private static int parseCoNLL09(
CompletePipelineCMDLineOptions options,
CompletePipeline pipeline,
BufferedReader in,
SentenceWriter writer)
throws IOException, Exception {
List<String> forms = new ArrayList<String>();
forms.add("<root>");
List<Boolean> isPred = new ArrayList<Boolean>();
isPred.add(false);
String str;
int senCount = 0;
while ((str = in.readLine()) != null) {
if (str.trim().equals("")) {
Sentence s;
if (options.desegment) {
s = pipeline.parse(ChineseDesegmenter.desegment(forms.toArray(new String[0])));
} else {
s = options.skipPI ? pipeline.parseOraclePI(forms, isPred) : pipeline.parse(forms);
}
forms.clear();
forms.add("<root>");
isPred.clear();
isPred.add(false); // Root is not a predicate
writer.write(s);
senCount++;
if (senCount % 100 == 0) { // TODO fix output in general, don't
// print to System.out. Wrap a
// printstream in some (static)
// class, and allow people to adjust
// this. While doing this, also add
// the option to make the output
// file be -, ie so it prints to
// stdout. All kinds of errors
// should goto stderr, and nothing
// should be printed to stdout by
// default
System.out.println("Processing sentence " + senCount);
}
} else {
String[] tokens = WHITESPACE_PATTERN.split(str);
forms.add(tokens[1]);
if (options.skipPI) isPred.add(tokens[12].equals("Y"));
}
}
if (forms.size() > 1) { // We have the root token too, remember!
writer.write(pipeline.parse(forms));
senCount++;
}
return senCount;
}
public static void main(String[] args) throws Exception {
long startTime = System.currentTimeMillis();
parseOptions = new ParseOptions(args);
SemanticRoleLabeler srl;
if (parseOptions.useReranker) {
srl = new Reranker(parseOptions);
// srl =
// Reranker.fromZipFile(zipFile,parseOptions.skipPI,parseOptions.global_alfa,parseOptions.global_aiBeam,parseOptions.global_acBeam);
} else {
ZipFile zipFile = new ZipFile(parseOptions.modelFile);
srl =
parseOptions.skipPD
? Pipeline.fromZipFile(zipFile, new Step[] {Step.ai, Step.ac})
: parseOptions.skipPI
? Pipeline.fromZipFile(zipFile, new Step[] {Step.pd, Step.ai, Step.ac /*
* ,Step.po,
* Step.ao
*/})
: Pipeline.fromZipFile(zipFile);
zipFile.close();
}
SentenceWriter writer = null;
if (parseOptions.printXML) writer = new FrameNetXMLWriter(parseOptions.output);
else writer = new CoNLL09Writer(parseOptions.output);
SentenceReader reader =
parseOptions.skipPI
? new SRLOnlyCoNLL09Reader(parseOptions.inputCorpus)
: new DepsOnlyCoNLL09Reader(parseOptions.inputCorpus);
int senCount = 0;
for (Sentence s : reader) {
senCount++;
if (senCount % 100 == 0) System.out.println("Parsing sentence " + senCount);
srl.parseSentence(s);
if (parseOptions.writeCoref) writer.specialwrite(s);
else writer.write(s);
}
writer.close();
reader.close();
long totalTime = System.currentTimeMillis() - startTime;
System.out.println("Done.");
System.out.println(srl.getStatus());
System.out.println();
System.out.println("Total execution time: " + Util.insertCommas(totalTime) + "ms");
}
private static int parseNonSegmentedLineByLine(
CompletePipelineCMDLineOptions options,
CompletePipeline pipeline,
BufferedReader in,
SentenceWriter writer)
throws IOException, Exception {
int senCount = 0;
String str;
while ((str = in.readLine()) != null) {
Sentence s = pipeline.parse(str);
writer.write(s);
senCount++;
if (senCount % 100 == 0) System.out.println("Processing sentence " + senCount); // TODO,
// same
// as
// below.
}
return senCount;
}
private static int parseFullDocument(
CompletePipelineCMDLineOptions options,
CompletePipeline pipeline,
BufferedReader in,
SentenceWriter writer)
throws IOException, Exception {
/** initialize * */
Properties props = new Properties();
props.put("annotators", "tokenize, ssplit, pos, lemma, ner, parse, dcoref");
props.put(
"dcoref.sievePasses",
"MarkRole,"
+ "DiscourseMatch,"
+ "ExactStringMatch,"
+ "RelaxedExactStringMatch,"
+ "PreciseConstructs,"
+ "StrictHeadMatch1,"
+ "StrictHeadMatch2,"
+ "StrictHeadMatch3,"
+ "StrictHeadMatch4,"
+ "RelaxedHeadMatch");
StanfordCoreNLP stanfordpipeline = new StanfordCoreNLP(props);
ExternalProcesses glove = new ExternalProcesses(options.glovedir);
/** read full text * */
int senCount = 0;
String str;
StringBuffer text = new StringBuffer();
while ((str = in.readLine()) != null) {
text.append(str);
text.append("\n");
}
/** document-level preprocessing * */
Annotation document = new Annotation(text.toString());
stanfordpipeline.annotate(document);
Map<String, Double[]> word2vecs = glove.createvecs(document);
Corpus c = new Corpus("tmp");
/** sentence-level preprocessing * */
for (CoreMap sentence : document.get(SentencesAnnotation.class)) {
StringBuffer posOutput = new StringBuffer();
for (CoreLabel token : sentence.get(TokensAnnotation.class)) {
if (posOutput.length() > 0) {
posOutput.append(" ");
}
posOutput.append(token.word());
posOutput.append("_");
posOutput.append(token.tag());
}
String parse =
ExternalProcesses.runProcess(
"nc " + options.mstserver.replaceAll(":", " "), posOutput.toString());
parse = parse.replaceAll("-\t-", "_\t_\n@#").replaceAll("@#\t", "").replaceAll("@#", "");
String[] lines = parse.split("\n");
String[] words = new String[lines.length + 1];
String[] lemmas = new String[lines.length + 1];
String[] tags = new String[lines.length + 1];
String[] morphs = new String[lines.length + 1];
int[] heads = new int[lines.length];
String[] deprels = new String[lines.length];
for (int i = 1; i < words.length; i++) {
String[] parts = lines[i - 1].split("\t");
words[i] = sentence.get(TokensAnnotation.class).get(i - 1).word();
tags[i] = sentence.get(TokensAnnotation.class).get(i - 1).tag();
lemmas[i] = sentence.get(TokensAnnotation.class).get(i - 1).lemma();
morphs[i] = "_";
heads[i - 1] = Integer.parseInt(parts[6]);
deprels[i - 1] = parts[7];
}
Sentence sen = new Sentence(words, lemmas, tags, morphs);
sen.setHeadsAndDeprels(heads, deprels);
/* add labeled predicates from SEMAFOR */
String json =
ExternalProcesses.runProcess("nc " + options.semaforserver.replaceAll(":", " "), parse);
Pattern pred_frame =
Pattern.compile(
"\\{\"target\":\\{\"name\":\"([A-Za-z_]*)\",\"spans\":\\[\\{\"start\":([0-9]*),\"");
Matcher m = pred_frame.matcher(json);
while (m.find()) {
String frame = m.group(1);
int index = Integer.parseInt(m.group(2));
System.out.println(index + "\t" + frame);
sen.makePredicate(index + 1);
((Predicate) sen.get(index + 1)).setSense(frame);
}
for (Word w : sen)
if (word2vecs.containsKey(w.getForm().toLowerCase()))
w.setRep(word2vecs.get(w.getForm().toLowerCase()));
new CorpusSentence(sen, c);
}
/* add coref output to corpus */
Map<Integer, CorefChain> coref = document.get(CorefChainAnnotation.class);
int num = 1;
for (Map.Entry<Integer, CorefChain> entry : coref.entrySet()) {
CorefChain cc = entry.getValue();
// skip singleton mentions
if (cc.getMentionsInTextualOrder().size() == 1) continue;
for (CorefMention m : cc.getMentionsInTextualOrder()) {
c.addMention(c.get(m.sentNum - 1), m.headIndex, num);
}
num++;
}
for (Sentence sen : c) {
pipeline.srl.parseSentence(sen);
senCount++;
if (senCount % 100 == 0) System.out.println("Processing sentence " + senCount);
writer.write(sen);
}
return senCount;
}