Add a FLAT relation for phrases such as 'en masse' which aren't consi…

…dered a FIXED or MWT expression. Addresses a tiny part of UniversalDependencies/docs#717

src/edu/stanford/nlp/trees/CoordinationTransformer.java

@@ -105,6 +105,11 @@ public Tree transformTree(Tree t) {
         log.info("After MWETransform:               " + t);
       }
 
+      t = MWFlatTransform(t);
+      if (VERBOSE) {
+        log.info("After MWFlatTransform:            " + t);
+      }
+
       t = prepCCTransform(t);
       if (VERBOSE) {
         log.info("After prepCCTransform:               " + t);
@@ -688,7 +693,6 @@ private static Tree findCCparent(Tree t, Tree root) {
     TregexPattern.compile("@QP|XS < ((JJR|RBR||RB|RP|IN=node1 < /^(?i)(up)$/) $+ (IN|TO=node2 < /^(?i)to$/))"), // up to
     TregexPattern.compile("@QP < ((JJR|RBR|RB|RP|IN=node1 < /^(?i)up$/) $+ (IN|TO=node2 < /^(?i)to$/))"), //up to
     TregexPattern.compile("@S|SQ|VP|ADVP|PP < (@ADVP < ((IN|RB=node1 < /^(?i)at$/) $+ (JJS|RBS=node2 < /^(?i)least$/)) !$+ (RB < /(?i)(once|twice)/))"), //at least
-
   };
 
   private static final TsurgeonPattern MWE_OPERATION = Tsurgeon.parseOperation("[createSubtree MWE node1 node2] [if exists node3 move node3 $- node2]");
@@ -727,7 +731,20 @@ public static Tree MWETransform(Tree t) {
     return t;
   }
 
-
+  private static final TregexPattern[] MW_FLAT_PATTERNS = {
+    TregexPattern.compile("@NP|ADVP <... {(__=node1 < /^(?i)en$/); (__=node2 < /^(?i)masse$/)}"), // en masse, which is tagged in different ways in PTB
+  };
+
+  private static final TsurgeonPattern MW_FLAT_OPERATION = Tsurgeon.parseOperation("[createSubtree FLAT node1 node2] [if exists node3 move node3 $- node2]");
+
+  public static Tree MWFlatTransform(Tree t) {
+    for (TregexPattern p : MW_FLAT_PATTERNS) {
+      Tsurgeon.processPattern(p, MW_FLAT_OPERATION, t);
+    }
+
+    return t;
+  }
+
   private static final TregexPattern FLAT_PREP_CC_PATTERN = TregexPattern.compile("PP <, (/^(IN|TO)$/=p1 $+ (CC=cc $+ /^(IN|TO)$/=p2))");
   private static final TsurgeonPattern FLAT_PREP_CC_OPERATION = Tsurgeon.parseOperation("[createSubtree PCONJP p1 cc] [move p2 $- cc]");
 

src/edu/stanford/nlp/trees/GrammaticalStructure.java

@@ -164,6 +164,7 @@ public GrammaticalStructure(Tree t, Collection<GrammaticalRelation> relations,
     // avoiding a wasteful copy of the labels.
     Trees.setLeafLabels(treeGraph, t.yield());
     Trees.setLeafTagsIfUnset(treeGraph);
+    //System.out.println(treeGraph.toPrettyString(2));
     if (transformer != null) {
       Tree transformed = transformer.transformTree(treeGraph);
       if (!(transformed instanceof TreeGraphNode)) {
@@ -173,6 +174,7 @@ public GrammaticalStructure(Tree t, Collection<GrammaticalRelation> relations,
     } else {
       this.root = treeGraph;
     }
+    //System.out.println(this.root.toPrettyString(2));
     indexNodes(this.root);
     // add head word and tag to phrase nodes
     if (hf == null) {

src/edu/stanford/nlp/trees/UniversalEnglishGrammaticalRelations.java

@@ -1235,6 +1235,11 @@ private UniversalEnglishGrammaticalRelations() {}
         MODIFIER, "MWE", tregexCompiler,
             "MWE < (IN|TO|RB|NP|NN|JJ|VB|CC|VBZ|VBD|ADVP|PP|JJS|RBS=target)");
 
+  public static final GrammaticalRelation FLAT_EXPRESSION =
+    new GrammaticalRelation(Language.UniversalEnglish, "flat", "flat expression",
+        MODIFIER, "FLAT", tregexCompiler,
+            "FLAT < (IN|FW=target)");
+
   /**
    * The "determiner" grammatical relation.
    * <br>
@@ -1613,6 +1618,7 @@ private UniversalEnglishGrammaticalRelations() {}
             ADVERBIAL_MODIFIER,
             NEGATION_MODIFIER,
             MULTI_WORD_EXPRESSION,
+            FLAT_EXPRESSION,
             DETERMINER,
             PREDETERMINER,
             PRECONJUNCT,

src/edu/stanford/nlp/trees/UniversalSemanticHeadFinder.java

@@ -194,6 +194,7 @@ private void ruleChanges() {
 
     // Special constituent for multi-word expressions
     nonTerminalInfo.put("MWE", new String[][]{{"left"}});
+    nonTerminalInfo.put("FLAT", new String[][]{{"left"}});
 
     nonTerminalInfo.put("PCONJP", new String[][]{{"left"}});