edu.stanford.nlp.parser.lexparser.RerankingParserQuery Maven / Gradle / Ivy

Show more of this group Show more artifacts with this name
Show all versions of stanford-corenlp Show documentation

Stanford CoreNLP provides a set of natural language analysis tools which can take raw English language text input and give the base forms of words, their parts of speech, whether they are names of companies, people, etc., normalize dates, times, and numeric quantities, mark up the structure of sentences in terms of phrases and word dependencies, and indicate which noun phrases refer to the same entities. It provides the foundational building blocks for higher level text understanding applications.

There is a newer version: 4.5.7

Show newest version

package edu.stanford.nlp.parser.lexparser;

import java.io.PrintWriter;
import java.util.ArrayList;
import java.util.Collections;
import java.util.List;

import edu.stanford.nlp.ling.HasWord;
import edu.stanford.nlp.parser.KBestViterbiParser;
import edu.stanford.nlp.parser.common.NoSuchParseException;
import edu.stanford.nlp.parser.common.ParserConstraint;
import edu.stanford.nlp.parser.common.ParserQuery;
import edu.stanford.nlp.trees.Tree;
import edu.stanford.nlp.util.Generics;
import edu.stanford.nlp.util.ScoredComparator;
import edu.stanford.nlp.util.ScoredObject;

/**
 * Rerank trees from the ParserQuery based on scores from a Reranker.
 * 

 * TODO: should handle Factored parsers as well
 *
 * @author John Bauer
 */
public class RerankingParserQuery implements ParserQuery {
  private final Options op;
  private final ParserQuery parserQuery;
  private final Reranker reranker;
  private final int rerankerKBest;

  private List> scoredTrees;

  /**
   * Data for this particular query stored by the Reranker will be
   * stored in this object
   */
  private RerankerQuery rerankerQuery;

  public RerankingParserQuery(Options op, ParserQuery parserQuery, Reranker reranker) {
    this.op = op;
    this.parserQuery = parserQuery;
    this.reranker = reranker;
    this.rerankerKBest = op.rerankerKBest;
  }

  @Override
  public boolean saidMemMessage() {
    return parserQuery.saidMemMessage();
  }

  @Override
  public void setConstraints(List constraints) {
    parserQuery.setConstraints(constraints);
  }

  @Override
  public boolean parse(List sentence) {
    boolean success = parserQuery.parse(sentence);
    if (!success) {
      return false;
    }

    List> bestKParses = parserQuery.getKBestPCFGParses(rerankerKBest);
    if (bestKParses.isEmpty()) {
      return false;
    }
    scoredTrees = rerank(sentence, bestKParses);
    return true;
  }

  @Override
  public boolean parseAndReport(List sentence, PrintWriter pwErr) {
    boolean success = parserQuery.parseAndReport(sentence, pwErr);
    if (!success) {
      return false;
    }

    List> bestKParses = parserQuery.getKBestPCFGParses(rerankerKBest);
    if (bestKParses.isEmpty()) {
      return false;
    }
    scoredTrees = rerank(sentence, bestKParses);
    return true;
  }

  List> rerank(List sentence, List> bestKParses) {
    this.rerankerQuery = reranker.process(sentence);

    List> reranked = new ArrayList<>();
    for (ScoredObject scoredTree : bestKParses) {
      double score = scoredTree.score();
      try {
        score = op.baseParserWeight * score + rerankerQuery.score(scoredTree.object());
      } catch (NoSuchParseException e) {
        score = Double.NEGATIVE_INFINITY;
      }
      reranked.add(new ScoredObject<>(scoredTree.object(), score));
    }
    Collections.sort(reranked, ScoredComparator.DESCENDING_COMPARATOR);
    return reranked;
  }

  @Override
  public Tree getBestParse() {
    if (scoredTrees == null || scoredTrees.isEmpty()) {
      return null;
    }
    return scoredTrees.get(0).object();
  }

  @Override
  public List> getKBestParses(int k) {
    return this.getKBestPCFGParses(k);
  }

  @Override
  public double getBestScore() {
    return this.getPCFGScore();
  }

  @Override
  public Tree getBestPCFGParse() {
    return getBestParse();
  }

  @Override
  public double getPCFGScore() {
    if (scoredTrees == null || scoredTrees.isEmpty()) {
      throw new AssertionError();
    }
    return scoredTrees.get(0).score();
  }

  @Override
  public Tree getBestDependencyParse(boolean debinarize) {
    // TODO: barf?
    return null;
  }

  @Override
  public Tree getBestFactoredParse() {
    // TODO: barf?
    return null;
  }

  @Override
  public List> getBestPCFGParses() {
    if (scoredTrees == null || scoredTrees.isEmpty()) {
      throw new AssertionError();
    }
    List> equalTrees = Generics.newArrayList();
    double score = scoredTrees.get(0).score();
    int treePos = 0;
    while (treePos < scoredTrees.size() && scoredTrees.get(treePos).score() == score) {
      equalTrees.add(scoredTrees.get(treePos));
    }
    return equalTrees;
  }

  @Override
  public void restoreOriginalWords(Tree tree) {
    parserQuery.restoreOriginalWords(tree);
  }

  @Override
  public boolean hasFactoredParse() {
    return false;
  }

  @Override
  public List> getKBestPCFGParses(int kbestPCFG) {
    List> trees = Generics.newArrayList();
    for (int treePos = 0; treePos < scoredTrees.size() && treePos < kbestPCFG; ++treePos) {
      trees.add(scoredTrees.get(treePos));
    }
    return trees;
  }

  @Override
  public List> getKGoodFactoredParses(int kbest) {
    // TODO: barf?
    return null;
  }

  @Override
  public KBestViterbiParser getPCFGParser() {
    return null;
  }

  @Override
  public KBestViterbiParser getFactoredParser() {
    return null;
  }

  @Override
  public KBestViterbiParser getDependencyParser() {
    return null;
  }


  /**
   * Parsing succeeded without any horrible errors or fallback
   */
  @Override
  public boolean parseSucceeded() {
    return parserQuery.parseSucceeded();
  }

  /**
   * The sentence was skipped, probably because it was too long or of length 0
   */
  @Override
  public boolean parseSkipped() {
    return parserQuery.parseSkipped();
  }

  /**
   * The model had to fall back to a simpler model on the previous parse
   */
  @Override
  public boolean parseFallback() {
    return parserQuery.parseFallback();
  }

  /**
   * The model ran out of memory on the most recent parse
   */
  @Override
  public boolean parseNoMemory() {
    return parserQuery.parseNoMemory();
  }

  /**
   * The model could not parse the most recent sentence for some reason
   */
  @Override
  public boolean parseUnparsable() {
    return parserQuery.parseUnparsable();
  }

  @Override
  public List originalSentence() {
    return parserQuery.originalSentence();
  }

  public RerankerQuery rerankerQuery() {
    return rerankerQuery;
  }
}