opennlp.tools.parser.AbstractContextGenerator 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.parser;
import java.util.Collection;
import java.util.List;
import java.util.Set;
/**
* Abstract class containing many of the methods used to generate contexts for parsing.
*/
public abstract class AbstractContextGenerator {
protected static final String EOS = "eos";
protected boolean zeroBackOff;
/** Set of punctuation to be used in generating features. */
protected Set punctSet;
protected boolean useLabel;
/**
* Creates punctuation feature for the specified punctuation at the specified index
* {@code i} based on the {@code punctuation} mark.
*
* @param punct The punctuation which is in context.
* @param i The index of the {@code punctuation} with relative to the parse.
* @return Punctuation feature for the parse and the punctuation at the specified index.
*/
protected String punct(Parse punct, int i) {
return i + "=" + punct.getCoveredText();
}
/**
* Creates punctuation feature for the specified punctuation at the specified index
* {@code i} based on the {@code punctuation}'s tag.
*
* @param punct The punctuation which is in context.
* @param i The index of the {@code punctuation} relative to the parse.
* @return Punctuation feature for the parse and the punctuation at the specified index.
*/
protected String punctbo(Parse punct, int i) {
return i + "=" + punct.getType();
}
protected String cons(Parse p, int i) {
StringBuilder feat = new StringBuilder(20);
feat.append(i).append("=");
if (p != null) {
if (useLabel && i < 0) {
feat.append(p.getLabel()).append("|");
}
feat.append(p.getType()).append("|").append(p.getHead().getCoveredText());
}
else {
feat.append(EOS);
}
return feat.toString();
}
protected String consbo(Parse p, int i) { //cons back-off
StringBuilder feat = new StringBuilder(20);
feat.append(i).append("*=");
if (p != null) {
if (useLabel && i < 0) {
feat.append(p.getLabel()).append("|");
}
feat.append(p.getType());
}
else {
feat.append(EOS);
}
return feat.toString();
}
/**
* Generates a string representing the grammar rule production that the specified parse
* is starting.
*
* The rule is of the form {@code p.type -> c.children[0..n].type}.
*
* @param p The {@link Parse} which stats the production.
* @param includePunctuation Whether punctuation should be included in the production.
* @return A string representing the grammar rule production that the specified parse
* is starting.
*/
protected String production(Parse p, boolean includePunctuation) {
StringBuilder production = new StringBuilder(20);
production.append(p.getType()).append("->");
Parse[] children = AbstractBottomUpParser.collapsePunctuation(p.getChildren(),punctSet);
for (int ci = 0; ci < children.length; ci++) {
production.append(children[ci].getType());
if (ci + 1 != children.length) {
production.append(",");
Collection nextPunct = children[ci].getNextPunctuationSet();
if (includePunctuation && nextPunct != null) {
//TODO: make sure multiple punctuation comes out the same
for (Parse punct : nextPunct) {
production.append(punct.getType()).append(",");
}
}
}
}
return production.toString();
}
protected void cons2(List features, Cons c0, Cons c1, Collection punct1s, boolean bigram) {
if (punct1s != null) {
for (Parse p : punct1s) {
String punctbo = punctbo(p, c1.index <= 0 ? c1.index - 1 : c1.index);
//punctbo(1);
features.add(punctbo);
if (c0.index == 0) { //TODO look at removing case
//cons(0)punctbo(1)
if (c0.unigram) features.add(c0.cons + "," + punctbo);
features.add(c0.consbo + "," + punctbo);
}
if (c1.index == 0) { //TODO look at removing case
//punctbo(1)cons(1)
if (c1.unigram) features.add(punctbo + "," + c1.cons);
features.add(punctbo + "," + c1.consbo);
}
//cons(0)punctbo(1)cons(1)
if (bigram) features.add(c0.cons + "," + punctbo + "," + c1.cons);
if (c1.unigram) features.add(c0.consbo + "," + punctbo + "," + c1.cons);
if (c0.unigram) features.add(c0.cons + "," + punctbo + "," + c1.consbo);
features.add(c0.consbo + "," + punctbo + "," + c1.consbo);
}
}
else {
//cons(0),cons(1)
if (bigram) features.add(c0.cons + "," + c1.cons);
if (c1.unigram) features.add(c0.consbo + "," + c1.cons);
if (c0.unigram) features.add(c0.cons + "," + c1.consbo);
features.add(c0.consbo + "," + c1.consbo);
}
}
/**
* Creates cons features involving the 3 specified nodes and adds them to the specified feature list.
*
* @param features The list of features.
* @param c0 The first {@link Cons node}.
* @param c1 The second {@link Cons node}.
* @param c2 The third {@link Cons node}.
* @param punct1s The punctuation between {@code c0} and {@code c1}.
* @param punct2s The punctuation between {@code c1} and {@code c2}.
* @param trigram Specifies whether lexical tri-gram features between these nodes
* should be generated.
* @param bigram1 Specifies whether lexical bi-gram features between {@code c0} and {@code c1}
* should be generated.
* @param bigram2 Specifies whether lexical bi-gram features between {@code c1} and {@code c2}
* should be generated.
*/
protected void cons3(List features, Cons c0, Cons c1, Cons c2, Collection punct1s,
Collection punct2s, boolean trigram, boolean bigram1, boolean bigram2) {
// features.add("stage=cons(0),cons(1),cons(2)");
if (punct1s != null) {
if (c0.index == -2) {
for (Parse p : punct1s) {
// String punct = punct(p,c1.index);
String punctbo = punctbo(p, c1.index <= 0 ? c1.index - 1 : c1.index);
//punct(-2)
//TODO consider changing
//features.add(punct);
//punctbo(-2)
features.add(punctbo);
}
}
}
if (punct2s != null) {
if (c2.index == 2) {
for (Parse p : punct2s) {
// String punct = punct(p,c2.index);
String punctbo = punctbo(p, c2.index);
//punct(2)
//TODO consider changing
//features.add(punct);
//punctbo(2)
features.add(punctbo);
}
}
if (punct1s != null) {
//cons(0),punctbo(1),cons(1),punctbo(2),cons(2)
for (Parse punct2 : punct2s) {
String punctbo2 = punctbo(punct2, c2.index <= 0 ? c2.index - 1 : c2.index);
for (Parse punct1 : punct1s) {
String punctbo1 = punctbo(punct1, c1.index <= 0 ? c1.index - 1 : c1.index);
if (trigram)
features.add(c0.cons + "," + punctbo1 + "," + c1.cons + "," + punctbo2 + "," + c2.cons);
if (bigram2)
features.add(c0.consbo + "," + punctbo1 + "," + c1.cons + "," + punctbo2 + "," + c2.cons);
if (c0.unigram && c2.unigram)
features.add(c0.cons + "," + punctbo1 + "," + c1.consbo + "," + punctbo2 + "," + c2.cons);
if (bigram1)
features.add(c0.cons + "," + punctbo1 + "," + c1.cons + "," + punctbo2 + "," + c2.consbo);
if (c2.unigram)
features.add(c0.consbo + "," + punctbo1 + "," + c1.consbo + "," + punctbo2 + "," + c2.cons);
if (c1.unigram)
features.add(c0.consbo + "," + punctbo1 + "," + c1.cons + "," + punctbo2 + "," + c2.consbo);
if (c0.unigram)
features.add(c0.cons + "," + punctbo1 + "," + c1.consbo + "," + punctbo2 + "," + c2.consbo);
features.add(c0.consbo + "," + punctbo1 + "," + c1.consbo + "," + punctbo2 + "," + c2.consbo);
if (zeroBackOff) {
if (bigram1) features.add(c0.cons + "," + punctbo1 + "," + c1.cons + "," + punctbo2);
if (c1.unigram) features.add(c0.consbo + "," + punctbo1 + "," + c1.cons + "," + punctbo2);
if (c0.unigram) features.add(c0.cons + "," + punctbo1 + "," + c1.consbo + "," + punctbo2);
features.add(c0.consbo + "," + punctbo1 + "," + c1.consbo + "," + punctbo2);
}
}
}
}
else { //punct1s == null
//cons(0),cons(1),punctbo(2),cons(2)
for (Parse punct2 : punct2s) {
String punctbo2 = punctbo(punct2, c2.index <= 0 ? c2.index - 1 : c2.index);
if (trigram) features.add(c0.cons + "," + c1.cons + "," + punctbo2 + "," + c2.cons);
if (bigram2)
features.add(c0.consbo + "," + c1.cons + "," + punctbo2 + "," + c2.cons);
if (c0.unigram && c2.unigram)
features.add(c0.cons + "," + c1.consbo + "," + punctbo2 + "," + c2.cons);
if (bigram1) features.add(c0.cons + "," + c1.cons + "," + punctbo2 + "," + c2.consbo);
if (c2.unigram) features.add(c0.consbo + "," + c1.consbo + "," + punctbo2 + "," + c2.cons);
if (c1.unigram) features.add(c0.consbo + "," + c1.cons + "," + punctbo2 + "," + c2.consbo);
if (c0.unigram) features.add(c0.cons + "," + c1.consbo + "," + punctbo2 + "," + c2.consbo);
features.add(c0.consbo + "," + c1.consbo + "," + punctbo2 + "," + c2.consbo);
if (zeroBackOff) {
if (bigram1) features.add(c0.cons + "," + c1.cons + "," + punctbo2);
if (c1.unigram) features.add(c0.consbo + "," + c1.cons + "," + punctbo2);
if (c0.unigram) features.add(c0.cons + "," + c1.consbo + "," + punctbo2);
features.add(c0.consbo + "," + c1.consbo + "," + punctbo2);
}
}
}
}
else {
if (punct1s != null) {
//cons(0),punctbo(1),cons(1),cons(2)
for (Parse punct1 : punct1s) {
String punctbo1 = punctbo(punct1, c1.index <= 0 ? c1.index - 1 : c1.index);
if (trigram)
features.add(c0.cons + "," + punctbo1 + "," + c1.cons + "," + c2.cons);
if (bigram2)
features.add(c0.consbo + "," + punctbo1 + "," + c1.cons + "," + c2.cons);
if (c0.unigram && c2.unigram)
features.add(c0.cons + "," + punctbo1 + "," + c1.consbo + "," + c2.cons);
if (bigram1)
features.add(c0.cons + "," + punctbo1 + "," + c1.cons + "," + c2.consbo);
if (c2.unigram)
features.add(c0.consbo + "," + punctbo1 + "," + c1.consbo + "," + c2.cons);
if (c1.unigram)
features.add(c0.consbo + "," + punctbo1 + "," + c1.cons + "," + c2.consbo);
if (c0.unigram)
features.add(c0.cons + "," + punctbo1 + "," + c1.consbo + "," + c2.consbo);
features.add(c0.consbo + "," + punctbo1 + "," + c1.consbo + "," + c2.consbo);
//zero backoff case covered by cons(0)cons(1)
}
}
else {
//cons(0),cons(1),cons(2)
if (trigram) features.add(c0.cons + "," + c1.cons + "," + c2.cons);
if (bigram2) features.add(c0.consbo + "," + c1.cons + "," + c2.cons);
if (c0.unigram && c2.unigram) features.add(c0.cons + "," + c1.consbo + "," + c2.cons);
if (bigram1) features.add(c0.cons + "," + c1.cons + "," + c2.consbo);
if (c2.unigram) features.add(c0.consbo + "," + c1.consbo + "," + c2.cons);
if (c1.unigram) features.add(c0.consbo + "," + c1.cons + "," + c2.consbo);
if (c0.unigram) features.add(c0.cons + "," + c1.consbo + "," + c2.consbo);
features.add(c0.consbo + "," + c1.consbo + "," + c2.consbo);
}
}
}
/**
* Generates features for nodes surrounding a completed node of the specified {@code type}.
*
* @param node A surrounding {@link Parse node}.
* @param i The index of the surrounding {@code node} with respect to the completed node.
* @param type The type of the completed node.
* @param punctuation The punctuation adjacent and between the specified surrounding node.
* @param features A list to which features are added.
*/
protected void surround(Parse node, int i, String type, Collection punctuation,
List features) {
StringBuilder feat = new StringBuilder(20);
feat.append("s").append(i).append("=");
if (punctuation != null) {
for (Parse punct : punctuation) {
if (node != null) {
feat.append(node.getHead().getCoveredText()).append("|").append(type)
.append("|").append(node.getType()).append("|").append(punct.getType());
} else {
feat.append(type).append("|").append(EOS).append("|").append(punct.getType());
}
features.add(feat.toString());
feat.setLength(0);
feat.append("s").append(i).append("*=");
if (node != null) {
feat.append(type).append("|").append(node.getType()).append("|").append(punct.getType());
} else {
feat.append(type).append("|").append(EOS).append("|").append(punct.getType());
}
features.add(feat.toString());
feat.setLength(0);
feat.append("s").append(i).append("*=");
feat.append(type).append("|").append(punct.getType());
features.add(feat.toString());
}
}
else {
if (node != null) {
feat.append(node.getHead().getCoveredText()).append("|").append(type)
.append("|").append(node.getType());
}
else {
feat.append(type).append("|").append(EOS);
}
features.add(feat.toString());
feat.setLength(0);
feat.append("s").append(i).append("*=");
if (node != null) {
feat.append(type).append("|").append(node.getType());
}
else {
feat.append(type).append("|").append(EOS);
}
features.add(feat.toString());
}
}
/**
* Produces features to determine whether the specified child node is part of
* a complete constituent of the specified type and adds those features to the
* specified list.
*
* @param child The {@link Parse node} to consider.
* @param i A string indicating the position of the child node.
* @param type The type of constituent being built.
* @param features List to add features to.
*/
protected void checkcons(Parse child, String i, String type, List features) {
StringBuilder feat = new StringBuilder(20);
feat.append("c").append(i).append("=").append(child.getType()).append("|")
.append(child.getHead().getCoveredText()).append("|").append(type);
features.add(feat.toString());
feat.setLength(0);
feat.append("c").append(i).append("*=").append(child.getType()).append("|").append(type);
features.add(feat.toString());
}
protected void checkcons(Parse p1, Parse p2, String type, List features) {
StringBuilder feat = new StringBuilder(20);
feat.append("cil=").append(type).append(",").append(p1.getType()).append("|")
.append(p1.getHead().getCoveredText()).append(",").append(p2.getType())
.append("|").append(p2.getHead().getCoveredText());
features.add(feat.toString());
feat.setLength(0);
feat.append("ci*l=").append(type).append(",").append(p1.getType()).append(",")
.append(p2.getType()).append("|").append(p2.getHead().getCoveredText());
features.add(feat.toString());
feat.setLength(0);
feat.append("cil*=").append(type).append(",").append(p1.getType()).append("|")
.append(p1.getHead().getCoveredText()).append(",").append(p2.getType());
features.add(feat.toString());
feat.setLength(0);
feat.append("ci*l*=").append(type).append(",").append(p1.getType())
.append(",").append(p2.getType());
features.add(feat.toString());
}
/**
* Populates specified nodes array with left-most right frontier
* node with a unique head. If the right frontier doesn't contain
* enough nodes, then {@code nulls} are placed in the array elements.
*
* @param rf The current right frontier.
* @param nodes The array to be populated.
*/
protected void getFrontierNodes(List rf, Parse[] nodes) {
int leftIndex = 0;
int prevHeadIndex = -1;
for (Parse fn : rf) {
int headIndex = fn.getHeadIndex();
if (headIndex != prevHeadIndex) {
nodes[leftIndex] = fn;
leftIndex++;
prevHeadIndex = headIndex;
if (leftIndex == nodes.length) {
break;
}
}
}
for (int ni = leftIndex; ni < nodes.length; ni++) {
nodes[ni] = null;
}
}
}