void LeftBinarize( SyntaxTree &tree, ParentNodes &parents )
{
for(ParentNodes::const_iterator p = parents.begin(); p != parents.end(); p++) {
const SplitPoints &point = *p;
if (point.size() > 3) {
const vector< SyntaxNode* >& topNodes
= tree.GetNodes( point[0], point[point.size()-1]-1);
string topLabel = topNodes[0]->GetLabel();
for(size_t i=2; i<point.size()-1; i++) {
// cerr << "LeftBin " << point[0] << "-" << (point[point.size()-1]-1) << ": " << point[0] << "-" << point[i]-1 << " ^" << topLabel << endl;
tree.AddNode( point[0], point[i]-1, "^" + topLabel );
}
}
}
}
void RightBinarize( SyntaxNodeCollection &tree, ParentNodes &parents )
{
for(ParentNodes::const_iterator p = parents.begin(); p != parents.end(); p++) {
const SplitPoints &point = *p;
if (point.size() > 3) {
int endPoint = point[point.size()-1]-1;
const vector< SyntaxNode* >& topNodes
= tree.GetNodes( point[0], endPoint);
string topLabel = topNodes[0]->label;
for(size_t i=1; i<point.size()-2; i++) {
// cerr << "RightBin " << point[0] << "-" << (point[point.size()-1]-1) << ": " << point[i] << "-" << endPoint << " ^" << topLabel << endl;
tree.AddNode( point[i], endPoint, "^" + topLabel );
}
}
}
}
ParentNodes determineSplitPoints(const SyntaxNodeCollection &nodeColl)
{
ParentNodes parents;
const std::size_t numWords = nodeColl.GetNumWords();
// looping through all spans of size >= 2
for( int length=2; length<=numWords; length++ ) {
for( int startPos = 0; startPos <= numWords-length; startPos++ ) {
if (nodeColl.HasNode( startPos, startPos+length-1 )) {
// processing one (parent) span
//std::cerr << "# " << startPos << "-" << (startPos+length-1) << ":";
SplitPoints splitPoints;
splitPoints.push_back( startPos );
//std::cerr << " " << startPos;
int first = 1;
int covered = 0;
int found_somehing = 1; // break loop if nothing found
while( covered < length && found_somehing ) {
// find largest covering subspan (child)
// starting at last covered position
found_somehing = 0;
for( int midPos=length-first; midPos>covered; midPos-- ) {
if( nodeColl.HasNode( startPos+covered, startPos+midPos-1 ) ) {
covered = midPos;
splitPoints.push_back( startPos+covered );
// std::cerr << " " << ( startPos+covered );
first = 0;
found_somehing = 1;
}
}
}
// std::cerr << std::endl;
parents.push_back( splitPoints );
}
}
}
return parents;
}
ParentNodes SyntaxTree::Parse()
{
ParentNodes parents;
int size = m_index.size();
// looping through all spans of size >= 2
for( int length=2; length<=size; length++ ) {
for( int startPos = 0; startPos <= size-length; startPos++ ) {
if (HasNode( startPos, startPos+length-1 )) {
// processing one (parent) span
//std::cerr << "# " << startPos << "-" << (startPos+length-1) << ":";
SplitPoints splitPoints;
splitPoints.push_back( startPos );
//std::cerr << " " << startPos;
int first = 1;
int covered = 0;
while( covered < length ) {
// find largest covering subspan (child)
// starting at last covered position
for( int midPos=length-first; midPos>covered; midPos-- ) {
if( HasNode( startPos+covered, startPos+midPos-1 ) ) {
covered = midPos;
splitPoints.push_back( startPos+covered );
// std::cerr << " " << ( startPos+covered );
first = 0;
}
}
}
// std::cerr << std::endl;
parents.push_back( splitPoints );
}
}
}
return parents;
}
void SAMT( SyntaxTree &tree, ParentNodes &parents )
{
int numWords = tree.GetNumWords();
SyntaxTree newTree; // to store new nodes
// look through parents to combine children
for(ParentNodes::const_iterator p = parents.begin(); p != parents.end(); p++)
{
const SplitPoints &point = *p;
// neighboring childen: DET+ADJ
if (point.size() >= 3) {
// cerr << "complex parent: ";
// for(int i=0;i<point.size();i++) cerr << point[i] << " ";
// cerr << endl;
for(int i = 0; i+2 < point.size(); i++)
{
// cerr << "\tadding " << point[i] << ";" << point[i+1] << ";" << (point[i+2]-1) << ": " << tree.GetNodes(point[i ],point[i+1]-1)[0]->GetLabel() << "+" << tree.GetNodes(point[i+1],point[i+2]-1)[0]->GetLabel() << endl;
newTree.AddNode( point[i],point[i+2]-1,
tree.GetNodes(point[i ],point[i+1]-1)[0]->GetLabel()
+ "+" +
tree.GetNodes(point[i+1],point[i+2]-1)[0]->GetLabel() );
}
}
if (point.size() >= 4) {
int ps = point.size();
string topLabel = tree.GetNodes(point[0],point[ps-1]-1)[0]->GetLabel();
// cerr << "\tadding " << topLabel + "\\" + tree.GetNodes(point[0],point[1]-1)[0]->GetLabel() << endl;
newTree.AddNode( point[1],point[ps-1]-1,
topLabel
+ "\\" +
tree.GetNodes(point[0],point[1]-1)[0]->GetLabel() );
// cerr << "\tadding " << topLabel + "/" + tree.GetNodes(point[ps-2],point[ps-1]-1)[0]->GetLabel() << endl;
newTree.AddNode( point[0],point[ps-2]-1,
topLabel
+ "/" +
tree.GetNodes(point[ps-2],point[ps-1]-1)[0]->GetLabel() );
}
}
// rules for any bordering constituents...
for(int size = 2; size < numWords; size++)
{
for(int start = 0; start < numWords-size+1; start++)
{
int end = start+size-1;
bool done = false;
if (tree.HasNode( start,end ) || newTree.HasNode( start,end )
|| SAMTLevel <= 1)
{
continue;
}
// if matching two adjacent parse constituents: use ++
for(int mid=start+1; mid<=end && !done; mid++)
{
if (tree.HasNode(start,mid-1) && tree.HasNode(mid,end)) {
// cerr << "\tadding " << tree.GetNodes(start,mid-1)[0]->GetLabel() << "++" << tree.GetNodes(mid, end )[0]->GetLabel() << endl;
newTree.AddNode( start, end,
tree.GetNodes(start,mid-1)[0]->GetLabel()
+ "++" +
tree.GetNodes(mid, end )[0]->GetLabel() );
done = true;
}
}
if (done) continue;
// if matching a constituent A right-minus const. B: use A//B
for(int postEnd=end+1; postEnd<numWords && !done; postEnd++)
{
if (tree.HasNode(start,postEnd) && tree.HasNode(end+1,postEnd))
{
newTree.AddNode( start, end,
tree.GetNodes(start,postEnd)[0]->GetLabel()
+ "//" +
tree.GetNodes(end+1,postEnd)[0]->GetLabel() );
done = true;
}
}
if (done) continue;
// if matching a constituent A left-minus constituent B: use A\\B
for(int preStart=start-1; preStart>=0; preStart--)
{
if (tree.HasNode(preStart,end) && tree.HasNode(preStart,start-1))
{
// cerr << "\tadding " << tree.GetNodes(preStart,end )[0]->GetLabel() << "\\\\" <<tree.GetNodes(preStart,start-1)[0]->GetLabel() << endl;
newTree.AddNode( start, end,
tree.GetNodes(preStart,end )[0]->GetLabel()
+ "\\\\" +
tree.GetNodes(preStart,start-1)[0]->GetLabel() );
done = true;
}
}
if (done) continue;
// if matching three consecutive constituents, use double-plus
// SAMT Level 3, not yet implemented
// else: assign default category _FAIL
if (SAMTLevel>=4)
{
newTree.AddNode( start, end, "_FAIL" );
}
}
}
// adding all new nodes
vector< SyntaxNode* > nodes = newTree.GetAllNodes();
for( int i=0; i<nodes.size(); i++ )
{
tree.AddNode( nodes[i]->GetStart(), nodes[i]->GetEnd(), nodes[i]->GetLabel());
}
}