package issco.nbest; import issco.eval.*; import java.io.*; import org.jdom.Element; import org.jdom.Document; import org.jdom.output.XMLOutputter; import org.jdom.*; /** Abstract class containing methods useful for data pre-processing * required before feature selection or machine learning based reranking. * *@see DataPreprocNfold *@author GEORGESCUL Maria, ISSCO/TIM, ETI, UNIVERSITY OF GENEVA */ abstract public class DataPreproc { private static int noHypotheses = 7; private static final String wordDeliminator = ","; // protected java.util.HashSet responseTypeFeatSet = new java.util.HashSet(); protected java.util.HashSet dialogueMoveFeatSet = new java.util.HashSet(); // Start defining feature attributes protected Feature rank = new Feature(100, true,"rank", -1); protected Feature lenRecWords = new Feature(200, false, "", 0); protected Feature conf = new Feature(300, false,"confidence", 0); protected Feature noWordsResponse = new Feature(400, false,"", 0); protected Feature availableResponse= new Feature(500, false,"", 0); protected Feature available_lf = new Feature(600, false,"lf_context_available", 0); protected Feature available_referent =new Feature(700, false,"referent_available", 0); protected Feature resolution = new Feature(800, false,"resolution", 0); protected Feature elliptical = new Feature(900, false,"elliptical_utterance", 0); protected Feature u_query = new Feature (1000, true,"underconstrained_query", -10); protected Feature i_tense = new Feature (1100, false,"inconsistent_tense", 0); protected Feature responseType = new Feature (1200, false,"response_type", 0); protected Feature dialogueMove = new Feature (1300, false,"dialogue_move", 0); protected Feature definiteMeetingReferent = new Feature (1500, false,"definite_meeting_and_meeting_referent", 0); protected Feature noDialogueMove = new Feature (1600, true,"no_dialogue_move", -50); protected Feature non_indefinite_existential = new Feature (1700, true,xmlTags.non_indefinite_existential, -10); protected Feature non_show_imperative = new Feature (1800, true,"non_show_imperative", -50); protected Feature definite_meeting = new Feature (1900, false, "definite_meeting", 0); protected Feature indefinite_meeting = new Feature (2000, false, "indefinite_meeting", 0); protected Feature indefinite_meeting_and_meeting_referent = new Feature (2100, true, "indefinite_meeting_and_meeting_referent", -2 ); protected Feature meeting_referent_available = new Feature (2200, false, "meeting_referent_available", 0); // end features /* Features used in first experiments: rank (100, true,"rank", -1), lenRecWords (200, false, "", 0), conf (300, true,"confidence", 0), noWordsResponse (400, false,"", 0), availableResponse (500, true,"", 0), available_lf (600, true,"lf_context_available", 0), available_referent (700, true,"referent_available", 0), resolution (800, true,"resolution", 0), elliptical (900, true,"elliptical_utterance", 0), u_query (1000, true,"underconstrained_query", -10), i_tense (1100, true,"inconsistent_tense", 0), responseType (1200, true,"response_type", 0), dialogueMove (1300, true,"dialogue_move", 0), definiteMeetingReferent (1500, true,"definite_meeting_and_meeting_referent", 0), // "definite_meeting_and_meeting_referent", 0), noDialogueMove(1600, true,"no_dialogue_move", -50), non_indefinite_existential(1700, true,xmlTags.non_indefinite_existential, -10), non_show_imperative(1800, true,"non_show_imperative", -50), definite_meeting(1900, true, "definite_meeting", 0), indefinite_meeting(2000, true, "indefinite_meeting", 0), indefinite_meeting_and_meeting_referent(2100, true, "indefinite_meeting_and_meeting_referent", -2 ), meeting_referent_available(2200, true, "meeting_referent_available", 0); */ private final static class xmlTags{ static String nbest = "nbest"; static String nbest_data = "nbest_data"; static String wavfile = "wavfile"; static String correct_words ="correct_words"; static String recognition = "recognition"; static String recognised_words = "recognised_words"; static String confidence = "confidence"; static String rank = "rank"; static String response = "response"; static String response_type = "response_type"; static String lf_context_available = "lf_context_available"; static String referent_list = "referent_list"; static String referent = "referent"; static String dialogue_move = "dialogue_move"; static String resolution = "resolution"; static String elliptical_utterance = "elliptical_utterance"; static String underconstrained_query ="underconstrained_query"; static String inconsistent_tense = "inconsistent_tense"; static String no_dialogue_move = "no_dialogue_move"; static String non_indefinite_existential = "non_indefinite_existential"; static String non_show_imperative ="non_show_imperative"; static String semantically_correct = "semantically_correct"; static String definite_meeting = "definite_meeting"; static String indefinite_meeting ="indefinite_meeting"; static String indefinite_meeting_and_meeting_referent = "indefinite_meeting_and_meeting_referent"; static String meeting_referent_available = "meeting_referent_available"; } /** * Reads the data which is in XML format and * transforms categorical features into numerical-valued features. * That is, any categorical feature having V different values is converted into V separate binary-valued features. * Then writes the (FEATURE, VALUES) PAIRS INTO A FILE HAVING THE FORMAT REQUIRED BY SVMlight (ranking module). * @param String inXMLFileName - name of the input file containing categorical features * @param String outFileName - name of the output file containing numerical-valued features in the format required by the classifier. * @param String targetRankValue - parameter taking two values ("WER" or "SER") specifying whether * Word Error Rate (WER) or Semantic error Rate (SER) should be used as target ranking value */ void transformCategFeatIntoNumFeat(String inFileName, String outFileName, String targetRankValue, boolean ranking){ try{ File f = new File(outFileName); FileOutputStream outS = new FileOutputStream(f); PrintStream outPStream = new PrintStream(outS); try{ Document d = new org.jdom.input.SAXBuilder().build(new File(inFileName)); // PARSE THE XML FILE BY USING THE SAX PARSER java.util.List nbestList = d.getRootElement().getChildren(xmlTags.nbest_data); int noUtt = 0; java.util.ArrayList arrayResTypeVal = new java.util.ArrayList(); java.util.ArrayList arrayDialMoveVal = new java.util.ArrayList(); for (int i = 0; i< nbestList.size(); i++){ Element nbestElem = (Element) nbestList.get(i); noUtt++; // noUtt will represent the "qid" in the SVM input file outPStream.println("# utterance " + noUtt); // In order to COMPUTE WORD ERROR RATE (wer), get the correct transcription Element transcriptionElem = nbestElem.getChild("correct_words"); String transcription = transcriptionElem.getValue(); java.util.List recList = nbestElem.getChildren("recognition"); Element ref_recElem = (Element) recList.get(0); Element ref_dm_Elem = ref_recElem.getChild("dialogue_move"); String ref_dialogue_move = ref_dm_Elem.getValue(); for (int j = 1; j< recList.size(); j++){ Element recElem = (Element) recList.get(j); if (targetRankValue.equalsIgnoreCase("WER")){ // compute WORD ERROR RATE (WER) and use WER as the target ranking value printWER(recElem, outPStream, transcription); } if (targetRankValue.equalsIgnoreCase("SER")){ // compute Semantic Error Rate (SER) and use it as the target ranking value // the reference dialogue move is in recList.get(0). Element dm_Elem = recElem.getChild("dialogue_move"); String hyp_dm = dm_Elem.getValue(); // printSER(outPStream, ref_dialogue_move, hyp_dm); // Element recElem, PrintStream outPStream, String featName, String trueVal printSemanticErrorRate(recElem, outPStream, "semantically_correct", "good" ); } // if ranking is true then prepare data for ranking, i.e. add "qid:uttID" // otherwise prepare data for classification if (ranking) outPStream.print("qid:" + noUtt + " "); // "definite_meeting_and_meeting_referent" if (this.rank.isUsed()){ printRankFeature(recElem, outPStream, rank.getID()); // printRankFeature_3vals(recElem, outPStream, rankID); // printInconsistentUnderconstrained(recElem, outPStream, u_query_ID, i_tense_ID); } if (this.lenRecWords.isUsed()) printLenRecWords(recElem, outPStream, this.lenRecWords.getID()); if (this.conf.isUsed()) printConfidence(recElem, outPStream, this.conf.getID()); if (this.noWordsResponse.isUsed()) printResponseAvailAndLen(recElem, outPStream, this.noWordsResponse.getID(), this.availableResponse.getID()); if (this.available_lf.isUsed()) printLF_Referent(recElem, outPStream, this.available_lf.getID(),this.available_referent.getID()); if (this.resolution.isUsed()){ printResolutionTrivial(recElem, outPStream, this.resolution.getID()); } if (this.elliptical.isUsed()) printElliptical(recElem, outPStream, this.elliptical.getID()); if (this.u_query.isUsed()) printUnderconstrainedQuery(recElem, outPStream, this.u_query.getID()); if (this.i_tense.isUsed()) printInconsistentTense(recElem, outPStream, this.i_tense.getID()); // TRANSFORM the categorical feature response_type into binary-valued features // eg: say_yes // eg: say_list_of_n_referents(3) if (this.responseType.isUsed()) arrayResTypeVal = printResponseType(recElem, outPStream, this.responseType.getID(), arrayResTypeVal); // // TRANSFORM the categorical feature dialogue_move into binary-valued features // eg: referent_from_context,tense_information,utterance_type if (this.dialogueMove.isUsed()) arrayDialMoveVal = printDialogueMove(recElem, outPStream, this.dialogueMove.getID(), arrayDialMoveVal); // System.out.println(" definiteMeetingReferent"); if (this.definiteMeetingReferent.isUsed()){ printFeatureBooleanValue(recElem, outPStream, this.definiteMeetingReferent.getID(), this.definiteMeetingReferent.name ); } if (this.noDialogueMove.isUsed()) { printFeatureBooleanValue(recElem, outPStream, this.noDialogueMove.getID(), this.noDialogueMove.name ); } if (this.non_indefinite_existential.isUsed()){ printFeatureBooleanValue(recElem, outPStream, this.non_indefinite_existential.getID(), this.non_indefinite_existential.name ); }; if (this.non_show_imperative.isUsed()){ printFeatureBooleanValue(recElem, outPStream, this.non_show_imperative.getID(), this.non_show_imperative.name ); } if (this.indefinite_meeting_and_meeting_referent.isUsed()) printFeatureBooleanValue(recElem, outPStream, this.indefinite_meeting_and_meeting_referent.getID(), this.indefinite_meeting_and_meeting_referent.name ); outPStream.println(); } }// end for } catch (IOException eIO) {eIO.printStackTrace();} catch (JDOMException eJDOM) {eJDOM.printStackTrace();} } catch(FileNotFoundException ex1){ ex1.printStackTrace(); } } /** * @deprecated Use transformCategFeatIntoNumFeat instead and set the Feature attributes * Choose only a subset of three features ((rank, underconstrained_query and incosistent_tense) * and write them into the format required by SVMlight. * @param inXMLFileName * @param svmFileName */ void takeSubsetOf3Feat(String inXMLFileName, String svmFileName){ try{ File f = new File(svmFileName); FileOutputStream outS = new FileOutputStream(f); PrintStream outPStream = new PrintStream(outS); final int rankID = this.rank.getID(); final int u_query_ID = this.u_query.getID(); final int i_tense_ID = this.i_tense.getID(); try{ Document d = new org.jdom.input.SAXBuilder().build(new File(inXMLFileName)); // PARSE THE XML FILE BY USING THE SAX PARSER java.util.List nbestList = d.getRootElement().getChildren("nbest_data"); int noUtt = 0; for (int i = 0; i< nbestList.size(); i++){ Element nbestElem = (Element) nbestList.get(i); noUtt++; // noUtt will represent the "qid" in the SVM input file outPStream.println("# utterance " + noUtt); // In order to COMPUTE WORD ERROR RATE (wer), get the correct transcription Element transcriptionElem = nbestElem.getChild("correct_words"); String transcription = transcriptionElem.getValue(); java.util.List recList = nbestElem.getChildren("recognition"); for (int j = 0; j< recList.size(); j++){ Element recElem = (Element) recList.get(j); // TODO: COMPUTE CONCEPT ERROR RATE AS A RANKING. That is, DO EXPERIMENTS IN PARALLEL WITH "SEMANTIC" ERROR RATE (INSTEAD OF WER) AS THE RANKING SCOPE printWER(recElem, outPStream, transcription); // COMPUTE WORD ERROR RATE (WER) and use WER AS A RANKING outPStream.print("qid:" + noUtt + " "); printRankFeature(recElem, outPStream, rankID); printInconsistentTense(recElem, outPStream, i_tense_ID); printUnderconstrainedQuery(recElem, outPStream, u_query_ID); outPStream.println(); } }// end for } catch (IOException eIO) {eIO.printStackTrace();} catch (JDOMException eJDOM) {eJDOM.printStackTrace();} } catch(FileNotFoundException ex1){ ex1.printStackTrace(); } } /** * Choose only a subset of three features ((rank, underconstrained_query and incosistent_tense) * and write them into the format required by SVMlight. * @deprecated Use transformCategFeatIntoNumFeat instead and set the Feature attributes * @param inXMLFileName * @param svmFileName */ void takeSubsetOf3Feat_rank3vals(String inXMLFileName, String svmFileName){ try{ File f = new File(svmFileName); FileOutputStream outS = new FileOutputStream(f); PrintStream outPStream = new PrintStream(outS); final int rankID = this.rank.getID(); final int u_query_ID = this.u_query.getID(); final int i_tense_ID = this.i_tense.getID(); try{ Document d = new org.jdom.input.SAXBuilder().build(new File(inXMLFileName)); // PARSE THE XML FILE BY USING THE SAX PARSER java.util.List nbestList = d.getRootElement().getChildren("nbest_data"); int noUtt = 0; for (int i = 0; i< nbestList.size(); i++){ Element nbestElem = (Element) nbestList.get(i); noUtt++; // noUtt will represent the "qid" in the SVM input file outPStream.println("# utterance " + noUtt); // In order to COMPUTE WORD ERROR RATE (wer), get the correct transcription Element transcriptionElem = nbestElem.getChild("correct_words"); String transcription = transcriptionElem.getValue(); java.util.List recList = nbestElem.getChildren("recognition"); for (int j = 0; j< recList.size(); j++){ Element recElem = (Element) recList.get(j); // TODO: COMPUTE CONCEPT ERROR RATE AS A RANKING. That is, DO EXPERIMENTS IN PARALLEL WITH "SEMANTIC" ERROR RATE (INSTEAD OF WER) AS THE RANKING SCOPE printWER(recElem, outPStream, transcription); // COMPUTE WORD ERROR RATE (WER) and use WER AS A RANKING outPStream.print("qid:" + noUtt + " "); printRankFeature_3vals(recElem, outPStream, rankID); printInconsistentTense(recElem, outPStream, i_tense_ID); printUnderconstrainedQuery(recElem, outPStream, u_query_ID); outPStream.println(); } }// end for } catch (IOException eIO) {eIO.printStackTrace();} catch (JDOMException eJDOM) {eJDOM.printStackTrace();} } catch(FileNotFoundException ex1){ ex1.printStackTrace(); } } /** * Computes Word Error Rate for two utterances and prints the result into the specified output stream. * @param recElem * @param outPStream * @param transcription */ private static void printWER(Element recElem, PrintStream outPStream, String transcription){ Element recWordsElem = recElem.getChild("recognised_words"); String recWords = recWordsElem.getValue(); float wer = computeWordErrorRate(transcription, recWords); outPStream.print(wer + " "); } /** * * @param recElem * @param outPStream * @param responseTypeID * @param arrayTypeValues * @return */ private java.util.ArrayList printResponseType(Element recElem, PrintStream outPStream, int responseTypeID, java.util.ArrayList arrayTypeValues ){ // eg: say_yes // say_list_of_n_referents(3) // TODO: only one feature for: say_list_of_n_referents(3), say_list_of_n_referents(2), say_list_of_n_referents(n), etc Element responseTypeElem = recElem.getChild("response_type"); String responseTypeVal = responseTypeElem.getValue(); if (responseTypeVal.contains("say_list_of_n_referents") ) responseTypeVal = "say_list_of_n_referents"; int responseTypeIdx = arrayTypeValues.indexOf(responseTypeVal); if ( responseTypeIdx == -1 ){ arrayTypeValues.add(responseTypeVal); responseTypeIdx = arrayTypeValues.indexOf(responseTypeVal); } int feat_ID = responseTypeID + responseTypeIdx; if (this.responseTypeFeatSet.contains(feat_ID)) outPStream.print( feat_ID + ":" + 1 + " "); return arrayTypeValues; } /** * * @param recElem * @param outPStream * @param dialogueMoveID * @param arrayDialogueMove * @return */ private java.util.ArrayList printDialogueMove(Element recElem, PrintStream outPStream, int dialogueMoveID, java.util.ArrayList arrayDialogueMove ){ // referent_from_context,tense_information,utterance_type Element dialogueMoveElem = recElem.getChild("dialogue_move"); if (dialogueMoveElem.getContent().size() > 0){ String dialogueMoveVal = dialogueMoveElem.getValue(); String[] dialogueMoveVal2 = dialogueMoveVal.split(","); int[] features = new int[dialogueMoveVal2.length]; for (int i = 0; i< dialogueMoveVal2.length; i++){ int dialogueMoveIdx = arrayDialogueMove.indexOf(dialogueMoveVal2[i]); if ( dialogueMoveIdx == -1 ){ arrayDialogueMove.add(dialogueMoveVal2[i]); dialogueMoveIdx = arrayDialogueMove.indexOf(dialogueMoveVal2[i]); } features[i] = dialogueMoveID + dialogueMoveIdx; } // FEATURES MUST BE IN INCREASING ORDER java.util.Arrays.sort(features); for (int i = 0; i< features.length; i++){ if ((i>0) && (features[i] != features[i-1])) if (this.dialogueMoveFeatSet.contains(features[i])) outPStream.print( features[i] + ":" + 1 + " "); } } return arrayDialogueMove; } /** * * @param recElem * @param outPStream * @param rankID */ private static void printRankFeature(Element recElem, PrintStream outPStream, int rankID){ // rank <-- the position of a hypothesis in the n-best list Element rankElem = recElem.getChild("rank"); Integer rankElemVal = new Integer(rankElem.getValue()); int rankValue = rankElemVal.intValue(); outPStream.print(rankID + ":0." + rankValue + " "); } /** * */ private static void printFeatureBooleanValue(Element recElem, PrintStream outPStream, int featID, String featName){ Element elem = recElem.getChild(featName); Integer elemVal = new Integer(elem.getValue()); int value = elemVal.intValue(); if (value == 0) outPStream.print(featID + ":0.5" + " "); else outPStream.print(featID + ":" + value + " "); } /** * */ private static void printSemanticErrorRate(Element recElem, PrintStream outPStream, String featName, String goodValue){ Element elem = recElem.getChild(featName); String elemVal = elem.getValue(); if (elemVal.equalsIgnoreCase(goodValue)) outPStream.print("0 "); else outPStream.print("1 "); } /** * * @param recElem * @param outPStream * @param rankID */ private static void printRankFeature_3vals(Element recElem, PrintStream outPStream, int rankID){ // rank <-- the position of a hypothesis in the n-best list Element rankElem = recElem.getChild("rank"); Integer rankElemVal = new Integer(rankElem.getValue()); int rankValue = rankElemVal.intValue(); if (rankValue >=3) rankValue = 3; outPStream.print(rankID + ":" + rankValue + " "); } /** * * @param recElem * @param outPStream * @param lenRecWordsID */ private static void printLenRecWords(Element recElem, PrintStream outPStream, int lenRecWordsID){ // number_recognized_words <-- the length in words of the hypothesized utterance String recWordsStr = recElem.getChild("recognised_words").getValue(); int lenRecWords = numberOf(recWordsStr, ",") + 1; outPStream.print(lenRecWordsID + ":0." + lenRecWords + " "); } /** * * @param recElem * @param outPStream * @param confID */ private static void printConfidence(Element recElem, PrintStream outPStream, int confID){ // confidence <-- the confidence assessment given by the recognizer Element confElem = recElem.getChild(xmlTags.confidence); String confidence = confElem.getValue(); if (confidence.equals("")){ System.out.println("Empty confidence for " + recElem.getChild(xmlTags.recognised_words).getValue()); } else{ Integer confElemVal = new Integer(confidence); int confValue = confElemVal.intValue(); outPStream.print(confID + ":0." + confValue + " "); } } /** * * @param recElem * @param outPStream * @param noWordsResponseID * @param availableResponseID */ private static void printResponseAvailAndLen(Element recElem, PrintStream outPStream, int noWordsResponseID, int availableResponseID){ // response_available and response_lenght String responseStr = recElem.getChild(xmlTags.response).getValue(); // response_lenght <-- the number of words inside the response int noWordsInResponse = numberOf(responseStr, " ") + 1; outPStream.print(noWordsResponseID + ":0." + noWordsInResponse + " "); // response_available: indicating whether the system gives a response // possible values: �yes�, �no�, // check whether the response is "sorry" : // if yes then it means that there was no response if (noWordsInResponse == 1){ if (responseStr.equalsIgnoreCase("sorry")){ double availableResponseVal = 0.5; // there was NO response from the system outPStream.print(availableResponseID + ":" + availableResponseVal + " "); } else{ int availableResponseVal = 1; // there was a response from the system outPStream.print(availableResponseID + ":" + availableResponseVal + " "); } } else{ int availableResponseVal = 1; // there was a response from the system outPStream.print(availableResponseID + ":" + availableResponseVal + " "); } } /** * * @param recElem * @param outPStream * @param available_lf_ID * @param available_referent_ID */ private static void printLF_Referent(Element recElem, PrintStream outPStream, int available_lf_ID, int available_referent_ID ){ // lf_context_available // --> indicating whether there was an LF context available for ellipsis resolution, String lf_context = recElem.getChild(xmlTags.lf_context_available).getValue(); if (lf_context.equals("yes")){ double available_lf_val = 1; // there was an available LF // System.out.print(available_lf_ID + ":" + nbestCurrent[available_lf_ID] + " "); outPStream.print(available_lf_ID + ":" + available_lf_val + " "); } else{ double available_lf_val = 0.5; // there was NO LF available // System.out.print(available_lf_ID + ":" + nbestCurrent[available_lf_ID] + " "); outPStream.print(available_lf_ID + ":" + available_lf_val + " "); } // referent_available // --> indicating whether there was a referent of the specified type available // // meeting // attendee // start_time // Element referentElem = recElem.getChild(xmlTags.referent_list); // eg: meeting if (referentElem == null){ double available_referent_val = 0.5; // there was NO available referent outPStream.print(available_referent_ID + ":" + available_referent_val + " "); } else{ if (referentElem.getChildren(xmlTags.referent).size() > 0){ int available_referent_val = 1; // there was an available referent outPStream.print(available_referent_ID + ":" + available_referent_val + " "); } else{ double available_referent_val = 0.5; // there was NO available referent outPStream.print(available_referent_ID + ":" + available_referent_val + " "); } } } /** * * @param recElem * @param outPStream * @param resolution_ID */ private static void printResolutionTrivial(Element recElem, PrintStream outPStream, int resolution_ID){ // resolution_trivial --> indicating whether the ellipsis resolution was trivial or non_trivial // eg: trivial String resolution = recElem.getChild(xmlTags.resolution).getValue(); if (resolution.equals("non_trivial")){ double resolution_val = 0.3; outPStream.print(resolution_ID + ":" + resolution_val + " "); }else{ if (resolution.equals("trivial")){ double resolution_val = 0.6; outPStream.print(resolution_ID + ":" + resolution_val + " "); } else{ double resolution_val = 0.9; outPStream.print(resolution_ID + ":" + resolution_val + " "); } } } /** * * @param recElem * @param outPStream * @param elliptical_ID */ private static void printElliptical(Element recElem, PrintStream outPStream, int elliptical_ID){ // elliptical_utterance --> indicating whether the utterance was elliptical or non-elliptical, // eg: yes String elliptical = recElem.getChild(xmlTags.elliptical_utterance).getValue(); if (elliptical.equals("no")){ double elliptical_val = 0.3; outPStream.print(elliptical_ID + ":" + elliptical_val + " "); } else{ if (elliptical.equals("yes")){ double elliptical_val = 0.6; outPStream.print(elliptical_ID + ":" + elliptical_val + " "); } else{ double elliptical_val = 0.9; outPStream.print(elliptical_ID + ":" + elliptical_val + " "); } } } private static void printInconsistentTense(Element recElem, PrintStream outPStream, int i_tense_ID){ // inconsistent_tense indicating whether tenses were consistent or not according to heuristic. // eg: 0 String it = recElem.getChild(xmlTags.inconsistent_tense).getValue(); if (it.equals("")){} else{ Integer i_tense = new Integer(it); int i_tense_val = i_tense.intValue(); if (i_tense_val==0) outPStream.print(i_tense_ID + ":" + 0.5 + " "); // print 0.5 instead 0 else outPStream.print(i_tense_ID + ":" + i_tense_val + " ");// print 1 } } private static void printUnderconstrainedQuery(Element recElem, PrintStream outPStream, int u_query_ID){ // underconstrained_query indicating whether the query was underconstrained or not underconstrained according to heuristic, // eg: 0 String uq = recElem.getChild(xmlTags.underconstrained_query).getValue(); if (uq.equals("")){ System.out.println("The underconstrained_query feature is not available"); } else{ Integer u_query = new Integer(uq); int u_query_val = u_query.intValue(); if (u_query_val == 0) outPStream.print(u_query_ID + ":" + 0.5 + " "); else outPStream.print(u_query_ID + ":" + u_query_val + " "); } } protected int getNoHyp(){ return noHypotheses; } /** * Computes how many word substitutions you need in order to obtain the "ref" data from "hyp" data * @param ref * @param hyp * @return */ private static float computeWordErrorRate(String ref, String hyp){ WordErrorRate wer = new WordErrorRate(ref, hyp, wordDeliminator); return (wer.computeNumerator() / wer.computeDenominator() ); } /** * The method returns the number of times the "substr" occurs inside the "str" * @param str * @param substr * @return */ private static int numberOf(String str, String substr){ String[] str1 = str.split(","); return (str1.length -1) ; }; /** * Reads the data which is in prologue format and puts (key, value) pairs into an xml format * @param String infileName - name of the input file containing features values in prolog format * @param String outXMLFileName - name of the output xml file * @param noHyp - number of nbest recognition variants */ void transformPrologIntoXML(String inFileName, String outXMLFileName, int noHyp){ Element root = new Element(xmlTags.nbest); java.io.FileInputStream fis; try{ fis = new java.io.FileInputStream(inFileName); } catch(java.io.FileNotFoundException ex){ fis = null; } if (fis !=null){ try { // read the file containing the features/values in prolog format BufferedReader br = new BufferedReader(new InputStreamReader(fis)); try { String nbestCurrent = ""; boolean first_nbest = true; String thisLine; while ((thisLine = br.readLine()) != null) { // while not end of file if (thisLine.contains("%")){} else{ if (thisLine.contains("nbest_data")){ if (first_nbest){ first_nbest = false; } else{ String correctWords = ParsePrologData.getCorrectWords(nbestCurrent); correctWords = correctWords.replace("\\'","{"); correctWords = correctWords.replace("'", ""); correctWords = correctWords.replace("{", "'"); if(! correctWords.contains("out-of-coverage")){ // Parse the content of the previous nbestCurrent and put it into the xml format Element item1 = new Element(xmlTags.nbest_data); String wavFile = ParsePrologData.getWavFile(nbestCurrent); Element wavFileItem = new Element(xmlTags.wavfile); wavFileItem.addContent(wavFile); item1.addContent(wavFileItem); Element correctWordsItem = new Element(xmlTags.correct_words); correctWordsItem.addContent(correctWords); item1.addContent(correctWordsItem); item1 = getFeat4OneUtt(item1, noHyp, nbestCurrent, "confidence"); // System.out.println("After proc the utterance"); root.addContent(item1); } // END HERE: Parse the content of the previous nbestCurrent // and put it into the xml format // THEN: simply reset the content of nbestCurrent nbestCurrent = ""; } } nbestCurrent += thisLine; } // end if line does not contain % } // end while } catch (Exception e) { System.out.println("Error: " + e); } } // end try catch (Exception e) { System.out.println("Error: " + e); } try{ FileOutputStream fos = new FileOutputStream (outXMLFileName); XMLOutputter outputter = new XMLOutputter(); outputter.output(new Document(root), fos); } catch(java.io.FileNotFoundException ex) { System.out.println("Problems when creating the xml file."); ex.printStackTrace(); } catch(java.io.IOException ex) { ex.printStackTrace(); } } } /** * Gets all features for one utterance and writes them into * an XML element. This static private method is used * by transformPrologIntoXML method. * @param item1 * @param noHyp * @param nbestCurrent * @param firstFeatureName * @return XML element containing the entire set of features * @see transformPrologIntoXML method * */ private Element getFeat4OneUtt(Element item1, int noHyp, String nbestCurrent, String firstFeatureName){ int i1 = 0; boolean stillRecVariants = true; while ( (i1 < noHyp) && stillRecVariants ){ i1++; Element recognitionItem = new Element(xmlTags.recognition); // get into "recCurrent" the substring containing the features for one of the nbest variants int idxRecStart = nbestCurrent.indexOf(firstFeatureName); int idxRecEnd = nbestCurrent.indexOf(firstFeatureName, idxRecStart + 1); if (idxRecEnd < 1){ idxRecEnd = nbestCurrent.length(); stillRecVariants = false; } String recCurrent = nbestCurrent.substring(idxRecStart + 1, idxRecEnd); nbestCurrent = nbestCurrent.substring(idxRecEnd, nbestCurrent.length()); String recognisedWords = ParsePrologData.getRecWords(recCurrent); recognisedWords = recognisedWords.replace("\\'","{"); recognisedWords = recognisedWords.replace("'", ""); recognisedWords = recognisedWords.replace("{", "'"); Element recWordsItem = new Element(xmlTags.recognised_words); recWordsItem.addContent(recognisedWords); recognitionItem.addContent(recWordsItem); String confidenceValue = ParsePrologData.getFirstFeature(recCurrent, "confidence"); confidenceValue = confidenceValue.replace("(",""); confidenceValue = confidenceValue.replace(")",""); Element confidenceItem = new Element(xmlTags.confidence); confidenceItem.addContent(confidenceValue); recognitionItem.addContent(confidenceItem); String rankValue = ParsePrologData.getFeatureValue(recCurrent, "rank"); rankValue = rankValue.replace("(",""); rankValue = rankValue.replace(")",""); Element rankItem = new Element(xmlTags.rank); rankItem.addContent(rankValue); recognitionItem.addContent(rankItem); String response = ParsePrologData.getResponse(recCurrent); response = response.replace("\\'","{"); response = response.replace("'", ""); response = response.replace("{", "'"); Element responseItem = new Element(xmlTags.response); responseItem.addContent(response); recognitionItem.addContent(responseItem); String responseType = ParsePrologData.getResponseType(recCurrent); Element responseTypeItem = new Element(xmlTags.response_type); responseTypeItem.addContent(responseType); recognitionItem.addContent(responseTypeItem); String lfContextAvailable = ParsePrologData.getLFContextAvailable(recCurrent); Element lfContextAvailableItem = new Element(xmlTags.lf_context_available); lfContextAvailableItem.addContent(lfContextAvailable); recognitionItem.addContent(lfContextAvailableItem); if (this.available_referent.used){ // put this feature in the xml file only if is used later // there might be more than one feature called "referent_available" java.util.Vector referentList = ParsePrologData.getReferentAvailable(recCurrent); if (referentList != null){ Element referentListItem = new Element(xmlTags.referent_list); for (int index = 0; index < referentList.size(); index++){ String referent = referentList.get(index); Element referentItem = new Element(xmlTags.referent); referentItem.addContent(referent); referentListItem.addContent(referentItem); } recognitionItem.addContent(referentListItem); } } String dialogueMove = ParsePrologData.getFeatBetweenSquarePar(recCurrent, "dialogue_move="); Element dialogueMoveItem = new Element(xmlTags.dialogue_move); dialogueMoveItem.addContent(dialogueMove); recognitionItem.addContent(dialogueMoveItem); String resolution = ParsePrologData.getFeature(recCurrent, "resolution="); Element resolutionItem = new Element(xmlTags.resolution); resolutionItem.addContent(resolution); recognitionItem.addContent(resolutionItem); String ellipticalUtt = ParsePrologData.getFeature(recCurrent, "elliptical_utterance="); Element ellipticalUttItem = new Element(xmlTags.elliptical_utterance); ellipticalUttItem.addContent(ellipticalUtt); recognitionItem.addContent(ellipticalUttItem); String underconstrainedQuery = ParsePrologData.getLastFeature(recCurrent, "underconstrained_query="); Element underconstrainedQueryItem = new Element(xmlTags.underconstrained_query); underconstrainedQueryItem.addContent(underconstrainedQuery); recognitionItem.addContent(underconstrainedQueryItem); String inconsistent_tense = ParsePrologData.getFeature(recCurrent, "inconsistent_tense="); // inconsistent_tense = inconsistent_tense.replace("]", ""); Element inconsistent_tenseItem = new Element(xmlTags.inconsistent_tense); inconsistent_tenseItem.addContent(inconsistent_tense); recognitionItem.addContent(inconsistent_tenseItem); String valueAux = ParsePrologData.getFeatureValue(recCurrent, this.definiteMeetingReferent.name); Element itemAux = new Element(this.definiteMeetingReferent.name); itemAux.addContent(valueAux); recognitionItem.addContent(itemAux); valueAux = ParsePrologData.getFeatureValue(recCurrent, "no_dialogue_move"); itemAux = new Element(xmlTags.no_dialogue_move); itemAux.addContent(valueAux); recognitionItem.addContent(itemAux); valueAux = ParsePrologData.getFeatureValue(recCurrent, "non_indefinite_existential"); itemAux = new Element(xmlTags.non_indefinite_existential); itemAux.addContent(valueAux); recognitionItem.addContent(itemAux); String valueAux1 = ParsePrologData.getFeatureValue(recCurrent, "non_show_imperative"); Element itemAux1 = new Element(xmlTags.non_show_imperative); itemAux1.addContent(valueAux1); recognitionItem.addContent(itemAux1); valueAux = ParsePrologData.getFeatureValue(recCurrent, "semantically_correct"); valueAux = valueAux.replace(")", ""); valueAux = valueAux.replace("(", ""); itemAux = new Element(xmlTags.semantically_correct); itemAux.addContent(valueAux); recognitionItem.addContent(itemAux); valueAux = ParsePrologData.getFeatureValue(recCurrent, "definite_meeting"); itemAux = new Element(xmlTags.definite_meeting); itemAux.addContent(valueAux); recognitionItem.addContent(itemAux); valueAux = ParsePrologData.getFeatureValue(recCurrent, "indefinite_meeting"); itemAux = new Element(xmlTags.indefinite_meeting); itemAux.addContent(valueAux); recognitionItem.addContent(itemAux); valueAux = ParsePrologData.getFeatureValue(recCurrent, "indefinite_meeting_and_meeting_referent"); itemAux = new Element(xmlTags.indefinite_meeting_and_meeting_referent); itemAux.addContent(valueAux); recognitionItem.addContent(itemAux); valueAux = ParsePrologData.getFeatureValue(recCurrent, "meeting_referent_available"); itemAux = new Element(xmlTags.meeting_referent_available); itemAux.addContent(valueAux); recognitionItem.addContent(itemAux); item1.addContent(recognitionItem); } return item1; } }