1) < fws) {
continue;
}
double[] probs = model.eval(cgen.getContext(sb, cint));
String bestOutcome = model.getBestOutcome(probs);
sentProb *= probs[model.getIndex(bestOutcome)];
if (bestOutcome.equals(SPLIT) && isAcceptableBreak(s, index, cint)) {
if (index != cint) {
if (useTokenEnd) {
positions.add(getFirstNonWS(s, getFirstWS(s,cint + 1)));
}
else {
positions.add(getFirstNonWS(s,cint));
}
sentProbs.add(new Double(probs[model.getIndex(bestOutcome)]));
}
index = cint + 1;
}
}
int[] starts = new int[positions.size()];
for (int i = 0; i < starts.length; i++) {
starts[i] = positions.get(i);
}
// string does not contain sentence end positions
if (starts.length == 0) {
// remove leading and trailing whitespace
int start = 0;
int end = s.length();
while (start < s.length() && Character.isWhitespace(s.charAt(start)))
start++;
while (end > 0 && Character.isWhitespace(s.charAt(end - 1)))
end--;
if ((end - start) > 0)
return new Span[] {new Span(start, end)};
else
return new Span[0];
}
// Now convert the sent indexes to spans
boolean leftover = starts[starts.length - 1] != s.length();
Span[] spans = new Span[leftover? starts.length + 1 : starts.length];
for (int si=0;si 0 && Character.isWhitespace(s.charAt(end-1))) {
end--;
}
spans[si]=new Span(start,end);
}
if (leftover) {
spans[spans.length-1] = new Span(starts[starts.length-1],s.length());
sentProbs.add(ONE);
}
return spans;
}
/**
* Returns the probabilities associated with the most recent
* calls to sentDetect().
*
* @return probability for each sentence returned for the most recent
* call to sentDetect. If not applicable an empty array is
* returned.
*/
public double[] getSentenceProbabilities() {
double[] sentProbArray = new double[sentProbs.size()];
for (int i = 0; i < sentProbArray.length; i++) {
sentProbArray[i] = ((Double) sentProbs.get(i)).doubleValue();
}
return sentProbArray;
}
/**
* Allows subclasses to check an overzealous (read: poorly
* trained) model from flagging obvious non-breaks as breaks based
* on some boolean determination of a break's acceptability.
*
* The implementation here always returns true, which means
* that the MaxentModel's outcome is taken as is.
*
* @param s the string in which the break occurred.
* @param fromIndex the start of the segment currently being evaluated
* @param candidateIndex the index of the candidate sentence ending
* @return true if the break is acceptable
*/
protected boolean isAcceptableBreak(String s, int fromIndex, int candidateIndex) {
return true;
}
public static SentenceModel train(String languageCode, ObjectStream samples,
boolean useTokenEnd, Dictionary abbreviations) throws IOException {
return train(languageCode, samples, useTokenEnd, abbreviations,5,100);
}
public static SentenceModel train(String languageCode, ObjectStream samples,
boolean useTokenEnd, Dictionary abbreviations, int cutoff, int iterations) throws IOException {
Map manifestInfoEntries = new HashMap();
ModelUtil.addCutoffAndIterations(manifestInfoEntries, cutoff, iterations);
Factory factory = new Factory();
// TODO: Fix the EventStream to throw exceptions when training goes wrong
EventStream eventStream = new SDEventStream(samples,
factory.createSentenceContextGenerator(languageCode),
factory.createEndOfSentenceScanner(languageCode));
HashSumEventStream hses = new HashSumEventStream(eventStream);
GISModel sentModel = GIS.trainModel(hses, iterations, cutoff);
manifestInfoEntries.put(BaseModel.TRAINING_EVENTHASH_PROPERTY,
hses.calculateHashSum().toString(16));
return new SentenceModel(languageCode, sentModel,
useTokenEnd, abbreviations, manifestInfoEntries);
}
private static void usage() {
System.err.println("Usage: SentenceDetectorME -encoding charset -lang language trainData modelName [cutoff iterations]");
System.err.println("-encoding charset specifies the encoding which should be used ");
System.err.println(" for reading and writing text.");
System.err.println("-lang language specifies the language which ");
System.err.println(" is being processed.");
System.err.println("trainData specifies the name of the input training file");
System.err.println(" to train the resulting model.");
System.err.println("modelName specifies the resulting saved model after");
System.err.println(" training.");
System.exit(1);
}
/**
* Trains a new sentence detection model.
*
* Usage: opennlp.tools.sentdetect.SentenceDetectorME data_file new_model_name (iterations cutoff)?
*
* @param args
* @throws IOException
*/
public static void main(String[] args) throws IOException {
int ai=0;
String encoding = null;
String lang = null;
if (args.length == 0) {
usage();
}
while (args[ai].startsWith("-")) {
if (args[ai].equals("-encoding")) {
ai++;
if (ai < args.length) {
encoding = args[ai];
ai++;
}
else {
usage();
}
}
else if (args[ai].equals("-lang")) {
ai++;
if (ai < args.length) {
lang = args[ai];
ai++;
}
else {
usage();
}
}
else {
usage();
}
}
File inFile = new File(args[ai++]);
File outFile = new File(args[ai++]);
try {
if ((lang == null) || (encoding == null)) {
usage();
}
SentenceModel model = train(lang, new SentenceSampleStream(new PlainTextByLineStream(
new InputStreamReader(new FileInputStream(inFile), encoding))), true, null);
// TODO: add support for iterations and cutoff settings
// if (args.length > ai)
// mod = train(es, Integer.parseInt(args[ai++]), Integer.parseInt(args[ai++]));
// else
// mod = train(es, 100, 5);
System.out.println("Saving the model as: " + outFile);
model.serialize(new FileOutputStream(outFile));
}
catch (Exception e) {
e.printStackTrace();
}
}
}