/**
* for testing -- CURRENTLY BROKEN!!!
*
* @param args input dir and output filename
* @throws IOException
*/
public static void main(String[] args) throws IOException {
if (args.length != 3) {
throw new RuntimeException("args: treebankPath trainNums testNums");
}
ChineseTreebankParserParams ctpp = new ChineseTreebankParserParams();
ctpp.charTags = true;
// TODO: these options are getting clobbered by reading in the
// parser object (unless it's a text file parser?)
Options op = new Options(ctpp);
op.doDep = false;
op.testOptions.maxLength = 90;
LexicalizedParser lp;
try {
FileFilter trainFilt = new NumberRangesFileFilter(args[1], false);
lp = LexicalizedParser.trainFromTreebank(args[0], trainFilt, op);
try {
String filename = "chineseCharTagPCFG.ser.gz";
System.err.println("Writing parser in serialized format to file " + filename + ' ');
System.err.flush();
ObjectOutputStream out = IOUtils.writeStreamFromString(filename);
out.writeObject(lp);
out.close();
System.err.println("done.");
} catch (IOException ioe) {
ioe.printStackTrace();
}
} catch (IllegalArgumentException e) {
lp = LexicalizedParser.loadModel(args[1], op);
}
FileFilter testFilt = new NumberRangesFileFilter(args[2], false);
MemoryTreebank testTreebank = ctpp.memoryTreebank();
testTreebank.loadPath(new File(args[0]), testFilt);
PrintWriter pw =
new PrintWriter(new OutputStreamWriter(new FileOutputStream("out.chi"), "GB18030"), true);
WordCatEquivalenceClasser eqclass = new WordCatEquivalenceClasser();
WordCatEqualityChecker eqcheck = new WordCatEqualityChecker();
EquivalenceClassEval eval = new EquivalenceClassEval(eqclass, eqcheck);
// System.out.println("Preterminals:" + preterminals);
System.out.println("Testing...");
for (Tree gold : testTreebank) {
Tree tree;
try {
tree = lp.parseTree(gold.yieldHasWord());
if (tree == null) {
System.out.println("Failed to parse " + gold.yieldHasWord());
continue;
}
} catch (Exception e) {
e.printStackTrace();
continue;
}
gold = gold.firstChild();
pw.println(Sentence.listToString(gold.preTerminalYield()));
pw.println(Sentence.listToString(gold.yield()));
gold.pennPrint(pw);
pw.println(tree.preTerminalYield());
pw.println(tree.yield());
tree.pennPrint(pw);
// Collection allBrackets = WordCatConstituent.allBrackets(tree);
// Collection goldBrackets = WordCatConstituent.allBrackets(gold);
// eval.eval(allBrackets, goldBrackets);
eval.displayLast();
}
System.out.println();
System.out.println();
eval.display();
}
/**
* transformTree does all language-specific tree transformations. Any parameterizations should be
* inside the specific TreebankLangParserParams class.
*/
@Override
public Tree transformTree(Tree t, Tree root) {
if (t == null || t.isLeaf()) {
return t;
}
String parentStr;
String grandParentStr;
Tree parent;
Tree grandParent;
if (root == null || t.equals(root)) {
parent = null;
parentStr = "";
} else {
parent = t.parent(root);
parentStr = parent.label().value();
}
if (parent == null || parent.equals(root)) {
grandParent = null;
grandParentStr = "";
} else {
grandParent = parent.parent(root);
grandParentStr = grandParent.label().value();
}
String baseParentStr = ctlp.basicCategory(parentStr);
String baseGrandParentStr = ctlp.basicCategory(grandParentStr);
CoreLabel lab = (CoreLabel) t.label();
String word = lab.word();
String tag = lab.tag();
String baseTag = ctlp.basicCategory(tag);
String category = lab.value();
String baseCategory = ctlp.basicCategory(category);
if (t.isPreTerminal()) { // it's a POS tag
List<String> leftAunts =
listBasicCategories(SisterAnnotationStats.leftSisterLabels(parent, grandParent));
List<String> rightAunts =
listBasicCategories(SisterAnnotationStats.rightSisterLabels(parent, grandParent));
// Chinese-specific punctuation splits
if (chineseSplitPunct && baseTag.equals("PU")) {
if (ChineseTreebankLanguagePack.chineseDouHaoAcceptFilter().accept(word)) {
tag = tag + "-DOU";
// System.out.println("Punct: Split dou hao"); // debugging
} else if (ChineseTreebankLanguagePack.chineseCommaAcceptFilter().accept(word)) {
tag = tag + "-COMMA";
// System.out.println("Punct: Split comma"); // debugging
} else if (ChineseTreebankLanguagePack.chineseColonAcceptFilter().accept(word)) {
tag = tag + "-COLON";
// System.out.println("Punct: Split colon"); // debugging
} else if (ChineseTreebankLanguagePack.chineseQuoteMarkAcceptFilter().accept(word)) {
if (chineseSplitPunctLR) {
if (ChineseTreebankLanguagePack.chineseLeftQuoteMarkAcceptFilter().accept(word)) {
tag += "-LQUOTE";
} else {
tag += "-RQUOTE";
}
} else {
tag = tag + "-QUOTE";
}
// System.out.println("Punct: Split quote"); // debugging
} else if (ChineseTreebankLanguagePack.chineseEndSentenceAcceptFilter().accept(word)) {
tag = tag + "-ENDSENT";
// System.out.println("Punct: Split end sent"); // debugging
} else if (ChineseTreebankLanguagePack.chineseParenthesisAcceptFilter().accept(word)) {
if (chineseSplitPunctLR) {
if (ChineseTreebankLanguagePack.chineseLeftParenthesisAcceptFilter().accept(word)) {
tag += "-LPAREN";
} else {
tag += "-RPAREN";
}
} else {
tag += "-PAREN";
// printlnErr("Just used -PAREN annotation");
// printlnErr(word);
// throw new RuntimeException();
}
// System.out.println("Punct: Split paren"); // debugging
} else if (ChineseTreebankLanguagePack.chineseDashAcceptFilter().accept(word)) {
tag = tag + "-DASH";
// System.out.println("Punct: Split dash"); // debugging
} else if (ChineseTreebankLanguagePack.chineseOtherAcceptFilter().accept(word)) {
tag = tag + "-OTHER";
} else {
printlnErr("Unknown punct (you should add it to CTLP): " + tag + " |" + word + "|");
}
} else if (chineseSplitDouHao) { // only split DouHao
if (ChineseTreebankLanguagePack.chineseDouHaoAcceptFilter().accept(word)
&& baseTag.equals("PU")) {
tag = tag + "-DOU";
}
}
// Chinese-specific POS tag splits (non-punctuation)
if (tagWordSize) {
int l = word.length();
tag += "-" + l + "CHARS";
}
if (mergeNNVV && baseTag.equals("NN")) {
tag = "VV";
}
if ((chineseSelectiveTagPA || chineseVerySelectiveTagPA)
&& (baseTag.equals("CC") || baseTag.equals("P"))) {
tag += "-" + baseParentStr;
}
if (chineseSelectiveTagPA && (baseTag.equals("VV"))) {
tag += "-" + baseParentStr;
}
if (markMultiNtag && tag.startsWith("N")) {
for (int i = 0; i < parent.numChildren(); i++) {
if (parent.children()[i].label().value().startsWith("N") && parent.children()[i] != t) {
tag += "=N";
// System.out.println("Found multi=N rewrite");
}
}
}
if (markVVsisterIP && baseTag.equals("VV")) {
boolean seenIP = false;
for (int i = 0; i < parent.numChildren(); i++) {
if (parent.children()[i].label().value().startsWith("IP")) {
seenIP = true;
}
}
if (seenIP) {
tag += "-IP";
// System.out.println("Found VV with IP sister"); // testing
}
}
if (markPsisterIP && baseTag.equals("P")) {
boolean seenIP = false;
for (int i = 0; i < parent.numChildren(); i++) {
if (parent.children()[i].label().value().startsWith("IP")) {
seenIP = true;
}
}
if (seenIP) {
tag += "-IP";
}
}
if (markADgrandchildOfIP && baseTag.equals("AD") && baseGrandParentStr.equals("IP")) {
tag += "~IP";
// System.out.println("Found AD with IP grandparent"); // testing
}
if (gpaAD && baseTag.equals("AD")) {
tag += "~" + baseGrandParentStr;
// System.out.println("Found AD with grandparent " + grandParentStr); // testing
}
if (markPostverbalP && leftAunts.contains("VV") && baseTag.equals("P")) {
// System.out.println("Found post-verbal P");
tag += "^=lVV";
}
// end Chinese-specific tag splits
Label label = new CategoryWordTag(tag, word, tag);
t.setLabel(label);
} else {
// it's a phrasal category
Tree[] kids = t.children();
// Chinese-specific category splits
List<String> leftSis = listBasicCategories(SisterAnnotationStats.leftSisterLabels(t, parent));
List<String> rightSis =
listBasicCategories(SisterAnnotationStats.rightSisterLabels(t, parent));
if (paRootDtr && baseParentStr.equals("ROOT")) {
category += "^ROOT";
}
if (markIPsisterBA && baseCategory.equals("IP")) {
if (leftSis.contains("BA")) {
category += "=BA";
// System.out.println("Found IP sister of BA");
}
}
if (dominatesV && hasV(t.preTerminalYield())) {
// mark categories containing a verb
category += "-v";
}
if (markIPsisterVVorP && baseCategory.equals("IP")) {
// todo: cdm: is just looking for "P" here selective enough??
if (leftSis.contains("VV") || leftSis.contains("P")) {
category += "=VVP";
}
}
if (markIPsisDEC && baseCategory.equals("IP")) {
if (rightSis.contains("DEC")) {
category += "=DEC";
// System.out.println("Found prenominal IP");
}
}
if (baseCategory.equals("VP")) {
// cdm 2008: this used to just check that it startsWith("VP"), but
// I think that was bad because it also matched VPT verb compounds
if (chineseSplitVP == 3) {
boolean hasCC = false;
boolean hasPU = false;
boolean hasLexV = false;
for (Tree kid : kids) {
if (kid.label().value().startsWith("CC")) {
hasCC = true;
} else if (kid.label().value().startsWith("PU")) {
hasPU = true;
} else if (StringUtils.lookingAt(
kid.label().value(), "(V[ACEV]|VCD|VCP|VNV|VPT|VRD|VSB)")) {
hasLexV = true;
}
}
if (hasCC || (hasPU && !hasLexV)) {
category += "-CRD";
// System.out.println("Found coordinate VP"); // testing
} else if (hasLexV) {
category += "-COMP";
// System.out.println("Found complementing VP"); // testing
} else {
category += "-ADJT";
// System.out.println("Found adjoining VP"); // testing
}
} else if (chineseSplitVP >= 1) {
boolean hasBA = false;
for (Tree kid : kids) {
if (kid.label().value().startsWith("BA")) {
hasBA = true;
} else if (chineseSplitVP == 2 && tlp.basicCategory(kid.label().value()).equals("VP")) {
for (Tree kidkid : kid.children()) {
if (kidkid.label().value().startsWith("BA")) {
hasBA = true;
}
}
}
}
if (hasBA) {
category += "-BA";
}
}
}
if (markVPadjunct && baseParentStr.equals("VP")) {
// cdm 2008: This used to use startsWith("VP") but changed to baseCat
Tree[] sisters = parent.children();
boolean hasVPsister = false;
boolean hasCC = false;
boolean hasPU = false;
boolean hasLexV = false;
for (Tree sister : sisters) {
if (tlp.basicCategory(sister.label().value()).equals("VP")) {
hasVPsister = true;
}
if (sister.label().value().startsWith("CC")) {
hasCC = true;
}
if (sister.label().value().startsWith("PU")) {
hasPU = true;
}
if (StringUtils.lookingAt(sister.label().value(), "(V[ACEV]|VCD|VCP|VNV|VPT|VRD|VSB)")) {
hasLexV = true;
}
}
if (hasVPsister && !(hasCC || hasPU || hasLexV)) {
category += "-VPADJ";
// System.out.println("Found adjunct of VP"); // testing
}
}
if (markNPmodNP && baseCategory.equals("NP") && baseParentStr.equals("NP")) {
if (rightSis.contains("NP")) {
category += "=MODIFIERNP";
// System.out.println("Found NP modifier of NP"); // testing
}
}
if (markModifiedNP && baseCategory.equals("NP") && baseParentStr.equals("NP")) {
if (rightSis.isEmpty()
&& (leftSis.contains("ADJP")
|| leftSis.contains("NP")
|| leftSis.contains("DNP")
|| leftSis.contains("QP")
|| leftSis.contains("CP")
|| leftSis.contains("PP"))) {
category += "=MODIFIEDNP";
// System.out.println("Found modified NP"); // testing
}
}
if (markNPconj && baseCategory.equals("NP") && baseParentStr.equals("NP")) {
if (rightSis.contains("CC")
|| rightSis.contains("PU")
|| leftSis.contains("CC")
|| leftSis.contains("PU")) {
category += "=CONJ";
// System.out.println("Found NP conjunct"); // testing
}
}
if (markIPconj && baseCategory.equals("IP") && baseParentStr.equals("IP")) {
Tree[] sisters = parent.children();
boolean hasCommaSis = false;
boolean hasIPSis = false;
for (Tree sister : sisters) {
if (ctlp.basicCategory(sister.label().value()).equals("PU")
&& ChineseTreebankLanguagePack.chineseCommaAcceptFilter()
.accept(sister.children()[0].label().toString())) {
hasCommaSis = true;
// System.out.println("Found CommaSis"); // testing
}
if (ctlp.basicCategory(sister.label().value()).equals("IP") && sister != t) {
hasIPSis = true;
}
}
if (hasCommaSis && hasIPSis) {
category += "-CONJ";
// System.out.println("Found IP conjunct"); // testing
}
}
if (unaryIP && baseCategory.equals("IP") && t.numChildren() == 1) {
category += "-U";
// System.out.println("Found unary IP"); //testing
}
if (unaryCP && baseCategory.equals("CP") && t.numChildren() == 1) {
category += "-U";
// System.out.println("Found unary CP"); //testing
}
if (splitBaseNP && baseCategory.equals("NP")) {
if (t.isPrePreTerminal()) {
category = category + "-B";
}
}
// if (Test.verbose) printlnErr(baseCategory + " " + leftSis.toString()); //debugging
if (markPostverbalPP && leftSis.contains("VV") && baseCategory.equals("PP")) {
// System.out.println("Found post-verbal PP");
category += "=lVV";
}
if ((markADgrandchildOfIP || gpaAD)
&& listBasicCategories(SisterAnnotationStats.kidLabels(t)).contains("AD")) {
category += "^ADVP";
}
if (markCC) {
// was: for (int i = 0; i < kids.length; i++) {
// This second version takes an idea from Collins: don't count
// marginal conjunctions which don't conjoin 2 things.
for (int i = 1; i < kids.length - 1; i++) {
String cat2 = kids[i].label().value();
if (cat2.startsWith("CC")) {
category += "-CC";
}
}
}
Label label = new CategoryWordTag(category, word, tag);
t.setLabel(label);
}
return t;
}
/** @param args */
public static void main(String[] args) {
if (args.length != 3) {
System.err.printf(
"Usage: java %s language filename features%n",
TreebankFactoredLexiconStats.class.getName());
System.exit(-1);
}
Language language = Language.valueOf(args[0]);
TreebankLangParserParams tlpp = language.params;
if (language.equals(Language.Arabic)) {
String[] options = {"-arabicFactored"};
tlpp.setOptionFlag(options, 0);
} else {
String[] options = {"-frenchFactored"};
tlpp.setOptionFlag(options, 0);
}
Treebank tb = tlpp.diskTreebank();
tb.loadPath(args[1]);
MorphoFeatureSpecification morphoSpec =
language.equals(Language.Arabic)
? new ArabicMorphoFeatureSpecification()
: new FrenchMorphoFeatureSpecification();
String[] features = args[2].trim().split(",");
for (String feature : features) {
morphoSpec.activate(MorphoFeatureType.valueOf(feature));
}
// Counters
Counter<String> wordTagCounter = new ClassicCounter<>(30000);
Counter<String> morphTagCounter = new ClassicCounter<>(500);
// Counter<String> signatureTagCounter = new ClassicCounter<String>();
Counter<String> morphCounter = new ClassicCounter<>(500);
Counter<String> wordCounter = new ClassicCounter<>(30000);
Counter<String> tagCounter = new ClassicCounter<>(300);
Counter<String> lemmaCounter = new ClassicCounter<>(25000);
Counter<String> lemmaTagCounter = new ClassicCounter<>(25000);
Counter<String> richTagCounter = new ClassicCounter<>(1000);
Counter<String> reducedTagCounter = new ClassicCounter<>(500);
Counter<String> reducedTagLemmaCounter = new ClassicCounter<>(500);
Map<String, Set<String>> wordLemmaMap = Generics.newHashMap();
TwoDimensionalIntCounter<String, String> lemmaReducedTagCounter =
new TwoDimensionalIntCounter<>(30000);
TwoDimensionalIntCounter<String, String> reducedTagTagCounter =
new TwoDimensionalIntCounter<>(500);
TwoDimensionalIntCounter<String, String> tagReducedTagCounter =
new TwoDimensionalIntCounter<>(300);
int numTrees = 0;
for (Tree tree : tb) {
for (Tree subTree : tree) {
if (!subTree.isLeaf()) {
tlpp.transformTree(subTree, tree);
}
}
List<Label> pretermList = tree.preTerminalYield();
List<Label> yield = tree.yield();
assert yield.size() == pretermList.size();
int yieldLen = yield.size();
for (int i = 0; i < yieldLen; ++i) {
String tag = pretermList.get(i).value();
String word = yield.get(i).value();
String morph = ((CoreLabel) yield.get(i)).originalText();
// Note: if there is no lemma, then we use the surface form.
Pair<String, String> lemmaTag = MorphoFeatureSpecification.splitMorphString(word, morph);
String lemma = lemmaTag.first();
String richTag = lemmaTag.second();
// WSGDEBUG
if (tag.contains("MW")) lemma += "-MWE";
lemmaCounter.incrementCount(lemma);
lemmaTagCounter.incrementCount(lemma + tag);
richTagCounter.incrementCount(richTag);
String reducedTag = morphoSpec.strToFeatures(richTag).toString();
reducedTagCounter.incrementCount(reducedTag);
reducedTagLemmaCounter.incrementCount(reducedTag + lemma);
wordTagCounter.incrementCount(word + tag);
morphTagCounter.incrementCount(morph + tag);
morphCounter.incrementCount(morph);
wordCounter.incrementCount(word);
tagCounter.incrementCount(tag);
reducedTag = reducedTag.equals("") ? "NONE" : reducedTag;
if (wordLemmaMap.containsKey(word)) {
wordLemmaMap.get(word).add(lemma);
} else {
Set<String> lemmas = Generics.newHashSet(1);
wordLemmaMap.put(word, lemmas);
}
lemmaReducedTagCounter.incrementCount(lemma, reducedTag);
reducedTagTagCounter.incrementCount(lemma + reducedTag, tag);
tagReducedTagCounter.incrementCount(tag, reducedTag);
}
++numTrees;
}
// Barf...
System.out.println("Language: " + language.toString());
System.out.printf("#trees:\t%d%n", numTrees);
System.out.printf("#tokens:\t%d%n", (int) wordCounter.totalCount());
System.out.printf("#words:\t%d%n", wordCounter.keySet().size());
System.out.printf("#tags:\t%d%n", tagCounter.keySet().size());
System.out.printf("#wordTagPairs:\t%d%n", wordTagCounter.keySet().size());
System.out.printf("#lemmas:\t%d%n", lemmaCounter.keySet().size());
System.out.printf("#lemmaTagPairs:\t%d%n", lemmaTagCounter.keySet().size());
System.out.printf("#feattags:\t%d%n", reducedTagCounter.keySet().size());
System.out.printf("#feattag+lemmas:\t%d%n", reducedTagLemmaCounter.keySet().size());
System.out.printf("#richtags:\t%d%n", richTagCounter.keySet().size());
System.out.printf("#richtag+lemma:\t%d%n", morphCounter.keySet().size());
System.out.printf("#richtag+lemmaTagPairs:\t%d%n", morphTagCounter.keySet().size());
// Extra
System.out.println("==================");
StringBuilder sbNoLemma = new StringBuilder();
StringBuilder sbMultLemmas = new StringBuilder();
for (Map.Entry<String, Set<String>> wordLemmas : wordLemmaMap.entrySet()) {
String word = wordLemmas.getKey();
Set<String> lemmas = wordLemmas.getValue();
if (lemmas.size() == 0) {
sbNoLemma.append("NO LEMMAS FOR WORD: " + word + "\n");
continue;
}
if (lemmas.size() > 1) {
sbMultLemmas.append("MULTIPLE LEMMAS: " + word + " " + setToString(lemmas) + "\n");
continue;
}
String lemma = lemmas.iterator().next();
Set<String> reducedTags = lemmaReducedTagCounter.getCounter(lemma).keySet();
if (reducedTags.size() > 1) {
System.out.printf("%s --> %s%n", word, lemma);
for (String reducedTag : reducedTags) {
int count = lemmaReducedTagCounter.getCount(lemma, reducedTag);
String posTags =
setToString(reducedTagTagCounter.getCounter(lemma + reducedTag).keySet());
System.out.printf("\t%s\t%d\t%s%n", reducedTag, count, posTags);
}
System.out.println();
}
}
System.out.println("==================");
System.out.println(sbNoLemma.toString());
System.out.println(sbMultLemmas.toString());
System.out.println("==================");
List<String> tags = new ArrayList<>(tagReducedTagCounter.firstKeySet());
Collections.sort(tags);
for (String tag : tags) {
System.out.println(tag);
Set<String> reducedTags = tagReducedTagCounter.getCounter(tag).keySet();
for (String reducedTag : reducedTags) {
int count = tagReducedTagCounter.getCount(tag, reducedTag);
// reducedTag = reducedTag.equals("") ? "NONE" : reducedTag;
System.out.printf("\t%s\t%d%n", reducedTag, count);
}
System.out.println();
}
System.out.println("==================");
}