Print out EmptyIndex in the CoNLLUDocumentWriter if needed on enhanced dependency graphs. Currently, no special technique to separate copy nodes from empty nodes - presumably no conllu file will ever have both

Author: AngledLuffa

src/edu/stanford/nlp/ling/IndexedWord.java

@@ -370,9 +370,14 @@ public String toPrimes() {
     return StringUtils.repeat('\'', copyCount);
   }
 
-  public String toCopyIndex() {
+  public String toCopyOrEmptyIndex() {
+    // TODO: if there's ever a CoNLLU or other situation that has both
+    // copy & empty indices, no idea what to do with such an abomination
     if (copyCount == 0) {
-      if (Double.isNaN(this.pseudoPosition)) {
+      Integer empty = this.get(CoreAnnotations.EmptyIndexAnnotation.class);
+      if (empty != null) {
+        return this.index() + "." + empty;
+      } else if (Double.isNaN(this.pseudoPosition)) {
         return String.valueOf(this.index());
       } else {
         return String.valueOf(this.pseudoPosition);

src/edu/stanford/nlp/trees/ud/CoNLLUDocumentWriter.java

@@ -81,7 +81,7 @@ public String printSemanticGraph(SemanticGraph basicSg, SemanticGraph enhancedSg
 
       // Try to find main governor and additional dependencies
       IndexedWord gov = basicSg.containsVertex(token) ? basicSg.getParent(token) : null;
-      String govIdx = gov != null ? gov.toCopyIndex() : null;
+      String govIdx = gov != null ? gov.toCopyOrEmptyIndex() : null;
       GrammaticalRelation reln = gov != null ? basicSg.getEdge(gov, token).getRelation() : null;
 
       HashMap<String, String> enhancedDependencies = new HashMap<>();
@@ -99,7 +99,7 @@ public String printSemanticGraph(SemanticGraph basicSg, SemanticGraph enhancedSg
           //} else if (edge.getWeight() == 5.0) {
           //    relationString = relationString + ":ENH_CONJ_PROP";
           //}
-          enhancedDependencies.put(parent.toCopyIndex(), relationString);
+          enhancedDependencies.put(parent.toCopyOrEmptyIndex(), relationString);
         }
       } else {
         // add enhanced ones stored with token
@@ -165,7 +165,7 @@ public String printSemanticGraph(SemanticGraph basicSg, SemanticGraph enhancedSg
         lemma = lemma.replaceAll(RRB_PATTERN, ")");
       }
 
-      sb.append(String.format("%s\t%s\t%s\t%s\t%s\t%s\t%s\t%s\t%s\t%s%n", token.toCopyIndex(), word,
+      sb.append(String.format("%s\t%s\t%s\t%s\t%s\t%s\t%s\t%s\t%s\t%s%n", token.toCopyOrEmptyIndex(), word,
                               lemma, upos, pos, featuresString, govIdx, relnName, additionalDepsString, misc));
     }
     sb.append(System.lineSeparator());