/**
* tokenizes a given JCas object's document text using the treetagger program and adds the
* recognized tokens to the JCas object.
*
* @param jcas JCas object supplied by the pipeline
*/
private void tokenize(JCas jcas) {
// read tokenized text to add tokens to the jcas
Logger.printDetail(component, "TreeTagger (tokenization) with: " + ttprops.abbFileName);
EnumSet<Flag> flags = Flag.getSet(ttprops.languageSwitch);
TreeTaggerTokenizer ttt;
ttprops.abbFileName = "english-abbreviations";
if (ttprops.abbFileName != null) {
ttt =
new TreeTaggerTokenizer(
ttprops.rootPath
+ ttprops.fileSeparator
+ "lib"
+ ttprops.fileSeparator
+ ttprops.abbFileName,
flags);
} else {
ttt = new TreeTaggerTokenizer(null, flags);
}
String docText = jcas.getDocumentText().replaceAll("\n\n", "\nEMPTYLINE\n");
List<String> tokenized = ttt.tokenize(docText);
int tokenOffset = 0;
// loop through all the lines in the treetagger output
for (String s : tokenized) {
// charset missmatch fallback: signal (invalid) s
if ((!(s.equals("EMPTYLINE"))) && (jcas.getDocumentText().indexOf(s, tokenOffset) < 0))
throw new RuntimeException(
"Opps! Could not find token "
+ s
+ " in JCas after tokenizing with TreeTagger."
+ " Hmm, there may exist a charset missmatch!"
+ " Default encoding is "
+ Charset.defaultCharset().name()
+ " and should always be UTF-8 (use -Dfile.encoding=UTF-8)."
+ " If input document is not UTF-8 use -e option to set it according to the input, additionally.");
// create tokens and add them to the jcas's indexes.
Token newToken = new Token(jcas);
if (s.equals("EMPTYLINE")) {
newToken.setBegin(tokenOffset);
newToken.setEnd(tokenOffset);
newToken.setPos("EMPTYLINE");
if (annotate_partofspeech) {
newToken.addToIndexes();
}
} else {
newToken.setBegin(jcas.getDocumentText().indexOf(s, tokenOffset));
newToken.setEnd(newToken.getBegin() + s.length());
newToken.addToIndexes();
tokenOffset = newToken.getEnd();
}
}
}
/**
* tokenizes a given JCas object's document text using the chinese tokenization script and adds
* the recognized tokens to the JCas object.
*
* @param jcas JCas object supplied by the pipeline
*/
private void tokenizeChinese(JCas jcas) {
try {
// read tokenized text to add tokens to the jcas
Process proc = ttprops.getChineseTokenizationProcess();
Logger.printDetail(component, "Chinese tokenization: " + ttprops.chineseTokenizerPath);
BufferedReader in = new BufferedReader(new InputStreamReader(proc.getInputStream(), "UTF-8"));
BufferedWriter out =
new BufferedWriter(new OutputStreamWriter(proc.getOutputStream(), "UTF-8"));
Integer tokenOffset = 0;
// loop through all the lines in the stdout output
String[] inSplits = jcas.getDocumentText().split("[\\r\\n]+");
for (String inSplit : inSplits) {
out.write(inSplit);
out.newLine();
out.flush();
// do one initial read
String s = in.readLine();
do {
// break out of the loop if we've read a null
if (s == null) break;
String[] outSplits = s.split("\\s+");
for (String tok : outSplits) {
if (jcas.getDocumentText().indexOf(tok, tokenOffset) < 0)
throw new RuntimeException(
"Could not find token "
+ tok
+ " in JCas after tokenizing with Chinese tokenization script.");
// create tokens and add them to the jcas's indexes.
Token newToken = new Token(jcas);
newToken.setBegin(jcas.getDocumentText().indexOf(tok, tokenOffset));
newToken.setEnd(newToken.getBegin() + tok.length());
newToken.addToIndexes();
tokenOffset = newToken.getEnd();
}
// break out of the loop if the next read will block
if (!in.ready()) break;
s = in.readLine();
} while (true);
}
// clean up
in.close();
proc.destroy();
} catch (Exception e) {
e.printStackTrace();
}
}
/**
* improve german sentences; the treetagger splits german sentences incorrectly on some occasions
*
* @param jcas JCas object supplied by the pipeline
*/
private void improveGermanSentences(JCas jcas) {
/*
* these POS tag sequences will decide whether we want to merge two sentences
* that have (supposedly wrongfully) been split.
*/
HashSet<String[]> posRules = new HashSet<String[]>();
posRules.add(new String[] {"CARD", "\\$.", "NN"});
posRules.add(new String[] {"CARD", "\\$.", "NE"});
FSIterator sentIter = jcas.getAnnotationIndex(Sentence.type).iterator();
// compare two sentences at a time in order to have access to all POS tags
HashSet<HashSet<Sentence>> toMerge = new HashSet<HashSet<Sentence>>();
Sentence prevSent = null, thisSent = null;
while (sentIter.hasNext()) {
if (thisSent == null) {
thisSent = (Sentence) sentIter.next();
continue;
}
prevSent = thisSent;
thisSent = (Sentence) sentIter.next();
/*
* select the last two tokens within the previous sentence as well as the
* first of the current one and check for matches.
*/
Token penultimateToken = null, ultimateToken = null, firstToken = null;
FSIterator tokIter = jcas.getAnnotationIndex(Token.type).subiterator(thisSent);
if (tokIter.hasNext()) {
firstToken = (Token) tokIter.next();
}
tokIter = jcas.getAnnotationIndex(Token.type).subiterator(prevSent);
while (tokIter.hasNext()) {
if (ultimateToken == null) {
ultimateToken = (Token) tokIter.next();
continue;
}
penultimateToken = ultimateToken;
ultimateToken = (Token) tokIter.next();
}
// check that all tokens for further analysis are present. if not: skip
if (penultimateToken == null || ultimateToken == null || firstToken == null) {
continue;
}
// check rules, memorize sentences to be merged
for (String[] posRule : posRules) {
/*
* either one of the pre-defined POS rules fit, or the first token's
* covered text begins with lower case characters.
*/
if ((penultimateToken.getPos() != null
&& penultimateToken.getPos().matches(posRule[0])
&& ultimateToken.getPos() != null
&& ultimateToken.getPos().matches(posRule[1])
&& firstToken.getPos() != null
&& firstToken.getPos().matches(posRule[2]))
|| (firstToken.getCoveredText().matches("^[a-z/].*"))) {
/*
* check whether one of the previous candidate pairs already
* contains one of our sentences.
*/
Boolean candidateExisted = false;
for (HashSet<Sentence> mergeCandidate : toMerge) {
if (mergeCandidate.contains(thisSent) || mergeCandidate.contains(prevSent)) {
// we add both here because sets ignore duplicates
mergeCandidate.add(prevSent);
mergeCandidate.add(thisSent);
candidateExisted = true;
break;
}
}
/*
* if one of the sentences was not already to be merged with another,
* create a new merge candidate set
*/
if (!candidateExisted) {
HashSet<Sentence> newCandidate = new HashSet<Sentence>();
newCandidate.add(prevSent);
newCandidate.add(thisSent);
toMerge.add(newCandidate);
}
break; // don't need to do the next rules; already merging.
}
}
}
// iterate over the previously collected merge candidates
for (HashSet<Sentence> mergeCandidate : toMerge) {
// find the earliest beginning and latest end for the set of sentences
Integer beginIndex = Integer.MAX_VALUE, endIndex = Integer.MIN_VALUE;
Sentence mergedSent = new Sentence(jcas);
for (Sentence s : mergeCandidate) {
if (s.getBegin() < beginIndex) {
beginIndex = s.getBegin();
}
if (s.getEnd() > endIndex) {
endIndex = s.getEnd();
}
s.removeFromIndexes();
}
// set values, add to jcas
mergedSent.setBegin(beginIndex);
mergedSent.setEnd(endIndex);
mergedSent.addToIndexes();
}
}
/**
* based on tokens from the jcas object, adds part of speech (POS) and sentence tags to the jcas
* object using the treetagger program.
*
* @param jcas JCas object supplied by the pipeline
*/
private void doTreeTag(JCas jcas) {
File tmpDocument = null;
BufferedWriter tmpFileWriter;
ArrayList<Token> tokens = new ArrayList<Token>();
try {
// create a temporary file and write our pre-existing tokens to it.
tmpDocument = File.createTempFile("postokens", null);
tmpFileWriter =
new BufferedWriter(new OutputStreamWriter(new FileOutputStream(tmpDocument), "UTF-8"));
// iterate over existing tokens
FSIterator ai = jcas.getAnnotationIndex(Token.type).iterator();
while (ai.hasNext()) {
Token t = (Token) ai.next();
tokens.add(t);
if (!(t.getBegin() == t.getEnd())) {
tmpFileWriter.write(t.getCoveredText() + ttprops.newLineSeparator);
}
}
tmpFileWriter.close();
} catch (IOException e) {
Logger.printError(
"Something went wrong creating a temporary file for the treetagger to process.");
System.exit(-1);
}
// Possible End-of-Sentence Tags
HashSet<String> hsEndOfSentenceTag = new HashSet<String>();
hsEndOfSentenceTag.add("SENT"); // ENGLISH, FRENCH, GREEK,
hsEndOfSentenceTag.add("$."); // GERMAN, DUTCH
hsEndOfSentenceTag.add("FS"); // SPANISH
hsEndOfSentenceTag.add("_Z_Fst"); // ESTONIAN
hsEndOfSentenceTag.add("_Z_Int"); // ESTONIAN
hsEndOfSentenceTag.add("_Z_Exc"); // ESTONIAN
hsEndOfSentenceTag.add("ew"); // CHINESE
try {
Process p = ttprops.getTreeTaggingProcess(tmpDocument);
Logger.printDetail(component, "TreeTagger (pos tagging) with: " + ttprops.parFileName);
BufferedReader in = new BufferedReader(new InputStreamReader(p.getInputStream(), "UTF-8"));
Sentence sentence = null;
// iterate over all the output lines and tokens array (which have the same source and are
// hence symmetric)
int i = 0;
String s = null;
while ((s = in.readLine()) != null) {
// grab a token
Token token = tokens.get(i++);
// modified (Aug 29, 2011): Handle empty tokens (such as empty lines) in input file
while (token.getCoveredText().equals("")) {
// if part of the configuration, also add sentences to the jcas document
if ((annotate_sentences)
&& (token.getPos() != null && token.getPos().equals("EMPTYLINE"))) {
// Establish sentence structure
if (sentence == null) {
sentence = new Sentence(jcas);
sentence.setBegin(token.getBegin());
}
// Finish current sentence if end-of-sentence pos was found or document ended
sentence.setEnd(token.getEnd());
if (sentence.getBegin() < sentence.getEnd()) {
sentence.addToIndexes();
}
// Make sure current sentence is not active anymore so that a new one might be created
sentence = null;
// sentence = new Sentence(jcas);
}
token.removeFromIndexes();
token = tokens.get(i++);
}
// remove tokens, otherwise they are in the index twice
token.removeFromIndexes();
// set part of speech tag and add to indexes again
if (!(token.getCoveredText().equals(""))) {
token.setPos(s);
token.addToIndexes();
}
// if part of the configuration, also add sentences to the jcas document
if (annotate_sentences) {
// Establish sentence structure
if (sentence == null) {
sentence = new Sentence(jcas);
sentence.setBegin(token.getBegin());
}
// Finish current sentence if end-of-sentence pos was found or document ended
if (hsEndOfSentenceTag.contains(s) || i == tokens.size()) {
sentence.setEnd(token.getEnd());
sentence.addToIndexes();
// Make sure current sentence is not active anymore so that a new one might be created
sentence = null;
// sentence = new Sentence(jcas);
}
}
}
while (i < tokens.size()) {
if (!(sentence == null)) {
sentence.setEnd(tokens.get(tokens.size() - 1).getEnd());
sentence.addToIndexes();
}
Token token = tokens.get(i++);
if (token.getPos() != null && token.getPos().equals("EMPTYLINE")) {
token.removeFromIndexes();
}
}
in.close();
p.destroy();
} catch (Exception e) {
e.printStackTrace();
} finally {
// Delete temporary files
tmpDocument.delete();
}
}