@Override
public Tuple parse(
IndexedWord gov, IndexedWord dep, SemanticGraph depGraph, Tuple t, Set<IndexedWord> visited) {
getPOSString(gov, dep);
Tuple t1;
/*
* Check for LeafNode
*/
if (depGraph.getChildren(dep).size() > 0) {
t1 = parse(dep, depGraph, visited);
} else {
Entity e = new Entity(dep.word(), EntityType.Notion);
t1 = new Tuple(e);
}
String s = depGraph.getEdge(gov, dep).getRelation().getSpecific();
Relation r = new Relation(s, RelationType.One2One);
if (t == null) {
Entity e1 = new Entity(gov.word(), EntityType.Object);
Tuple t2 = new Tuple(e1);
t = new Tuple(t1, r, t2);
} else {
t = new Tuple(t1, r, t);
}
logger.info(t.toString());
return t;
}
public boolean isCopy(IndexedWord otherWord) {
Integer myInd = get(CoreAnnotations.IndexAnnotation.class);
Integer otherInd = otherWord.get(CoreAnnotations.IndexAnnotation.class);
if (myInd == null) {
if (otherInd != null) return false;
} else if (!myInd.equals(otherInd)) {
return false;
}
Integer mySentInd = get(CoreAnnotations.SentenceIndexAnnotation.class);
Integer otherSentInd = otherWord.get(CoreAnnotations.SentenceIndexAnnotation.class);
if (mySentInd == null) {
if (otherSentInd != null) return false;
} else if (!mySentInd.equals(otherSentInd)) {
return false;
}
String myDocID = getString(CoreAnnotations.DocIDAnnotation.class);
String otherDocID = otherWord.getString(CoreAnnotations.DocIDAnnotation.class);
if (myDocID == null) {
if (otherDocID != null) return false;
} else if (!myDocID.equals(otherDocID)) {
return false;
}
if (copyCount() == 0 || otherWord.copyCount() != 0) {
return false;
}
return true;
}
public Entity(IndexedWord... wrd) {
this.name = "";
for (IndexedWord w : wrd) {
this.name = this.name + " " + w.word();
}
this.name = this.name.trim();
this.type = EntityType.Unknown;
}
public void testGetCommonAncestor() {
IndexedWord common =
graph.getCommonAncestor(graph.getNodeByIndex(43), graph.getNodeByIndex(44));
assertEquals(45, common.index());
common = graph.getCommonAncestor(graph.getNodeByIndex(41), graph.getNodeByIndex(39));
assertEquals(41, common.index());
common = graph.getCommonAncestor(graph.getNodeByIndex(39), graph.getNodeByIndex(41));
assertEquals(41, common.index());
common = graph.getCommonAncestor(graph.getNodeByIndex(40), graph.getNodeByIndex(42));
assertEquals(41, common.index());
// too far for this method
common = graph.getCommonAncestor(graph.getNodeByIndex(10), graph.getNodeByIndex(42));
assertEquals(null, common);
common = graph.getCommonAncestor(graph.getNodeByIndex(10), graph.getNodeByIndex(10));
assertEquals(10, common.index());
common = graph.getCommonAncestor(graph.getNodeByIndex(40), graph.getNodeByIndex(40));
assertEquals(40, common.index());
// a couple tests at the top of the graph
common = graph.getCommonAncestor(graph.getNodeByIndex(10), graph.getNodeByIndex(1));
assertEquals(10, common.index());
common = graph.getCommonAncestor(graph.getNodeByIndex(1), graph.getNodeByIndex(10));
assertEquals(10, common.index());
}
public void testShortestPath() {
graph.prettyPrint();
IndexedWord word1 = graph.getNodeByIndex(10);
IndexedWord word2 = graph.getNodeByIndex(14);
System.out.println("word1: " + word1);
System.out.println("word1: " + word1.hashCode());
System.out.println("word2: " + word2);
System.out.println("word2: " + word2.hashCode());
System.out.println("word eq: " + word1.equals(word2));
System.out.println("word eq: " + (word1.hashCode() == word2.hashCode()));
System.out.println("word eq: " + (word1.toString().equals(word2.toString())));
List<SemanticGraphEdge> edges = graph.getShortestUndirectedPathEdges(word1, word2);
System.out.println("path: " + edges);
assertNotNull(edges);
List<IndexedWord> nodes = graph.getShortestUndirectedPathNodes(word1, word2);
System.out.println("path: " + nodes);
assertNotNull(nodes);
assertEquals(word1, nodes.get(0));
assertEquals(word2, nodes.get(nodes.size() - 1));
edges = graph.getShortestUndirectedPathEdges(word1, word1);
System.out.println("path: " + edges);
assertNotNull(edges);
assertEquals(0, edges.size());
nodes = graph.getShortestUndirectedPathNodes(word1, word1);
System.out.println("path: " + nodes);
assertNotNull(nodes);
assertEquals(1, nodes.size());
assertEquals(word1, nodes.get(0));
}
private static void verifySet(Collection<IndexedWord> nodes, int... expected) {
Set<Integer> results = Generics.newTreeSet();
for (IndexedWord node : nodes) {
results.add(node.index());
}
Set<Integer> expectedIndices = Generics.newTreeSet();
for (Integer index : expected) {
expectedIndices.add(index);
}
assertEquals(expectedIndices, results);
}
public void testHasChildren() {
SemanticGraph gr = SemanticGraph.valueOf("[ate subj>Bill dobj>[muffins compound>blueberry]]");
List<IndexedWord> vertices = gr.vertexListSorted();
for (IndexedWord word : vertices) {
if (word.word().equals("ate") || word.word().equals("muffins")) {
assertTrue(gr.hasChildren(word));
} else {
assertFalse(gr.hasChildren(word));
}
}
}
public static DependencyParse parse(String text) {
if (pipeline == null) {
loadModels();
}
DependencyParse parse = new DependencyParse();
Annotation document = new Annotation(text);
pipeline.annotate(document);
List<CoreMap> sentences = document.get(SentencesAnnotation.class);
for (CoreMap sentence : sentences) {
SemanticGraph dependencies = sentence.get(CollapsedCCProcessedDependenciesAnnotation.class);
IndexedWord root = dependencies.getFirstRoot();
parse.setHeadNode(root.index());
List<SemanticGraphEdge> edges = dependencies.edgeListSorted();
// System.out.println(edges);
for (SemanticGraphEdge t : edges) {
String dep = t.getDependent().originalText();
int depIndex = t.getDependent().index();
String depPOS = t.getDependent().tag();
int depStart = t.getDependent().beginPosition();
int depEnd = t.getDependent().endPosition();
String gov = t.getGovernor().originalText();
int govIndex = t.getGovernor().index();
String govPOS = t.getGovernor().tag();
int govStart = t.getGovernor().beginPosition();
int govEnd = t.getGovernor().endPosition();
parse.addNode(govIndex, gov, govPOS, govStart, govEnd);
parse.addNode(depIndex, dep, depPOS, depStart, depEnd);
parse.addEdge(depIndex, govIndex, t.getRelation().getShortName());
}
}
return parse;
}
public IndexedWord makeSoftCopy() {
if (original != null) {
return original.makeSoftCopy();
} else {
return makeSoftCopy(++numCopies);
}
}
@SuppressWarnings("unchecked")
public boolean nodeAttrMatch(IndexedWord node, final SemanticGraph sg, boolean ignoreCase) {
// System.out.println(node.word());
if (isRoot) return (negDesc ? !sg.getRoots().contains(node) : sg.getRoots().contains(node));
// System.out.println("not root");
if (isEmpty)
return (negDesc ? !node.equals(IndexedWord.NO_WORD) : node.equals(IndexedWord.NO_WORD));
// System.err.println("Attributes are: " + attributes);
for (Map.Entry<String, Pattern> attr : attributes.entrySet()) {
String key = attr.getKey();
// System.out.println(key);
String nodeValue;
// if (key.equals("idx"))
// nodeValue = Integer.toString(node.index());
// else {
Class c = Env.lookupAnnotationKey(env, key);
// find class for the key
Object value = node.get(c);
if (value == null) nodeValue = null;
else nodeValue = value.toString();
// }
// System.out.println(nodeValue);
if (nodeValue == null) return negDesc;
Pattern valuePattern = attr.getValue();
boolean matches = false;
if (ignoreCase) {
if (Pattern.compile(valuePattern.pattern(), Pattern.CASE_INSENSITIVE)
.matcher(nodeValue)
.matches()) matches = true;
} else {
if (nodeValue.matches(valuePattern.pattern())) matches = true;
}
if (!matches) {
// System.out.println("doesn't match");
// System.out.println("");
return negDesc;
}
}
// System.out.println("matches");
// System.out.println("");
return !negDesc;
}
private void testParseTree() {
try {
Properties props = new Properties();
props.setProperty("annotators", "tokenize, ssplit, pos, lemma, ner, parse, dcoref");
StanfordCoreNLP pipeline = new StanfordCoreNLP(props);
// read some text in the text variable
String text = "Give me a list of all bandleaders that play trumpet.";
// create an empty Annotation just with the given text
Annotation document = new Annotation(text);
// run all Annotators on this text
pipeline.annotate(document);
// these are all the sentences in this document
// a CoreMap is essentially a Map that uses class objects as keys and has values with custom
// types
List<CoreMap> sentences = document.get(SentencesAnnotation.class);
for (CoreMap sentence : sentences) {
// traversing the words in the current sentence
// a CoreLabel is a CoreMap with additional token-specific methods
// this is the parse tree of the current sentence
Tree tree = sentence.get(TreeAnnotation.class);
// this is the Stanford dependency graph of the current sentence
SemanticGraph dependencies = sentence.get(CollapsedCCProcessedDependenciesAnnotation.class);
Set<IndexedWord> vertices = dependencies.vertexSet();
List<SemanticGraphEdge> edges = dependencies.edgeListSorted();
for (SemanticGraphEdge e : edges) {}
for (IndexedWord i : vertices) {
System.out.println(i.toString());
}
}
} catch (Exception e) {
}
}
/**
* This method attempts to resolve noun phrases which consist of more than one word. More
* precisely, it looks for nn dependencies below {@code head} and creates an entity.
*
* @param head The head of the noun phrase
* @param graph The sentence to look in.
* @param words The words which make up the noun phrase
* @return A distinct word
*/
public static String resolveNN(
IndexedWord head, SemanticGraph graph, ArrayList<IndexedWord> words) {
List<IndexedWord> nns =
graph.getChildrenWithReln(head, EnglishGrammaticalRelations.NOUN_COMPOUND_MODIFIER);
String name = "";
// check for nulls. if there is nothing here, we have nothing to do.
if (nns != null) {
for (IndexedWord part : nns) {
name += part.word();
name += " ";
words.add(part); // save this word as a part of the results
}
// append the head word ("starting" word)
name += head.word();
words.add(head); // save this word as a part of the results
return name;
} else {
return null;
}
}
public void testCommonAncestor() {
IndexedWord word1 = graph.getNodeByIndex(43);
IndexedWord word2 = graph.getNodeByIndex(44);
IndexedWord common = graph.getCommonAncestor(word1, word2);
System.out.println("word1: " + word1);
System.out.println("word2: " + word2);
System.out.println("common: " + common);
System.out.println(
"common ancestor between "
+ word1.value()
+ "-"
+ word1.index()
+ " and "
+ word2.value()
+ "-"
+ word2.index()
+ " is "
+ common.value()
+ "-"
+ common.index());
assertEquals(45, common.index());
}
/**
* This .equals is dependent only on docID, sentenceIndex, and index. It doesn't consider the
* actual word value, but assumes that it is validly represented by token position. All
* IndexedWords that lack these fields will be regarded as equal.
*/
@Override
public boolean equals(Object o) {
if (this == o) return true;
if (!(o instanceof IndexedWord)) return false;
// now compare on appropriate keys
final IndexedWord otherWord = (IndexedWord) o;
Integer myInd = get(CoreAnnotations.IndexAnnotation.class);
Integer otherInd = otherWord.get(CoreAnnotations.IndexAnnotation.class);
if (myInd == null) {
if (otherInd != null) return false;
} else if (!myInd.equals(otherInd)) {
return false;
}
Integer mySentInd = get(CoreAnnotations.SentenceIndexAnnotation.class);
Integer otherSentInd = otherWord.get(CoreAnnotations.SentenceIndexAnnotation.class);
if (mySentInd == null) {
if (otherSentInd != null) return false;
} else if (!mySentInd.equals(otherSentInd)) {
return false;
}
String myDocID = getString(CoreAnnotations.DocIDAnnotation.class);
String otherDocID = otherWord.getString(CoreAnnotations.DocIDAnnotation.class);
if (myDocID == null) {
if (otherDocID != null) return false;
} else if (!myDocID.equals(otherDocID)) {
return false;
}
if (copyCount() != otherWord.copyCount()) {
return false;
}
// Compare pseudo-positions
if ((!Double.isNaN(this.pseudoPosition) || !Double.isNaN(otherWord.pseudoPosition))
&& this.pseudoPosition != otherWord.pseudoPosition) {
return false;
}
return true;
}
/**
* NOTE: This compareTo is based on and made to be compatible with the one from
* IndexedFeatureLabel. You <em>must</em> have a DocIDAnnotation, SentenceIndexAnnotation, and
* IndexAnnotation for this to make sense and be guaranteed to work properly. Currently, it won't
* error out and will try to return something sensible if these are not defined, but that really
* isn't proper usage!
*
* <p>This compareTo method is based not by value elements like the word(), but on passage
* position. It puts NO_WORD elements first, and then orders by document, sentence, and word
* index. If these do not differ, it returns equal.
*
* @param w The IndexedWord to compare with
* @return Whether this is less than w or not in the ordering
*/
@Override
public int compareTo(IndexedWord w) {
if (this.equals(IndexedWord.NO_WORD)) {
if (w.equals(IndexedWord.NO_WORD)) {
return 0;
} else {
return -1;
}
}
if (w.equals(IndexedWord.NO_WORD)) {
return 1;
}
// Override the default comparator if pseudo-positions are set.
// This is needed for splicing trees together awkwardly in OpenIE.
if (!Double.isNaN(w.pseudoPosition) || !Double.isNaN(this.pseudoPosition)) {
double val = this.pseudoPosition() - w.pseudoPosition();
if (val < 0) {
return -1;
}
if (val > 0) {
return 1;
} else {
return 0;
}
}
// Otherwise, compare using the normal doc/sentence/token index hierarchy
String docID = this.getString(CoreAnnotations.DocIDAnnotation.class);
int docComp = docID.compareTo(w.getString(CoreAnnotations.DocIDAnnotation.class));
if (docComp != 0) return docComp;
int sentComp = sentIndex() - w.sentIndex();
if (sentComp != 0) return sentComp;
int indexComp = index() - w.index();
if (indexComp != 0) return indexComp;
return copyCount() - w.copyCount();
}
// when finished = false; break; is called, it means I successfully matched.
@SuppressWarnings("null")
private void goToNextNodeMatch() {
decommitVariableGroups(); // make sure variable groups are free.
decommitNamedNodes();
decommitNamedRelations();
finished = true;
Matcher m = null;
while (nodeMatchCandidateIterator.hasNext()) {
if (myNode.reln.getName() != null) {
String foundReln = namesToRelations.get(myNode.reln.getName());
nextMatchReln = ((GraphRelation.SearchNodeIterator) nodeMatchCandidateIterator).getReln();
if ((foundReln != null) && (!nextMatchReln.equals(foundReln))) {
nextMatch = nodeMatchCandidateIterator.next();
continue;
}
}
nextMatch = nodeMatchCandidateIterator.next();
// System.err.println("going to next match: " + nextMatch.word() + " " +
// myNode.descString + " " + myNode.isLink);
if (myNode.descString.equals("{}") && myNode.isLink) {
IndexedWord otherNode = namesToNodes.get(myNode.name);
if (otherNode != null) {
if (otherNode.equals(nextMatch)) {
if (!myNode.negDesc) {
finished = false;
break;
}
} else {
if (myNode.negDesc) {
finished = false;
break;
}
}
} else {
boolean found = myNode.nodeAttrMatch(nextMatch, hyp ? sg : sg_aligned, ignoreCase);
if (found) {
for (Pair<Integer, String> varGroup : myNode.variableGroups) {
// if variables have been captured from a regex, they
// must match any previous matchings
String thisVariable = varGroup.second();
String thisVarString = variableStrings.getString(thisVariable);
if (thisVarString != null && !thisVarString.equals(m.group(varGroup.first()))) {
// failed to match a variable
found = false;
break;
}
}
// nodeAttrMatch already checks negDesc, so no need to
// check for that here
finished = false;
break;
}
}
} else { // try to match the description pattern.
boolean found = myNode.nodeAttrMatch(nextMatch, hyp ? sg : sg_aligned, ignoreCase);
if (found) {
for (Pair<Integer, String> varGroup : myNode.variableGroups) {
// if variables have been captured from a regex, they
// must match any previous matchings
String thisVariable = varGroup.second();
String thisVarString = variableStrings.getString(thisVariable);
if (thisVarString != null && !thisVarString.equals(m.group(varGroup.first()))) {
// failed to match a variable
found = false;
break;
}
}
// nodeAttrMatch already checks negDesc, so no need to
// check for that here
finished = false;
break;
}
}
} // end while
if (!finished) { // I successfully matched.
resetChild();
if (myNode.name != null) {
// note: have to fill in the map as we go for backreferencing
if (!namesToNodes.containsKey(myNode.name)) {
// System.err.println("making namedFirst");
namedFirst = true;
}
// System.err.println("adding named node: " + myNode.name + "=" +
// nextMatch.word());
namesToNodes.put(myNode.name, nextMatch);
}
if (myNode.reln.getName() != null) {
if (!namesToRelations.containsKey(myNode.reln.getName())) relnNamedFirst = true;
namesToRelations.put(myNode.reln.getName(), nextMatchReln);
}
commitVariableGroups(m); // commit my variable groups.
}
// finished is false exiting this if and only if nextChild exists
// and has a label or backreference that matches
// (also it will just have been reset)
}
/**
* This method searches for an index word in a sentence tree
*
* @param wordToFind
* @param treeToSearch
* @param expectedPOS The expected POS tag for the result. If this is NULL, the method tries to
* find a phrase.
* @param canGoUp If TRUE the method will walk up the tree to find a phrase.
* @param skip Set to "1" if you want to find the phrase for "in front of". Set to "0" otherwise.
* @return The largest matching tree.
*/
public static Tree match(
IndexedWord wordToFind, Tree treeToSearch, String expectedPOS, boolean canGoUp, int skip) {
int end = wordToFind.get(EndIndexAnnotation.class);
int begin = wordToFind.get(BeginIndexAnnotation.class);
// first, find whatever is at the word's index
for (Tree tree : treeToSearch) {
CoreLabel lbl = ((CoreLabel) tree.label());
if (lbl != null
&& lbl.get(EndIndexAnnotation.class) != null
&& lbl.get(EndIndexAnnotation.class) == end) {
if (lbl.get(BeginIndexAnnotation.class) == begin) {
// we found the first subtree at the word's index
// now, check if the word here is our searchword
if (tree.getLeaves().get(0).label().value().equals(wordToFind.value())) {
// we have found the label.
Tree candidate = tree;
if (expectedPOS != null) {
// if we know our desired POS, just keep walking up the tree to find the first
// instance of the expected pos
while (!expectedPOS.equals(candidate.value())) {
// if we don't have the right POS, just try our parent
candidate = candidate.parent(treeToSearch);
if (candidate == null) {
return null;
}
}
candidate = skip(candidate, treeToSearch, expectedPOS, skip);
} else {
// else walk up the tree again to find the corresponding phrase
while (!candidate.isPhrasal()) {
candidate =
candidate.parent(treeToSearch); // edu.stanford.nlp.trees.Tree.parent(Tree root)
if (candidate == null) {
return null;
}
}
}
if (canGoUp) {
// now keep walking as long as the phrase does not change. this should yield the
// largest representative phrase for this word.
String phrase = candidate.value();
while (phrase.equals(candidate.parent(treeToSearch).value())) {
candidate = candidate.parent(treeToSearch);
if (candidate == null) {
return null;
}
}
}
return candidate;
}
}
}
}
return null;
}
@Override
public double computeValue(IndexedWord label) {
double result = (Objects.equals(label.ner(), tag)) ? 1.0 : 0.0;
return result;
}
@Override
public void handle(HttpExchange httpExchange) throws IOException {
// Set common response headers
httpExchange.getResponseHeaders().add("Access-Control-Allow-Origin", "*");
Future<String> json =
corenlpExecutor.submit(
() -> {
try {
// Get the document
Properties props =
new Properties() {
{
setProperty("annotators", "tokenize,ssplit,pos,lemma,ner,depparse");
}
};
Annotation doc = getDocument(props, httpExchange);
if (!doc.containsKey(CoreAnnotations.SentencesAnnotation.class)) {
StanfordCoreNLP pipeline = mkStanfordCoreNLP(props);
pipeline.annotate(doc);
}
// Construct the matcher
Map<String, String> params = getURLParams(httpExchange.getRequestURI());
// (get the pattern)
if (!params.containsKey("pattern")) {
respondError("Missing required parameter 'pattern'", httpExchange);
return "";
}
String pattern = params.get("pattern");
// (get whether to filter / find)
String filterStr = params.getOrDefault("filter", "false");
final boolean filter =
filterStr.trim().isEmpty()
|| "true".equalsIgnoreCase(filterStr.toLowerCase());
// (create the matcher)
final SemgrexPattern regex = SemgrexPattern.compile(pattern);
// Run TokensRegex
return JSONOutputter.JSONWriter.objectToJSON(
(docWriter) -> {
if (filter) {
// Case: just filter sentences
docWriter.set(
"sentences",
doc.get(CoreAnnotations.SentencesAnnotation.class)
.stream()
.map(
sentence ->
regex
.matcher(
sentence.get(
SemanticGraphCoreAnnotations
.CollapsedCCProcessedDependenciesAnnotation
.class))
.matches())
.collect(Collectors.toList()));
} else {
// Case: find matches
docWriter.set(
"sentences",
doc.get(CoreAnnotations.SentencesAnnotation.class)
.stream()
.map(
sentence ->
(Consumer<JSONOutputter.Writer>)
(JSONOutputter.Writer sentWriter) -> {
SemgrexMatcher matcher =
regex.matcher(
sentence.get(
SemanticGraphCoreAnnotations
.CollapsedCCProcessedDependenciesAnnotation
.class));
int i = 0;
while (matcher.find()) {
sentWriter.set(
Integer.toString(i),
(Consumer<JSONOutputter.Writer>)
(JSONOutputter.Writer matchWriter) -> {
IndexedWord match = matcher.getMatch();
matchWriter.set("text", match.word());
matchWriter.set(
"begin", match.index() - 1);
matchWriter.set("end", match.index());
for (String capture :
matcher.getNodeNames()) {
matchWriter.set(
"$" + capture,
(Consumer<JSONOutputter.Writer>)
groupWriter -> {
IndexedWord node =
matcher.getNode(
capture);
groupWriter.set(
"text", node.word());
groupWriter.set(
"begin",
node.index() - 1);
groupWriter.set(
"end", node.index());
});
}
});
i += 1;
}
sentWriter.set("length", i);
}));
}
});
} catch (Exception e) {
e.printStackTrace();
try {
respondError(e.getClass().getName() + ": " + e.getMessage(), httpExchange);
} catch (IOException ignored) {
}
}
return "";
});
// Send response
byte[] response = new byte[0];
try {
response = json.get(5, TimeUnit.SECONDS).getBytes();
} catch (InterruptedException | ExecutionException | TimeoutException e) {
respondError("Timeout when executing Semgrex query", httpExchange);
}
if (response.length > 0) {
httpExchange.getResponseHeaders().add("Content-Type", "text/json");
httpExchange.getResponseHeaders().add("Content-Length", Integer.toString(response.length));
httpExchange.sendResponseHeaders(HTTP_OK, response.length);
httpExchange.getResponseBody().write(response);
httpExchange.close();
}
}
/**
* Fix some bizarre peculiarities with certain trees. So far, these include:
*
* <ul>
* <li>Sometimes there's a node from a word to itself. This seems wrong.
* </ul>
*
* @param tree The tree to clean (in place!).
* @return A list of extra edges, which are valid but were removed.
*/
public static List<SemanticGraphEdge> cleanTree(SemanticGraph tree) {
// assert !isCyclic(tree);
// Clean nodes
List<IndexedWord> toDelete = new ArrayList<>();
for (IndexedWord vertex : tree.vertexSet()) {
// Clean punctuation
if (vertex.tag() == null) {
continue;
}
char tag = vertex.backingLabel().tag().charAt(0);
if (tag == '.' || tag == ',' || tag == '(' || tag == ')' || tag == ':') {
if (!tree.outgoingEdgeIterator(vertex)
.hasNext()) { // This should really never happen, but it does.
toDelete.add(vertex);
}
}
}
toDelete.forEach(tree::removeVertex);
// Clean edges
Iterator<SemanticGraphEdge> iter = tree.edgeIterable().iterator();
while (iter.hasNext()) {
SemanticGraphEdge edge = iter.next();
if (edge.getDependent().index() == edge.getGovernor().index()) {
// Clean self-edges
iter.remove();
} else if (edge.getRelation().toString().equals("punct")) {
// Clean punctuation (again)
if (!tree.outgoingEdgeIterator(edge.getDependent())
.hasNext()) { // This should really never happen, but it does.
iter.remove();
}
}
}
// Remove extra edges
List<SemanticGraphEdge> extraEdges = new ArrayList<>();
for (SemanticGraphEdge edge : tree.edgeIterable()) {
if (edge.isExtra()) {
if (tree.incomingEdgeList(edge.getDependent()).size() > 1) {
extraEdges.add(edge);
}
}
}
extraEdges.forEach(tree::removeEdge);
// Add apposition edges (simple coref)
for (SemanticGraphEdge extraEdge :
new ArrayList<>(extraEdges)) { // note[gabor] prevent concurrent modification exception
for (SemanticGraphEdge candidateAppos : tree.incomingEdgeIterable(extraEdge.getDependent())) {
if (candidateAppos.getRelation().toString().equals("appos")) {
extraEdges.add(
new SemanticGraphEdge(
extraEdge.getGovernor(),
candidateAppos.getGovernor(),
extraEdge.getRelation(),
extraEdge.getWeight(),
extraEdge.isExtra()));
}
}
for (SemanticGraphEdge candidateAppos : tree.outgoingEdgeIterable(extraEdge.getDependent())) {
if (candidateAppos.getRelation().toString().equals("appos")) {
extraEdges.add(
new SemanticGraphEdge(
extraEdge.getGovernor(),
candidateAppos.getDependent(),
extraEdge.getRelation(),
extraEdge.getWeight(),
extraEdge.isExtra()));
}
}
}
// Brute force ensure tree
// Remove incoming edges from roots
List<SemanticGraphEdge> rootIncomingEdges = new ArrayList<>();
for (IndexedWord root : tree.getRoots()) {
for (SemanticGraphEdge incomingEdge : tree.incomingEdgeIterable(root)) {
rootIncomingEdges.add(incomingEdge);
}
}
rootIncomingEdges.forEach(tree::removeEdge);
// Loop until it becomes a tree.
boolean changed = true;
while (changed) { // I just want trees to be trees; is that so much to ask!?
changed = false;
List<IndexedWord> danglingNodes = new ArrayList<>();
List<SemanticGraphEdge> invalidEdges = new ArrayList<>();
for (IndexedWord vertex : tree.vertexSet()) {
// Collect statistics
Iterator<SemanticGraphEdge> incomingIter = tree.incomingEdgeIterator(vertex);
boolean hasIncoming = incomingIter.hasNext();
boolean hasMultipleIncoming = false;
if (hasIncoming) {
incomingIter.next();
hasMultipleIncoming = incomingIter.hasNext();
}
// Register actions
if (!hasIncoming && !tree.getRoots().contains(vertex)) {
danglingNodes.add(vertex);
} else {
if (hasMultipleIncoming) {
for (SemanticGraphEdge edge : new IterableIterator<>(incomingIter)) {
invalidEdges.add(edge);
}
}
}
}
// Perform actions
for (IndexedWord vertex : danglingNodes) {
tree.removeVertex(vertex);
changed = true;
}
for (SemanticGraphEdge edge : invalidEdges) {
tree.removeEdge(edge);
changed = true;
}
}
// Return
assert isTree(tree);
return extraEdges;
}
public IndexedWord makeCopy(int count) {
CoreLabel labelCopy = new CoreLabel(label);
IndexedWord copy = new IndexedWord(labelCopy);
copy.setCopyCount(count);
return copy;
}
public IndexedWord makeSoftCopy(int count) {
IndexedWord copy = new IndexedWord(label);
copy.setCopyCount(count);
copy.original = this;
return copy;
}
@Override
public void process(JCas jCas) throws AnalysisEngineProcessException {
Annotation document = this.processor.process(jCas.getDocumentText());
String lastNETag = "O";
int lastNEBegin = -1;
int lastNEEnd = -1;
for (CoreMap tokenAnn : document.get(TokensAnnotation.class)) {
// create the token annotation
int begin = tokenAnn.get(CharacterOffsetBeginAnnotation.class);
int end = tokenAnn.get(CharacterOffsetEndAnnotation.class);
String pos = tokenAnn.get(PartOfSpeechAnnotation.class);
String lemma = tokenAnn.get(LemmaAnnotation.class);
Token token = new Token(jCas, begin, end);
token.setPos(pos);
token.setLemma(lemma);
token.addToIndexes();
// hackery to convert token-level named entity tag into phrase-level tag
String neTag = tokenAnn.get(NamedEntityTagAnnotation.class);
if (neTag.equals("O") && !lastNETag.equals("O")) {
NamedEntityMention ne = new NamedEntityMention(jCas, lastNEBegin, lastNEEnd);
ne.setMentionType(lastNETag);
ne.addToIndexes();
} else {
if (lastNETag.equals("O")) {
lastNEBegin = begin;
} else if (lastNETag.equals(neTag)) {
// do nothing - begin was already set
} else {
NamedEntityMention ne = new NamedEntityMention(jCas, lastNEBegin, lastNEEnd);
ne.setMentionType(lastNETag);
ne.addToIndexes();
lastNEBegin = begin;
}
lastNEEnd = end;
}
lastNETag = neTag;
}
if (!lastNETag.equals("O")) {
NamedEntityMention ne = new NamedEntityMention(jCas, lastNEBegin, lastNEEnd);
ne.setMentionType(lastNETag);
ne.addToIndexes();
}
// add sentences and trees
for (CoreMap sentenceAnn : document.get(SentencesAnnotation.class)) {
// add the sentence annotation
int sentBegin = sentenceAnn.get(CharacterOffsetBeginAnnotation.class);
int sentEnd = sentenceAnn.get(CharacterOffsetEndAnnotation.class);
Sentence sentence = new Sentence(jCas, sentBegin, sentEnd);
sentence.addToIndexes();
// add the syntactic tree annotation
List<CoreLabel> tokenAnns = sentenceAnn.get(TokensAnnotation.class);
Tree tree = sentenceAnn.get(TreeAnnotation.class);
if (tree.children().length != 1) {
throw new RuntimeException("Expected single root node, found " + tree);
}
tree = tree.firstChild();
tree.indexSpans(0);
TopTreebankNode root = new TopTreebankNode(jCas);
root.setTreebankParse(tree.toString());
// TODO: root.setTerminals(v)
this.addTreebankNodeToIndexes(root, jCas, tree, tokenAnns);
// get the dependencies
SemanticGraph dependencies =
sentenceAnn.get(CollapsedCCProcessedDependenciesAnnotation.class);
// convert Stanford nodes to UIMA annotations
List<Token> tokens = JCasUtil.selectCovered(jCas, Token.class, sentence);
Map<IndexedWord, DependencyNode> stanfordToUima = new HashMap<IndexedWord, DependencyNode>();
for (IndexedWord stanfordNode : dependencies.vertexSet()) {
int indexBegin = stanfordNode.get(BeginIndexAnnotation.class);
int indexEnd = stanfordNode.get(EndIndexAnnotation.class);
int tokenBegin = tokens.get(indexBegin).getBegin();
int tokenEnd = tokens.get(indexEnd - 1).getEnd();
DependencyNode node;
if (dependencies.getRoots().contains(stanfordNode)) {
node = new TopDependencyNode(jCas, tokenBegin, tokenEnd);
} else {
node = new DependencyNode(jCas, tokenBegin, tokenEnd);
}
stanfordToUima.put(stanfordNode, node);
}
// create relation annotations for each Stanford dependency
ArrayListMultimap<DependencyNode, DependencyRelation> headRelations =
ArrayListMultimap.create();
ArrayListMultimap<DependencyNode, DependencyRelation> childRelations =
ArrayListMultimap.create();
for (SemanticGraphEdge stanfordEdge : dependencies.edgeIterable()) {
DependencyRelation relation = new DependencyRelation(jCas);
DependencyNode head = stanfordToUima.get(stanfordEdge.getGovernor());
DependencyNode child = stanfordToUima.get(stanfordEdge.getDependent());
String relationType = stanfordEdge.getRelation().toString();
if (head == null || child == null || relationType == null) {
throw new RuntimeException(
String.format(
"null elements not allowed in relation:\nrelation=%s\nchild=%s\nhead=%s\n",
relation, child, head));
}
relation.setHead(head);
relation.setChild(child);
relation.setRelation(relationType);
relation.addToIndexes();
headRelations.put(child, relation);
childRelations.put(head, relation);
}
// set the relations for each node annotation
for (DependencyNode node : stanfordToUima.values()) {
List<DependencyRelation> heads = headRelations.get(node);
node.setHeadRelations(new FSArray(jCas, heads == null ? 0 : heads.size()));
if (heads != null) {
FSCollectionFactory.fillArrayFS(node.getHeadRelations(), heads);
}
List<DependencyRelation> children = childRelations.get(node);
node.setChildRelations(new FSArray(jCas, children == null ? 0 : children.size()));
if (children != null) {
FSCollectionFactory.fillArrayFS(node.getChildRelations(), children);
}
node.addToIndexes();
}
}
// map from spans to named entity mentions
Map<Span, NamedEntityMention> spanMentionMap = new HashMap<Span, NamedEntityMention>();
for (NamedEntityMention mention : JCasUtil.select(jCas, NamedEntityMention.class)) {
spanMentionMap.put(new Span(mention.getBegin(), mention.getEnd()), mention);
}
// add mentions for all entities identified by the coreference system
List<NamedEntity> entities = new ArrayList<NamedEntity>();
List<List<Token>> sentenceTokens = new ArrayList<List<Token>>();
for (Sentence sentence : JCasUtil.select(jCas, Sentence.class)) {
sentenceTokens.add(JCasUtil.selectCovered(jCas, Token.class, sentence));
}
Map<Integer, CorefChain> corefChains = document.get(CorefChainAnnotation.class);
for (CorefChain chain : corefChains.values()) {
List<NamedEntityMention> mentions = new ArrayList<NamedEntityMention>();
for (CorefMention corefMention : chain.getMentionsInTextualOrder()) {
// figure out the character span of the token
List<Token> tokens = sentenceTokens.get(corefMention.sentNum - 1);
int begin = tokens.get(corefMention.startIndex - 1).getBegin();
int end = tokens.get(corefMention.endIndex - 2).getEnd();
// use an existing named entity mention when possible; otherwise create a new one
NamedEntityMention mention = spanMentionMap.get(new Span(begin, end));
if (mention == null) {
mention = new NamedEntityMention(jCas, begin, end);
mention.addToIndexes();
}
mentions.add(mention);
}
// create an entity for the mentions
Collections.sort(
mentions,
new Comparator<NamedEntityMention>() {
@Override
public int compare(NamedEntityMention m1, NamedEntityMention m2) {
return m1.getBegin() - m2.getBegin();
}
});
// create mentions and add them to entity
NamedEntity entity = new NamedEntity(jCas);
entity.setMentions(new FSArray(jCas, mentions.size()));
int index = 0;
for (NamedEntityMention mention : mentions) {
mention.setMentionedEntity(entity);
entity.setMentions(index, mention);
index += 1;
}
entities.add(entity);
}
// add singleton entities for any named entities not picked up by coreference system
for (NamedEntityMention mention : JCasUtil.select(jCas, NamedEntityMention.class)) {
if (mention.getMentionedEntity() == null) {
NamedEntity entity = new NamedEntity(jCas);
entity.setMentions(new FSArray(jCas, 1));
entity.setMentions(0, mention);
mention.setMentionedEntity(entity);
entity.getMentions();
entities.add(entity);
}
}
// sort entities by document order
Collections.sort(
entities,
new Comparator<NamedEntity>() {
@Override
public int compare(NamedEntity o1, NamedEntity o2) {
return getFirstBegin(o1) - getFirstBegin(o2);
}
private int getFirstBegin(NamedEntity entity) {
int min = Integer.MAX_VALUE;
for (NamedEntityMention mention :
JCasUtil.select(entity.getMentions(), NamedEntityMention.class)) {
if (mention.getBegin() < min) {
min = mention.getBegin();
}
}
return min;
}
});
// add entities to document
for (NamedEntity entity : entities) {
entity.addToIndexes();
}
}
/** {@inheritDoc} */
@Override
public LabelFactory labelFactory() {
return IndexedWord.factory();
}