/// <summary>
/// Gets lexically related words for the current synset. Many of the relations in WordNet are lexical instead of semantic. Whereas
/// the latter indicate relations between entire synsets (e.g., hypernym), the former indicate relations between specific
/// words in synsets. This method retrieves all lexical relations and the words related thereby.
/// </summary>
/// <returns>Mapping from relations to mappings from words in the current synset to related words in the related synsets</returns>
public Dictionary <WordNetEngine.SynSetRelation, Dictionary <string, Set <string> > > GetLexicallyRelatedWords()
{
Dictionary <WordNetEngine.SynSetRelation, Dictionary <string, Set <string> > > relatedWords = new Dictionary <WordNetEngine.SynSetRelation, Dictionary <string, Set <string> > >();
foreach (WordNetEngine.SynSetRelation relation in _lexicalRelations.Keys)
{
DictionaryExtensions.EnsureContainsKey(relatedWords, relation, typeof(Dictionary <string, Set <string> >)); //ZK changed to static method
foreach (SynSet relatedSynSet in _lexicalRelations[relation].Keys)
{
// make sure related synset is initialized
if (!relatedSynSet.Instantiated)
{
relatedSynSet.Instantiate();
}
foreach (int sourceWordIndex in _lexicalRelations[relation][relatedSynSet].Keys)
{
string sourceWord = _words[sourceWordIndex - 1];
DictionaryExtensions.EnsureContainsKey(relatedWords[relation], sourceWord, typeof(Set <string>), false); //ZK changed to static method
foreach (int targetWordIndex in _lexicalRelations[relation][relatedSynSet][sourceWordIndex])
{
string targetWord = relatedSynSet.Words[targetWordIndex - 1];
relatedWords[relation][sourceWord].Add(targetWord);
}
}
}
}
return(relatedWords);
}
/// <summary>
/// Instantiates the current synset. If idSynset is non-null, related synsets references are set to those from
/// idSynset; otherwise, related synsets are created as shells.
/// </summary>
/// <param name="definition">Definition line of synset from data file</param>
/// <param name="idSynset">Lookup for related synsets. If null, all related synsets will be created as shells.</param>
internal void Instantiate(string definition, Dictionary <string, SynSet> idSynset)
{
// don't re-instantiate
if (_instantiated)
{
throw new Exception("Synset has already been instantiated");
}
/* get lexicographer file name...the enumeration lines up precisely with the wordnet spec (see the lexnames file) except that
* it starts with None, so we need to add 1 to the definition line's value to get the correct file name */
int lexicographerFileNumber = int.Parse(GetField(definition, 1)) + 1;
if (lexicographerFileNumber <= 0)
{
throw new Exception("Invalid lexicographer file name number. Should be >= 1.");
}
_lexicographerFileName = (WordNetEngine.LexicographerFileName)lexicographerFileNumber;
// get number of words in the synset and the start character of the word list
int wordStart;
int numWords = int.Parse(GetField(definition, 3, out wordStart), NumberStyles.HexNumber);
wordStart = definition.IndexOf(' ', wordStart) + 1;
// get words in synset
_words = new List <string>(numWords);
for (int i = 0; i < numWords; ++i)
{
int wordEnd = definition.IndexOf(' ', wordStart + 1) - 1;
int wordLen = wordEnd - wordStart + 1;
string word = definition.Substring(wordStart, wordLen);
if (Enumerable.Contains(word, ' ')) //ZK changed to static method
{
throw new Exception("Unexpected space in word: " + word);
}
_words.Add(word);
// skip lex_id field
wordStart = definition.IndexOf(' ', wordEnd + 2) + 1;
}
// get gloss
_gloss = definition.Substring(definition.IndexOf('|') + 1).Trim();
if (Enumerable.Contains(_gloss, '|')) //ZK changed to static method
{
throw new Exception("Unexpected pipe in gloss");
}
// get number and start of relations
int relationCountField = 3 + (_words.Count * 2) + 1;
int relationFieldStart;
int numRelations = int.Parse(GetField(definition, relationCountField, out relationFieldStart));
relationFieldStart = definition.IndexOf(' ', relationFieldStart) + 1;
// grab each related synset
_relationSynSets = new Dictionary <WordNetEngine.SynSetRelation, Set <SynSet> >();
_lexicalRelations = new Dictionary <WordNetEngine.SynSetRelation, Dictionary <SynSet, Dictionary <int, Set <int> > > >();
for (int relationNum = 0; relationNum < numRelations; ++relationNum)
{
string relationSymbol = null;
int relatedSynSetOffset = -1;
WordNetEngine.POS relatedSynSetPOS = WordNetEngine.POS.None;
int sourceWordIndex = -1;
int targetWordIndex = -1;
// each relation has four columns
for (int relationField = 0; relationField <= 3; ++relationField)
{
int fieldEnd = definition.IndexOf(' ', relationFieldStart + 1) - 1;
int fieldLen = fieldEnd - relationFieldStart + 1;
string fieldValue = definition.Substring(relationFieldStart, fieldLen);
// relation symbol
if (relationField == 0)
{
relationSymbol = fieldValue;
}
// related synset offset
else if (relationField == 1)
{
relatedSynSetOffset = int.Parse(fieldValue);
}
// related synset POS
else if (relationField == 2)
{
relatedSynSetPOS = GetPOS(fieldValue);
}
// source/target word for lexical relation
else if (relationField == 3)
{
sourceWordIndex = int.Parse(fieldValue.Substring(0, 2), NumberStyles.HexNumber);
targetWordIndex = int.Parse(fieldValue.Substring(2), NumberStyles.HexNumber);
}
else
{
throw new Exception();
}
relationFieldStart = definition.IndexOf(' ', relationFieldStart + 1) + 1;
}
// get related synset...create shell if we don't have a lookup
SynSet relatedSynSet;
if (idSynset == null)
{
relatedSynSet = new SynSet(relatedSynSetPOS, relatedSynSetOffset, _wordNetEngine);
}
// look up related synset directly
else
{
relatedSynSet = idSynset[relatedSynSetPOS + ":" + relatedSynSetOffset];
}
// get relation
WordNetEngine.SynSetRelation relation = WordNetEngine.GetSynSetRelation(_pos, relationSymbol);
// add semantic relation if we have neither a source nor a target word index
if (sourceWordIndex == 0 && targetWordIndex == 0)
{
DictionaryExtensions.EnsureContainsKey(_relationSynSets, relation, typeof(Set <SynSet>)); //ZK changed to static method
_relationSynSets[relation].Add(relatedSynSet);
}
// add lexical relation
else
{
DictionaryExtensions.EnsureContainsKey(_lexicalRelations, relation, typeof(Dictionary <SynSet, Dictionary <int, Set <int> > >)); //ZK changed to static method
DictionaryExtensions.EnsureContainsKey(_lexicalRelations[relation], relatedSynSet, typeof(Dictionary <int, Set <int> >)); //ZK changed to static method
DictionaryExtensions.EnsureContainsKey(_lexicalRelations[relation][relatedSynSet], sourceWordIndex, typeof(Set <int>)); //ZK changed to static method
if (!_lexicalRelations[relation][relatedSynSet][sourceWordIndex].Contains(targetWordIndex))
{
_lexicalRelations[relation][relatedSynSet][sourceWordIndex].Add(targetWordIndex);
}
}
}
// release the wordnet engine if we have one...don't need it anymore
if (_wordNetEngine != null)
{
_wordNetEngine = null;
}
_instantiated = true;
}
/// <summary>
/// Constructor
/// </summary>
/// <param name="wordNetDirectory">Path to WorNet directory (the one with the data and index files in it)</param>
/// <param name="inMemory">Whether or not to store all data in memory. In-memory storage requires quite a bit of space
/// but it is also very quick. The alternative (false) will cause the data to be searched on-disk with an efficient
/// binary search algorithm.</param>
public WordNetEngine(string wordNetDirectory, bool inMemory)
{
_wordNetDirectory = wordNetDirectory;
_inMemory = inMemory;
_posIndexWordSearchStream = null;
_posSynSetDataFile = null;
if (!System.IO.Directory.Exists(_wordNetDirectory))
{
throw new DirectoryNotFoundException("Non-existent WordNet directory: " + _wordNetDirectory);
}
// get data and index paths
string[] dataPaths = new string[]
{
Path.Combine(_wordNetDirectory, "data.adj"),
Path.Combine(_wordNetDirectory, "data.adv"),
Path.Combine(_wordNetDirectory, "data.noun"),
Path.Combine(_wordNetDirectory, "data.verb")
};
string[] indexPaths = new string[]
{
Path.Combine(_wordNetDirectory, "index.adj"),
Path.Combine(_wordNetDirectory, "index.adv"),
Path.Combine(_wordNetDirectory, "index.noun"),
Path.Combine(_wordNetDirectory, "index.verb")
};
// make sure all files exist
foreach (string path in Enumerable.Union(dataPaths, indexPaths)) //ZK change to static method
{
if (!System.IO.File.Exists(path))
{
throw new FileNotFoundException("Failed to find WordNet file: " + path);
}
}
#region index file sorting
string sortFlagPath = Path.Combine(_wordNetDirectory, ".sorted_for_dot_net");
if (!System.IO.File.Exists(sortFlagPath))
{
/* make sure the index files are sorted according to the current sort order. the index files in the
* wordnet distribution are sorted in the order needed for (presumably) the java api, which uses
* a different sort order than the .net runtime. thus, unless we resort the lines in the index
* files, we won't be able to do a proper binary search over the data. */
foreach (string indexPath in indexPaths)
{
// create temporary file for sorted lines
string tempPath = Path.GetTempFileName();
StreamWriter tempFile = new StreamWriter(tempPath);
// get number of words (lines) in file
int numWords = 0;
StreamReader indexFile = new StreamReader(indexPath);
string line;
while (StreamReaderExtensions.TryReadLine(indexFile, out line)) //ZK change to static method
{
if (!line.StartsWith(" "))
{
++numWords;
}
}
// get lines in file, sorted by first column (i.e., the word)
Dictionary <string, string> wordLine = new Dictionary <string, string>(numWords);
indexFile = new StreamReader(indexPath);
while (StreamReaderExtensions.TryReadLine(indexFile, out line)) //ZK change to static method
// write header lines to temp file immediately
{
if (line.StartsWith(" "))
{
tempFile.WriteLine(line);
}
else
{
// trim useless blank spaces from line and map line to first column
line = line.Trim();
wordLine.Add(line.Substring(0, line.IndexOf(' ')), line);
}
}
// get sorted words
List <string> sortedWords = new List <string>(wordLine.Count);
sortedWords.AddRange(wordLine.Keys);
sortedWords.Sort();
// write lines sorted by word
foreach (string word in sortedWords)
{
tempFile.WriteLine(wordLine[word]);
}
tempFile.Close();
// replace original index file with properly sorted one
System.IO.File.Delete(indexPath);
System.IO.File.Move(tempPath, indexPath);
}
// create flag file, indicating that we've sorted the data
StreamWriter sortFlagFile = new StreamWriter(sortFlagPath);
sortFlagFile.WriteLine("This file serves no purpose other than to indicate that the WordNet distribution data in the current directory has been sorted for use by the .NET API.");
sortFlagFile.Close();
}
#endregion
#region engine init
if (inMemory)
{
// pass 1: get total number of synsets
int totalSynsets = 0;
foreach (string dataPath in dataPaths)
{
// scan synset data file for lines that don't start with a space...these are synset definition lines
StreamReader dataFile = new StreamReader(dataPath);
string line;
while (StreamReaderExtensions.TryReadLine(dataFile, out line)) //ZK change to static method
{
int firstSpace = line.IndexOf(' ');
if (firstSpace > 0)
{
++totalSynsets;
}
}
}
// pass 2: create synset shells (pos and offset only)
_idSynset = new Dictionary <string, SynSet>(totalSynsets);
foreach (string dataPath in dataPaths)
{
POS pos = GetFilePOS(dataPath);
// scan synset data file
StreamReader dataFile = new StreamReader(dataPath);
string line;
while (StreamReaderExtensions.TryReadLine(dataFile, out line)) //ZK change to static method
{
int firstSpace = line.IndexOf(' ');
if (firstSpace > 0)
{
// get offset and create synset shell
int offset = int.Parse(line.Substring(0, firstSpace));
SynSet synset = new SynSet(pos, offset, null);
_idSynset.Add(synset.ID, synset);
}
}
}
// pass 3: instantiate synsets (hooks up relations, set glosses, etc.)
foreach (string dataPath in dataPaths)
{
POS pos = GetFilePOS(dataPath);
// scan synset data file
StreamReader dataFile = new StreamReader(dataPath);
string line;
while (StreamReaderExtensions.TryReadLine(dataFile, out line)) //ZK change to static method
{
int firstSpace = line.IndexOf(' ');
if (firstSpace > 0)
{
// instantiate synset defined on current line, using the instantiated synsets for all references
_idSynset[pos + ":" + int.Parse(line.Substring(0, firstSpace))].Instantiate(line, _idSynset);
}
}
}
// organize synsets by pos and words...also set most common synset for word-pos pairs that have multiple synsets
_posWordSynSets = new Dictionary <POS, Dictionary <string, Set <SynSet> > >();
foreach (string indexPath in indexPaths)
{
POS pos = GetFilePOS(indexPath);
DictionaryExtensions.EnsureContainsKey(_posWordSynSets, pos, typeof(Dictionary <string, Set <SynSet> >)); //ZK change to static method
// scan word index file, skipping header lines
StreamReader indexFile = new StreamReader(indexPath);
string line;
while (StreamReaderExtensions.TryReadLine(indexFile, out line)) //ZK change to static method
{
int firstSpace = line.IndexOf(' ');
if (firstSpace > 0)
{
// grab word and synset shells, along with the most common synset
string word = line.Substring(0, firstSpace);
SynSet mostCommonSynSet;
Set <SynSet> synsets = GetSynSetShells(line, pos, out mostCommonSynSet, null);
// set flag on most common synset if it's ambiguous
if (synsets.Count > 1)
{
_idSynset[mostCommonSynSet.ID].SetAsMostCommonSynsetFor(word);
}
// use reference to the synsets that we instantiated in our three-pass routine above
_posWordSynSets[pos].Add(word, new Set <SynSet>(synsets.Count));
foreach (SynSet synset in synsets)
{
_posWordSynSets[pos][word].Add(_idSynset[synset.ID]);
}
}
}
}
}
else
{
// open binary search streams for index files
_posIndexWordSearchStream = new Dictionary <POS, BinarySearchTextStream>();
foreach (string indexPath in indexPaths)
{
// create binary search stream for index file
BinarySearchTextStream searchStream = new BinarySearchTextStream(indexPath, new BinarySearchTextStream.SearchComparisonDelegate(delegate(object searchWord, string currentLine)
{
// if we landed on the header text, search further down
if (currentLine[0] == ' ')
{
return(1);
}
// get word on current line
string currentWord = currentLine.Substring(0, currentLine.IndexOf(' '));
// compare searched-for word to the current word
return(((string)searchWord).CompareTo(currentWord));
}));
// add search stream for current POS
_posIndexWordSearchStream.Add(GetFilePOS(indexPath), searchStream);
}
// open readers for synset data files
_posSynSetDataFile = new Dictionary <POS, StreamReader>();
foreach (string dataPath in dataPaths)
{
_posSynSetDataFile.Add(GetFilePOS(dataPath), new StreamReader(dataPath));
}
}
#endregion
}
