/**
* Strip away case edges, if the incoming edge is a preposition. This replicates the behavior of
* the old Stanford dependencies on universal dependencies.
*
* @param tree The tree to modify in place.
*/
public static void stripPrepCases(SemanticGraph tree) {
// Find incoming case edges that have an 'nmod' incoming edge
List<SemanticGraphEdge> toClean = new ArrayList<>();
for (SemanticGraphEdge edge : tree.edgeIterable()) {
if ("case".equals(edge.getRelation().toString())) {
boolean isPrepTarget = false;
for (SemanticGraphEdge incoming : tree.incomingEdgeIterable(edge.getGovernor())) {
if ("nmod".equals(incoming.getRelation().getShortName())) {
isPrepTarget = true;
break;
}
}
if (isPrepTarget && !tree.outgoingEdgeIterator(edge.getDependent()).hasNext()) {
toClean.add(edge);
}
}
}
// Delete these edges
for (SemanticGraphEdge edge : toClean) {
tree.removeEdge(edge);
tree.removeVertex(edge.getDependent());
assert isTree(tree);
}
}
/**
* The basic method for splitting off a clause of a tree. This modifies the tree in place. This
* method addtionally follows ref edges.
*
* @param tree The tree to split a clause from.
* @param toKeep The edge representing the clause to keep.
*/
@SuppressWarnings("unchecked")
private void simpleClause(SemanticGraph tree, SemanticGraphEdge toKeep) {
splitToChildOfEdge(tree, toKeep);
// Follow 'ref' edges
Map<IndexedWord, IndexedWord> refReplaceMap = new HashMap<>();
// (find replacements)
for (IndexedWord vertex : tree.vertexSet()) {
for (SemanticGraphEdge edge : extraEdgesByDependent.get(vertex)) {
if ("ref".equals(edge.getRelation().toString())
&& // it's a ref edge...
!tree.containsVertex(
edge.getGovernor())) { // ...that doesn't already exist in the tree.
refReplaceMap.put(vertex, edge.getGovernor());
}
}
}
// (do replacements)
for (Map.Entry<IndexedWord, IndexedWord> entry : refReplaceMap.entrySet()) {
Iterator<SemanticGraphEdge> iter = tree.incomingEdgeIterator(entry.getKey());
if (!iter.hasNext()) {
continue;
}
SemanticGraphEdge incomingEdge = iter.next();
IndexedWord governor = incomingEdge.getGovernor();
tree.removeVertex(entry.getKey());
addSubtree(
tree,
governor,
incomingEdge.getRelation().toString(),
this.tree,
entry.getValue(),
this.tree.incomingEdgeList(tree.getFirstRoot()));
}
}
/**
* Stips aux and mark edges when we are splitting into a clause.
*
* @param toModify The tree we are stripping the edges from.
*/
private void stripAuxMark(SemanticGraph toModify) {
List<SemanticGraphEdge> toClean = new ArrayList<>();
for (SemanticGraphEdge edge : toModify.outgoingEdgeIterable(toModify.getFirstRoot())) {
String rel = edge.getRelation().toString();
if (("aux".equals(rel) || "mark".equals(rel))
&& !toModify.outgoingEdgeIterator(edge.getDependent()).hasNext()) {
toClean.add(edge);
}
}
for (SemanticGraphEdge edge : toClean) {
toModify.removeEdge(edge);
toModify.removeVertex(edge.getDependent());
}
}
/**
* 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;
}