// get signs with additional args for a known special token const, target pred and target rel
private SignHash getSignsFromWord(
Word w, String specialTokenConst, String targetPred, String targetRel) throws LexException {
Collection<MorphItem> morphItems =
(specialTokenConst == null) ? (Collection<MorphItem>) _words.get(w) : null;
if (morphItems == null) {
// check for special tokens
if (specialTokenConst == null) {
specialTokenConst =
tokenizer.getSpecialTokenConstant(tokenizer.isSpecialToken(w.getForm()));
targetPred = w.getForm();
}
if (specialTokenConst != null) {
Word key = Word.createSurfaceWord(w, specialTokenConst);
morphItems = (Collection<MorphItem>) _words.get(key);
}
// otherwise throw lex exception
if (morphItems == null) throw new LexException(w + " not in lexicon");
}
SignHash result = new SignHash();
for (Iterator<MorphItem> MI = morphItems.iterator(); MI.hasNext(); ) {
getWithMorphItem(w, MI.next(), targetPred, targetRel, result);
}
return result;
}
// look up and apply coarts for w to each sign in result
@SuppressWarnings("unchecked")
private void applyCoarts(Word w, SignHash result) throws LexException {
List<Sign> inputSigns = new ArrayList<Sign>(result.asSignSet());
result.clear();
List<Sign> outputSigns = new ArrayList<Sign>(inputSigns.size());
// for each surface attr, lookup coarts and apply to input signs, storing results in output
// signs
for (Iterator<Pair<String, String>> it = w.getSurfaceAttrValPairs(); it.hasNext(); ) {
Pair<String, String> p = it.next();
String attr = (String) p.a;
if (!_indexedCoartAttrs.contains(attr)) continue;
String val = (String) p.b;
Word coartWord = Word.createWord(attr, val);
SignHash coartResult = getSignsFromWord(coartWord, null, null, null);
for (Iterator<Sign> it2 = coartResult.iterator(); it2.hasNext(); ) {
Sign coartSign = it2.next();
// apply to each input
for (int j = 0; j < inputSigns.size(); j++) {
Sign sign = inputSigns.get(j);
grammar.rules.applyCoart(sign, coartSign, outputSigns);
}
}
// switch output to input for next iteration
inputSigns.clear();
inputSigns.addAll(outputSigns);
outputSigns.clear();
}
// add results back
result.addAll(inputSigns);
}
/**
* For a given word, return all of its surface word's lexical entries. If the word is not listed
* in the lexicon, the tokenizer is consulted to see if it is a special token (date, time, etc.);
* otherwise an exception is thrown. If the word has coarticulations, all applicable
* coarticulation entries are applied to the base word, in an arbitrary order.
*
* @param w the word
* @return a sign hash
* @exception LexException thrown if word not found
*/
public SignHash getSignsFromWord(Word w) throws LexException {
// reduce word to its core, removing coart attrs if any
Word surfaceWord = Word.createSurfaceWord(w);
Word coreWord =
(surfaceWord.attrsIntersect(_coartAttrs))
? Word.createCoreSurfaceWord(surfaceWord, _coartAttrs)
: surfaceWord;
// lookup core word
SignHash result = getSignsFromWord(coreWord, null, null, null);
if (result.size() == 0) {
throw new LexException(coreWord + " not found in lexicon");
}
// return signs if no coart attrs
if (coreWord == surfaceWord) return result;
// otherwise apply coarts for word
applyCoarts(surfaceWord, result);
return result;
}
// get signs using an additional arg for a target rel
private Collection<Sign> getSignsFromPredAndTargetRel(String pred, String targetRel) {
Collection<Word> words = (Collection<Word>) _predToWords.get(pred);
String specialTokenConst = null;
// for robustness, when using supertagger, add words for pred sans sense index
int dotIndex = -1;
if (_supertagger != null
&& !Character.isDigit(pred.charAt(0))
&& // skip numbers
(dotIndex = pred.lastIndexOf('.')) > 0
&& pred.length() > dotIndex + 1
&& pred.charAt(dotIndex + 1) != '_') // skip titles, eg Mr._Smith
{
String barePred = pred.substring(0, dotIndex);
Collection<Word> barePredWords = (Collection<Word>) _predToWords.get(barePred);
if (words == null) words = barePredWords;
else if (barePredWords != null) {
Set<Word> unionWords = new HashSet<Word>(words);
unionWords.addAll(barePredWords);
words = unionWords;
}
}
if (words == null) {
specialTokenConst = tokenizer.getSpecialTokenConstant(tokenizer.isSpecialToken(pred));
if (specialTokenConst == null) return null;
// lookup words with pred = special token const
Collection<Word> specialTokenWords = (Collection<Word>) _predToWords.get(specialTokenConst);
// replace special token const with pred
if (specialTokenWords == null) return null;
words = new ArrayList<Word>(specialTokenWords.size());
for (Iterator<Word> it = specialTokenWords.iterator(); it.hasNext(); ) {
Word stw = it.next();
Word w = Word.createSurfaceWord(stw, pred);
words.add(w);
}
}
List<Sign> retval = new ArrayList<Sign>();
for (Iterator<Word> it = words.iterator(); it.hasNext(); ) {
Word w = it.next();
try {
SignHash signs = getSignsFromWord(w, specialTokenConst, pred, targetRel);
retval.addAll(signs.asSignSet());
}
// shouldn't happen
catch (LexException exc) {
System.err.println("Unexpected lex exception for word " + w + ": " + exc);
}
}
return retval;
}
// given EntriesItem
private void getWithEntriesItem(
Word w,
MorphItem mi,
String stem,
String pred,
String targetPred,
String targetRel,
EntriesItem item,
MacroAdder macAdder,
Map<String, Double> supertags,
Set<String> supertagsFound,
SignHash result) {
// ensure apropos
if (targetPred != null && !targetPred.equals(pred)) return;
if (targetRel != null
&& !targetRel.equals(item.getIndexRel())
&& !targetRel.equals(item.getCoartRel())) return;
if (!item.getActive().booleanValue()) return;
if (mi.excluded(item)) return;
try {
// copy and add macros
Category cat = item.getCat().copy();
macAdder.addMacros(cat);
// replace DEFAULT_VAL with pred, after first
// unifying type of associated nom var(s) with sem class
unifySemClass(cat, mi.getWord().getSemClass());
REPLACEMENT = pred;
cat.deepMap(defaultReplacer);
// check supertag
// TODO: think about earlier checks for efficiency, for grammars where macros and preds don't
// matter
// Double lexprob = null; // nb: skipping lex log probs, don't seem to be helpful
if (supertags != null) {
// skip if not found
String stag = cat.getSupertag();
if (!supertags.containsKey(stag)) return;
// otherwise update found supertags
supertagsFound.add(stag);
// get lex prob
// lexprob = supertags.get(stag);
}
// propagate types of nom vars
propagateTypes(cat);
// handle distrib attrs and inherits-from
propagateDistributiveAttrs(cat);
expandInheritsFrom(cat);
// merge stem, pos, sem class from morph item, plus supertag from cat
Word word = Word.createFullWord(w, mi.getWord(), cat.getSupertag());
// set origin and lexprob
Sign sign = new Sign(word, cat);
sign.setOrigin();
// if (lexprob != null) {
// sign.addData(new SupertaggerAdapter.LexLogProb((float) Math.log10(lexprob)));
// }
// return sign
result.insert(sign);
} catch (RuntimeException exc) {
System.err.println(
"Warning: ignoring entry: "
+ item.getName()
+ " of family: "
+ item.getFamilyName()
+ " for stem: "
+ stem
+ " b/c: "
+ exc.toString());
}
}
/** Loads the lexicon and morph files. */
public void init(URL lexiconUrl, URL morphUrl) throws IOException {
List<Family> lexicon = null;
List<MorphItem> morph = null;
List<MacroItem> macroModel = null;
// load category families (lexicon), morph forms and macros
lexicon = getLexicon(lexiconUrl);
Pair<List<MorphItem>, List<MacroItem>> morphInfo = getMorph(morphUrl);
morph = morphInfo.a;
macroModel = morphInfo.b;
// index words; also index stems to words, as default preds
// store indexed coarticulation attrs too
_words = new GroupMap<Word, MorphItem>();
_predToWords = new GroupMap<String, Word>();
_coartAttrs = new HashSet<String>();
_indexedCoartAttrs = new HashSet<String>();
for (MorphItem morphItem : morph) {
Word surfaceWord = morphItem.getSurfaceWord();
_words.put(surfaceWord, morphItem);
_predToWords.put(morphItem.getWord().getStem(), surfaceWord);
if (morphItem.isCoart()) {
Word indexingWord = morphItem.getCoartIndexingWord();
_words.put(indexingWord, morphItem);
Pair<String, String> first = indexingWord.getSurfaceAttrValPairs().next();
_indexedCoartAttrs.add(first.a);
for (Iterator<Pair<String, String>> it = surfaceWord.getSurfaceAttrValPairs();
it.hasNext(); ) {
Pair<String, String> p = it.next();
_coartAttrs.add(p.a);
}
}
}
// index entries based on stem+pos
_stems = new GroupMap<String, Object>();
_posToEntries = new GroupMap<String, EntriesItem[]>();
// index entries by supertag+pos, for supertagging
_stagToEntries = new GroupMap<String, EntriesItem>();
// also index rels and coart rels to preds
_relsToPreds = new GroupMap<String, String>();
_coartRelsToPreds = new GroupMap<String, String>();
// and gather list of attributes used per atomic category type
_catsToAttrs = new GroupMap<String, String>();
_lfAttrs = new HashSet<String>();
// and remember family and ent, names, for checking excluded list on morph items
HashSet<String> familyAndEntryNames = new HashSet<String>();
// index each family
for (Family family : lexicon) {
familyAndEntryNames.add(family.getName());
EntriesItem[] entries = family.getEntries();
DataItem[] data = family.getData();
// for generic use when we get an unknown stem
// from the morphological analyzer
if (!family.isClosed()) {
_posToEntries.put(family.getPOS(), entries);
}
// scan through entries
for (int j = 0; j < entries.length; j++) {
// index
EntriesItem eItem = entries[j];
_stagToEntries.put(eItem.getSupertag() + family.getPOS(), eItem);
if (eItem.getStem().length() > 0) {
_stems.put(eItem.getStem() + family.getPOS(), eItem);
}
try {
// gather features
eItem.getCat().forall(gatherAttrs);
// record names
familyAndEntryNames.add(eItem.getName());
familyAndEntryNames.add(eItem.getQualifiedName());
} catch (RuntimeException exc) {
System.err.println("exception for: " + family.getName() + ": " + exc);
}
}
// scan through data
for (int j = 0; j < data.length; j++) {
DataItem dItem = data[j];
_stems.put(
dItem.getStem() + family.getPOS(), new Pair<DataItem, EntriesItem[]>(dItem, entries));
// index non-default preds to words
if (!dItem.getStem().equals(dItem.getPred())) {
Collection<Word> words = (Collection<Word>) _predToWords.get(dItem.getStem());
if (words == null) {
if (!openlex) {
System.out.print("Warning: couldn't find words for pred '");
System.out.println(dItem.getPred() + "' with stem '" + dItem.getStem() + "'");
}
} else {
for (Iterator<Word> it = words.iterator(); it.hasNext(); ) {
_predToWords.put(dItem.getPred(), it.next());
}
}
}
}
// index rels to preds
// nb: this covers relational (eg @x<GenRel>e) and featural (eg @e<tense>past)
// elementary predications
List<String> indexRels = new ArrayList<String>(3);
String familyIndexRel = family.getIndexRel();
if (familyIndexRel.length() > 0) {
indexRels.add(familyIndexRel);
}
for (int j = 0; j < entries.length; j++) {
EntriesItem eItem = entries[j];
String indexRel = eItem.getIndexRel();
if (indexRel.length() > 0 && !indexRel.equals(familyIndexRel)) {
indexRels.add(indexRel);
}
}
for (Iterator<String> it = indexRels.iterator(); it.hasNext(); ) {
String indexRel = it.next();
// nb: not indexing on entries items, b/c some stems are still defaults
for (int j = 0; j < data.length; j++) {
DataItem dItem = data[j];
_relsToPreds.put(indexRel, dItem.getPred());
}
}
// index coart rels (features, really) to preds
String coartRel = family.getCoartRel();
if (coartRel.length() > 0) {
for (int j = 0; j < data.length; j++) {
_coartRelsToPreds.put(coartRel, data[j].getPred());
}
}
}
// index the macros
_macros = new GroupMap<String, FeatureStructure>();
// nb: could just index MacroItem objects for feature structures too;
// this might be a bit cleaner, but life is short
_macroItems = new HashMap<String, MacroItem>();
for (MacroItem mi : macroModel) {
String macName = mi.getName();
FeatureStructure[] specs = mi.getFeatureStructures();
for (int j = 0; j < specs.length; j++) {
_macros.put(macName, specs[j]);
}
// this is for handling LF part of macros
_macroItems.put(macName, mi);
}
// with morph items, check POS, macro names, excluded list for xref
for (MorphItem morphItem : morph) {
Word w = morphItem.getWord();
if (!openlex
&& !_stems.containsKey(w.getStem() + w.getPOS())
&& !_posToEntries.containsKey(w.getPOS())) {
System.err.println(
"Warning: no entries for stem '"
+ w.getStem()
+ "' and POS '"
+ w.getPOS()
+ "' found for word '"
+ w
+ "'");
}
String[] macroNames = morphItem.getMacros();
for (int j = 0; j < macroNames.length; j++) {
if (!_macroItems.containsKey(macroNames[j])) {
System.err.println(
"Warning: macro "
+ macroNames[j]
+ " not found for word '"
+ morphItem.getWord()
+ "'");
}
}
String[] excludedNames = morphItem.getExcluded();
for (int j = 0; j < excludedNames.length; j++) {
if (!familyAndEntryNames.contains(excludedNames[j])) {
System.err.println(
"Warning: excluded family or entry '"
+ excludedNames[j]
+ "' not found for word '"
+ morphItem.getWord()
+ "'");
}
}
}
}