/// <summary>
/// Constructor. WARNING: this will accept an instance of any class derived from TreeBankNode (e.g., PropBankNode), but
/// will return a PropBankNode with only TreeBankNode members instantiated.
/// </summary>
/// <param name="treeBankNode">TreeBankNode from which to construct this PropBankNode</param>
/// <param name="parent">Parent of this PropBank node</param>
protected PropBankNode(TreeBankNode treeBankNode, PropBankNode parent)
: base(treeBankNode, parent, new TreeBankNodeConstructor(PropBankChildConstructor))
{
_label = null;
_information = null;
_labeledNodeCollections = new List <PropBankLabeledNodeCollection>();
}
/// <summary>
/// Loads a VerbInfo list from a file starting at a specific position
/// </summary>
/// <param name="file">File to load list from</param>
/// <param name="position">Position to start at</param>
/// <returns>List of VerbInfo</returns>
private List <VerbInfo> LoadVerbInfo(string file, long position)
{
// read line at specified position
StreamReader loadFile = new StreamReader(file);
loadFile.BaseStream.Position = position;
string infoLine = loadFile.ReadLine();
loadFile.Close();
// parse line...pipe-delimited series of verb entries
List <VerbInfo> viList = new List <VerbInfo>();
int pipeLoc = infoLine.IndexOf('|');
while (pipeLoc != infoLine.Length - 1)
{
// get entry line
int nextPipeLoc = infoLine.IndexOf('|', pipeLoc + 1);
string entry = infoLine.Substring(pipeLoc + 1, nextPipeLoc - pipeLoc - 1);
// parse verb info object
VerbInfo vi = new VerbInfo(entry);
// attach frame if we can find one for the verb
Frame frame;
if (_verbFrame.TryGetValue(vi.Verb, out frame))
{
vi.VerbFrame = frame;
}
// if we didn't get a frame for the verb, throw exception if we're using a gold-standard tagger (non-gold taggers might identify predicates that propbank doesn't know about)
else if (vi.Tagger == "gold")
{
throw new Exception("Failed to get PropBank frame for verb: " + vi.Verb);
}
// set full file path
vi.File = Path.Combine(MrgPath, vi.File);
if (!File.Exists(vi.File))
{
throw new Exception("Invalid PropBank file: " + vi.File);
}
viList.Add(vi);
// start at next pipe
pipeLoc = nextPipeLoc;
}
return(viList);
}
/// <summary>
/// Gets a predicate tree for a PropBank propositions entry
/// </summary>
/// <param name="vi">VerbInfo specifying tree to look up</param>
/// <returns>PropBankNode</returns>
public PropBankNode GetPropBankTree(VerbInfo vi)
{
TreeBankNode parse = GetParseTree(vi.File, vi.SentenceNumber);
PropBankNode predTree = new PropBankNode(parse);
predTree.Information = vi;
// label information is space-delimited
string[] labels = vi.LabeledNodeLocations.Split(' ');
foreach (string label in labels)
{
// label columns are dash-delimited
string[] labelCols = label.Split('-');
// get label type
PropBankNodeLabel.NodeType labelType = PropBankNodeLabel.GetNodeType(labelCols[1]);
// get label feature if any
PropBankNodeLabel.NodeFeature labelFeature = PropBankNodeLabel.NodeFeature.None;
if (labelCols.Length > 2)
{
// sometimes the feature is the actual preposition, so this might fail
string featureStr = labelCols[2];
if (!PropBankNodeLabel.TryGetNodeFeature(featureStr, out labelFeature))
{
// use PRP as the feature, which we have added for this case
featureStr = "PRP";
labelFeature = PropBankNodeLabel.GetNodeFeature(featureStr);
}
if (labelCols.Length > 3)
{
throw new Exception("Missed feature");
}
}
// create new labeled node collection
PropBankLabeledNodeCollection labeledNodes = new PropBankLabeledNodeCollection(new PropBankNodeLabel(labelType, labelFeature, 1));
AddNodesToCollection(predTree, labelCols[0], labeledNodes);
// add to root's list of nodes
predTree.LabeledNodeCollections.Add(labeledNodes);
}
// make sure one of the predicate leaves has the leaf number from the propositions file entry
bool foundMatch = false;
foreach (PropBankNode predicateNode in predTree.PredicateNodes)
{
foreach (PropBankNode leaf in predicateNode.Leaves)
{
if (leaf.LeafNumber == vi.LeafNumber)
{
foundMatch = true;
break;
}
}
if (foundMatch)
{
break;
}
}
if (!foundMatch)
{
throw new Exception("Mismatch between VerbInfo predicate leaf number and actual predicate leaf number");
}
return(predTree);
}
/// <summary>
/// Loads the propositions file
/// </summary>
/// <param name="propsPath">Path to prop.txt file</param>
private void LoadProps(string propsPath)
{
if (!File.Exists(propsPath))
{
throw new Exception("Invalid PropBank propositions file: \"" + propsPath + "\"");
}
// reuse existing index files if present
if (File.Exists(VerbInfoPath) &&
File.Exists(VerbInfoFilePositionPath) &&
File.Exists(MrgSentenceInfoPath) &&
File.Exists(MrgSentenceInfoFilePositionsPath))
{
// load verb info positions
_verbInfoFilePosition = new Dictionary <string, long>();
StreamReader positionsFile = new StreamReader(VerbInfoFilePositionPath);
string line;
while ((line = positionsFile.ReadLine()) != null)
{
// format: position verb
int spaceLoc = line.IndexOf(' ');
_verbInfoFilePosition.Add(line.Substring(spaceLoc + 1), long.Parse(line.Substring(0, spaceLoc)));
}
positionsFile.Close();
// load mrg-sentence verb info positions
_mrgSentInfoFilePosition = new Dictionary <string, Dictionary <int, long> >();
positionsFile = new StreamReader(MrgSentenceInfoFilePositionsPath);
while ((line = positionsFile.ReadLine()) != null)
{
// each line lists a MRG file then a series of pipe-delimited sentence-position pairs...get MRG file first
int pipeLoc = line.IndexOf('|');
string mrgFile = line.Substring(0, pipeLoc);
_mrgSentInfoFilePosition.Add(mrgFile, new Dictionary <int, long>());
// get sentence-position pairs
while (pipeLoc != line.Length - 1)
{
// find next pipe and space
int nextPipe = line.IndexOf('|', pipeLoc + 1);
int spaceLoc = line.IndexOf(' ', pipeLoc + 1);
// get sentence and position
int sent = int.Parse(line.Substring(pipeLoc + 1, spaceLoc - pipeLoc - 1));
long position = long.Parse(line.Substring(spaceLoc + 1, nextPipe - spaceLoc - 1));
// add to index
_mrgSentInfoFilePosition[mrgFile].Add(sent, position);
pipeLoc = nextPipe;
}
}
positionsFile.Close();
return;
}
// maps each verb to a list of VerbInfo objects, each of which stores an annotation instance
Dictionary <string, List <VerbInfo> > verbInfo = new Dictionary <string, List <VerbInfo> >();
// maps each mrg file and sentence number to a list of VerbInfo objects for that sentence
Dictionary <string, Dictionary <int, List <VerbInfo> > > mrgSentInfo = new Dictionary <string, Dictionary <int, List <VerbInfo> > >();
// read propositions file
StreamReader propFile = new StreamReader(propsPath);
string propLine;
while ((propLine = propFile.ReadLine()) != null)
{
VerbInfo vi = new VerbInfo(propLine);
// add to mapping from verbs to their information
verbInfo.EnsureContainsKey(vi.Verb, typeof(List <VerbInfo>));
verbInfo[vi.Verb].Add(vi);
// add to mapping from file-sentence pairs to their information
mrgSentInfo.EnsureContainsKey(vi.File, typeof(Dictionary <int, List <VerbInfo> >));
mrgSentInfo[vi.File].EnsureContainsKey(vi.SentenceNumber, typeof(List <VerbInfo>));
mrgSentInfo[vi.File][vi.SentenceNumber].Add(vi);
}
propFile.Close();
// write verb index to disk and record file positions of verb info lists
_verbInfoFilePosition = new Dictionary <string, long>();
FileStream saveStream = new FileStream(VerbInfoPath, FileMode.Create);
foreach (string verb in verbInfo.Keys)
{
// save position of VerbInfo list
_verbInfoFilePosition.Add(verb, saveStream.Position);
WriteVerbInfoList(verbInfo[verb], saveStream);
}
saveStream.Close();
// save file positions
StreamWriter verbInfoPositionFile = new StreamWriter(VerbInfoFilePositionPath);
foreach (string verb in _verbInfoFilePosition.Keys)
{
verbInfoPositionFile.WriteLine(_verbInfoFilePosition[verb] + " " + verb);
}
verbInfoPositionFile.Close();
// save mrg-sentence info
_mrgSentInfoFilePosition = new Dictionary <string, Dictionary <int, long> >();
saveStream = new FileStream(MrgSentenceInfoPath, FileMode.Create);
foreach (string mrgFile in mrgSentInfo.Keys)
{
// add each sentence, recording position
Dictionary <int, long> sentInfoPosition = new Dictionary <int, long>();
foreach (int sentNum in mrgSentInfo[mrgFile].Keys)
{
// add index of sentence to file position
sentInfoPosition.Add(sentNum, saveStream.Position);
// write VerbInfo list for sentence
WriteVerbInfoList(mrgSentInfo[mrgFile][sentNum], saveStream);
}
_mrgSentInfoFilePosition.Add(mrgFile, sentInfoPosition);
}
saveStream.Close();
// save file positions for MRG file index
StreamWriter mrgSentInfoPositionsFile = new StreamWriter(MrgSentenceInfoFilePositionsPath);
foreach (string mrgFile in _mrgSentInfoFilePosition.Keys)
{
mrgSentInfoPositionsFile.Write(mrgFile);
foreach (int sent in _mrgSentInfoFilePosition[mrgFile].Keys)
{
mrgSentInfoPositionsFile.Write("|" + sent + " " + _mrgSentInfoFilePosition[mrgFile][sent]);
}
mrgSentInfoPositionsFile.WriteLine("|");
}
mrgSentInfoPositionsFile.Close();
}
/// <summary>
/// Marks argument nodes from the current node in the corresponding parse from a different TreeBank. This is used when
/// transferring PropBank annotations to parse trees other than those distributed in the TreeBank (e.g., those produced
/// by an automatic syntactic parser).
/// </summary>
/// <param name="treeBankEngine">Initialized TreeBank engine from which to pull the parse tree to mark PropBank arguments within</param>
/// <returns>PropBank node, or null if all arguments couldn't be minimally transferred to the other parse tree. An argument
/// is minimally transferred if the corresponding node in the other parse tree subsumes precisely the same text as the node in the
/// current parse tree. Sometimes this is not possible due to parse errors.</returns>
public PropBankNode MarkArgumentNodesIn(TreeBankEngine treeBankEngine)
{
if (!IsRoot)
{
throw new Exception("Attempted to transform non-root node");
}
// get mrg file in other tree bank
string treeBankMrgFile = treeBankEngine.GetFullMrgPath(MrgFile.Substring(MrgFile.LastIndexOf(Path.DirectorySeparatorChar) + 1));
// need a PropBank root to mark arguments within
PropBankNode pbRoot = new PropBankNode(treeBankEngine.GetParseTree(treeBankMrgFile, SentenceNumber));
// make sure we got the right sentence
if (pbRoot.SurfaceText != SurfaceText)
{
throw new Exception("Failed to convert root to Charniak-parsed version");
}
// Add information to root. Ignore leaf number and argument info for now - we'll set them at the end.
treeBankMrgFile = treeBankMrgFile.Substring(treeBankEngine.MrgPath.Length);
VerbInfo pbInfo = Information;
pbRoot.Information = new VerbInfo(pbInfo.Verb, treeBankMrgFile, pbInfo.SentenceNumber,
-1, pbInfo.Tagger, pbInfo.RoleSetId,
pbInfo.VForm, pbInfo.VTense, pbInfo.VAspect,
pbInfo.VPerson, pbInfo.VVoice, "");
// transfer all argument node lists
foreach (PropBankLabeledNodeCollection nodeCollection in LabeledNodeCollections)
{
// new node collection
PropBankLabeledNodeCollection otherNodeCollection = new PropBankLabeledNodeCollection(new PropBankNodeLabel(nodeCollection.Label.Type, nodeCollection.Label.Feature, nodeCollection.Label.Confidence));
// get single nodes
foreach (PropBankNode singleNode in nodeCollection.SingleNodes)
{
if (!singleNode.IsNullElement)
{
// get argument node from other parse tree
PropBankNode otherArgNode = (PropBankNode)pbRoot.GetMinimallySubsumingNode(singleNode.FirstToken, singleNode.LastToken);
if (otherArgNode == null)
{
return(null);
}
otherNodeCollection.AddSingleNode(otherArgNode);
}
}
// get split arguments
foreach (List <TreeBankNode> splitArg in nodeCollection.SplitNodes)
{
List <TreeBankNode> otherSplitArg = new List <TreeBankNode>();
// get each node in the split argument
foreach (PropBankNode splitArgNode in splitArg)
{
if (!splitArgNode.IsNullElement)
{
// get split node in other tree
PropBankNode otherSplitArgNode = (PropBankNode)pbRoot.GetMinimallySubsumingNode(splitArgNode.FirstToken, splitArgNode.LastToken);
if (otherSplitArgNode == null)
{
return(null);
}
otherSplitArg.Add(otherSplitArgNode);
}
}
// if only one node of the split arg was non-null, at that node as a single
if (otherSplitArg.Count == 1)
{
otherNodeCollection.AddSingleNode(otherSplitArg.First());
}
// otherwise, add the split arg normally
else if (otherSplitArg.Count >= 2)
{
otherNodeCollection.AddSplitNode(otherSplitArg);
}
}
// add coref list if we found non-null nodes
if (otherNodeCollection.SingleNodes.Count > 0 || otherNodeCollection.SplitNodes.Count > 0)
{
pbRoot.LabeledNodeCollections.Add(otherNodeCollection);
}
}
// return null if we didn't find any argument node lists with non-null nodes
if (pbRoot.LabeledNodeCollections.Count == 0)
{
return(null);
}
// set leaf number and argument information
pbRoot.Information.LeafNumber = pbRoot.PredicateNodes.First().Leaves[0].LeafNumber;
pbRoot.Information.LabeledNodeLocations = pbRoot.LabeledNodeLocations;
return(pbRoot);
}