answerparser.cpp

OSB-PIK-OpenVXI-3.0.0源代码 “中国XML论坛 - 专业的XML技术讨论区--XML在语音技术中的应用”

 
 
 // This sets all of the application.lastresult$ array when <instance> contains
 // a complex result.
 //
 bool NLSMLSetVars(AnswerTranslator & translator, SimpleLogger & log,
                   int nbest, const XMLCh * confidence,
                   const XMLCh * utterance, const XMLCh * mode,
                   const DOMNode * interp)
 {
   std::basic_ostringstream<VXIchar> out;
   out << L"application.lastresult$[" << nbest << L"]";
   translator.EvaluateExpression(out.str() + L" = new Object();");
 
   // Set everything but interpretation.
   NLSMLSetSomeVars(out.str(), translator, confidence, utterance, mode);
 
   // Set 'interpretation'.
   vxistring path = out.str() + L".interpretation";
   return NLSMLSetInterp(translator, log, path, interp);
 }
 
 // This function set up lastresult$ n best
 //
 static int SetUpNBestList(int *nbest, bool *isAmbigous, SimpleLogger & log, 
                           AnswerTranslator & translator, AnswerHolder & ans) 
 {
   // if there is an instance of DOMElement type
   // setting these instance
   int index = 0;
   if( ans.instanceStrList.empty() ) {
     AnswerHolder::DOMLIST::iterator pos;
     for(pos = ans.instanceDomList.begin(); 
         pos != ans.instanceDomList.end(); ++pos, ++index) {
 
       *nbest += index;
       if (index > 0)
         *isAmbigous = true;
 
       if (log.IsLogging(2)) {
          log.StartDiagnostic(2) << L"SetUpNBestList(" << *nbest
          << L", " << XMLChToVXIchar(ans.grammarid).c_str() << L")";
          log.EndDiagnostic();
       }     
 
       if (!NLSMLSetVars(translator, log, *nbest, ans.confidence, 
                         ans.utterance.c_str(), ans.inputmode, (*pos)))
       {
         log.LogDiagnostic(0, L"AnswerParser: <instance> element "
                       L"processing failed");
         return -1;
       }                    
     }  
   } 
   else {
     // Use the input text(utterance) as instance
     AnswerHolder::STRLIST::iterator pos;
     for(pos = ans.instanceStrList.begin(); 
         pos != ans.instanceStrList.end(); ++pos, ++index) {
 
       *nbest += index;
       if (index > 0)
         *isAmbigous = true;
 
       if (log.IsLogging(2)) {
          log.StartDiagnostic(2) << L"SetUpNBestList(" << *nbest
          << L", " << XMLChToVXIchar(ans.grammarid).c_str() << L")";
          log.EndDiagnostic();
       }
 
       NLSMLSetVars(translator, *nbest, ans.confidence, 
                    ans.utterance.c_str(), ans.inputmode, (*pos).c_str());
     }
   } 
   ++(*nbest);
   return 0;                             
 }
 
 // This function converts an NLSML result into ECMAScript.
 //
 int AnswerParser::ProcessNLSML(SimpleLogger & log,
                                AnswerTranslator & translator,
                                GrammarManager & gm, int maxnbest,
                                vxistring & bestmatch)
 {
   DOMNode * doc = nlsmlParser->getDocument();
   if (doc == NULL || !doc->hasChildNodes()) {
     log.LogDiagnostic(0, L"AnswerParser::ProcessNLSML - "
                       L"Document parse failed");
     return -1;
   }
 
   // (1) <result> element
   DOMElement * result = NULL;
   DOMNode * temp = NULL;
   for (temp = doc->getFirstChild(); temp != NULL;
        temp = temp->getNextSibling())
   {
     switch (temp->getNodeType()) {
     case DOMNode::ELEMENT_NODE:
       if (result == NULL)
         result = reinterpret_cast<DOMElement *>(temp);
       else {
         log.LogDiagnostic(0,L"AnswerParser: document must have a single root");
         return -1;
       }
       break;
     case DOMNode::TEXT_NODE:
       log.LogDiagnostic(0, L"AnswerParser: PCDATA not allowed after <xml>");
       return -1;
     default:
       continue;
     }
   }
 
   if (result == NULL || !Compare(result->getNodeName(), NODE_RESULT)) {
     log.LogDiagnostic(0, L"AnswerParser: document must have a single "
                       L"root named 'result'");
     return -1;
   }
 
   const XMLCh * resultGrammar = result->getAttribute(ATTR_GRAMMAR);
 
   // (2) <interpretation> elements
 
   int nbest = -1;
   ANSWERHOLDERMAP answerMap;
 
   for (DOMNode * temp2 = result->getFirstChild(); temp2 != NULL;
        temp2 = temp2->getNextSibling())
   {
     switch (temp2->getNodeType()) {
     case DOMNode::ELEMENT_NODE:
       break;   
     case DOMNode::TEXT_NODE: 
       // check for white space
       if (IsAllWhiteSpace(temp2->getNodeValue())) continue;               
       log.LogDiagnostic(0, L"AnswerParser: PCDATA not allowed after <result>");
       return -1;
     default:
       continue;
     }
 
     DOMElement * interpretation = reinterpret_cast<DOMElement *>(temp2);
 
     // (2.1) Is this an interpretation element?
     if (!Compare(interpretation->getNodeName(), NODE_INTERPRETATION)) {
       log.LogDiagnostic(0, L"AnswerParser: only <interpretation> elements may "
                         L"follow <result>");
       return -1;
     }
 
     // (2.2) Get the grammar attribute.
     const XMLCh * grammar = interpretation->getAttribute(ATTR_GRAMMAR);
     if (grammar == NULL || grammar[0] == 0) grammar = resultGrammar;
 
     // (2.3) Get the confidence attribute.
     const XMLCh * interpretationConfidence
       = interpretation->getAttribute(ATTR_CONFIDENCE);
 
     // (3) <input> element.
 
     // (3.1) Find the <input> element.  There can be only one.
     DOMElement * input = NULL;
 
     for (temp = interpretation->getFirstChild(); temp != NULL;
          temp = temp->getNextSibling())
     {
       switch (temp->getNodeType()) {
       case DOMNode::TEXT_NODE:
         // check for white space
         if( IsAllWhiteSpace(temp->getNodeValue())) continue;               
         log.LogDiagnostic(0, L"AnswerParser: PCDATA not allowed after "
                           L"<interpretation>");
         return -1;
       case DOMNode::ELEMENT_NODE:
         if (Compare(temp->getNodeName(), NODE_INSTANCE)) continue;
         if (!Compare(temp->getNodeName(), NODE_INPUT)) {
           log.LogDiagnostic(0, L"AnswerParser: Only <input> and <instance> "
                             L"elements are allowed in an <interpretation>");
           return -1;
         }
         if (input != NULL) {
           log.LogDiagnostic(0, L"AnswerParser: Only one <input> element is "
                             L"allowed in an <interpretation>");
           return -1;
         }
         input = reinterpret_cast<DOMElement *>(temp);
         break;
       default:
         continue;
       }
     }
 
     // (3.2) Look at the data inside of <input>.  Is this an event?
 
     xmlcharstring inputText;
     for (temp = input->getFirstChild(); temp != NULL;
          temp = temp->getNextSibling())
     {
       short type = temp->getNodeType();
       switch (type) {
       case DOMNode::CDATA_SECTION_NODE:
       case DOMNode::TEXT_NODE:
         inputText += temp->getNodeValue();
         break;
       case DOMNode::ELEMENT_NODE:
         // NOTE: The nbest count here is one less than the array value.  The
         //       value is not incremented until the <instance> element is
         //       processed.  Thus, the checks are against -1 NOT 0.
         //
         if (Compare(temp->getNodeName(), NODE_NOINPUT)) {
           if (nbest == -1) return 1;
           return 0; // Ignore this and subsequent interpretations.
         }
         if (Compare(temp->getNodeName(), NODE_NOMATCH)) {
           if (nbest == -1) return 2;
           return 0; // Ignore this and subsequent interpretations.
         }
 
         log.LogDiagnostic(0, L"AnswerParser: Only <noinput> and <nomatch> "
                           L"elements or PCDATA is allowed in an <input>");
         return -1;
       default:
         continue;
       }
     }
      
     if (inputText.empty()) {
       log.LogDiagnostic(0, L"AnswerParser: <input> element may not be empty");
       return -1;
     }
 
     if (grammar == NULL) {
       log.LogDiagnostic(0, L"AnswerParser: the grammar attribute must be "
                         L"defined in either <result> or <interpretation>");
       return -1;
     }
 
     // (3.3) Get the confidence.
     const XMLCh * confidence = input->getAttribute(ATTR_CONFIDENCE);
     if (confidence == NULL || confidence[0] == 0)
       confidence = interpretationConfidence;
     if (confidence == NULL || confidence[0] == 0)
       confidence = DEFAULT_CONFIDENCE;
     
     // (3.3.1) convert confidence score to float
     VXIflt32 vConfidence = 0.5;
     std::basic_stringstream<VXIchar> confStream(XMLChToVXIchar(confidence).c_str());
     confStream >> vConfidence;    
     
     // (3.4) Get the input mode.
     const XMLCh * mode = input->getAttribute(ATTR_MODE);
     if (mode == NULL || mode[0] == 0) mode = DEFAULT_MODE;
 
     // (3.5) Determine precedence of this grammar
     
     // (3.5.1) retrieve the grammar for this interpretation
 
     // create the answer info.
     AnswerHolder newanswer(inputText.c_str(), mode, grammar, confidence, gm);
     
     // find this answer grammar
     VXMLElement answerElement;
     if (!gm.FindMatchedElement(XMLChToVXIchar(newanswer.grammarid).c_str(), 
                                answerElement, newanswer.grammarInfo) ||
         answerElement == 0)
     {
       // No match found!
       log.LogError(433);
       return 2;
     }
     
     // find the current confidence in answer map
     ANSWERHOLDERMAP::iterator mit = answerMap.find(vConfidence);
     AnswerHolder *newAnswerPtr = NULL;
     // insert answer with confidence sorted in order
     if( mit == answerMap.end()) {
       // this score is first one in this list, create it
       ANSWERHOLDERVECTOR newlist;
       newlist.push_back(newanswer);
       answerMap[vConfidence] = newlist;
       newAnswerPtr = &(answerMap[vConfidence].back());
     } 
     else {      
       (*mit).second.push_back(newanswer);
       newAnswerPtr = &((*mit).second.back());     
     }         
 
     // (4) Look at each <instance>.
     //
     // At this point, we have DOMStrings with the 'grammar' name, 'confidence'
     // score, input 'mode', and 'inputText'.  We have already verified that
     // only <input> and <instance> nodes exist at this level.
 
     bool hasInstance = false;
 
     for (DOMNode * temp3 = interpretation->getFirstChild(); temp3 != NULL;
          temp3 = temp3->getNextSibling())
     {
       if (temp3->getNodeType() != DOMNode::ELEMENT_NODE) continue;
       if (!Compare(temp3->getNodeName(), NODE_INSTANCE)) continue;
       hasInstance = true;
       ++nbest;
 
       // (4.2) Process the instance
 
       // (4.2.1) Does the instance have data?
 
       DOMElement * instance = reinterpret_cast<DOMElement *>(temp3);
       bool hasInstanceData = false;
 
       for (DOMNode * temp4 = instance->getFirstChild();
            temp4 != NULL && !hasInstanceData;
            temp4 = temp4->getNextSibling())
       {
         short type = temp4->getNodeType();
         switch (type) {
         case DOMNode::CDATA_SECTION_NODE:
         case DOMNode::TEXT_NODE:
         case DOMNode::ELEMENT_NODE:
           newAnswerPtr->SetInstance(instance);
           hasInstanceData = true;
           break;
         default:
           continue;
         }
       }
       
       // (4.2.2) If the contents were empty - use input text as interpretation.
       if (!hasInstanceData) {
         newAnswerPtr->SetInstance(inputText);       
       }
     }
     
     // (4.2.3) If <instance> did not appear - use input text as interpretation.
     if (!hasInstance) {
       newAnswerPtr->SetInstance(inputText);             
     }
   }
 
   // Setting up the lastresult$ structure
   ANSWERHOLDERMAP::iterator it = answerMap.begin();
   if( it == answerMap.end() ) {
     // something is wrong, return error
     log.LogDiagnostic(0, L"AnswerParser: the result structure is corrupted");                         
     return -1;
   }
   else {    
     // go through the vector and sort its precedence using selection sort
     for( it = answerMap.begin(); it != answerMap.end(); ++it ) {
       ANSWERHOLDERVECTOR & a = (*it).second;
       int vsize = a.size();
       if( vsize <= 1 ) continue;
       for(int i = 0; i < vsize; i++ ) {
         int min = i;
         for( int j = i+1; j < vsize; j++)
           if (a[j] < a[min]) min = j;
         AnswerHolder t = a[min]; a[min] = a[i]; a[i] = t;
       }
     }
     
     // go through the vector and setup lastresult structure
     int nbestIndex = 0;
     bool done = false, isAmbigous = false;
     for( it = answerMap.begin(); !done && it != answerMap.end(); ++it ) {
       ANSWERHOLDERVECTOR & a = (*it).second;
       int vsize = a.size();
       for( int i = 0; !done && i < vsize; i++) {
         if( nbestIndex == 0 ) {
           // create the result structure at first element
           translator.EvaluateExpression(L"application.lastresult$ = "
                                         L"new Array();");
           bestmatch = XMLChToVXIchar(a[i].grammarid).c_str();
         }                 
 
         if( SetUpNBestList(&nbestIndex, &isAmbigous, log, translator, a[i]) < 0 )
           return -1;
 
         // check for maxnbest, if there is ambiguity, ignore maxnbest
         // and setup appropriate nbest list
         if( isAmbigous && nbestIndex > maxnbest-1 ) done = true;
         else if( nbestIndex > maxnbest -1 ) done = true;
       }       
     }     
   }
   return 0;
 }