Merge branch 'master' of origin

stanfordnlp · Apr 28, 2015 · 3cc3604 · 3cc3604
1 parent 2352624
commit 3cc3604
Show file tree

Hide file tree

Showing 4 changed files with 33 additions and 4 deletions.
diff --git a/src/edu/stanford/nlp/naturalli/ClauseSplitterSearchProblem.java b/src/edu/stanford/nlp/naturalli/ClauseSplitterSearchProblem.java
@@ -64,6 +64,10 @@ public class ClauseSplitterSearchProblem {
       add("clone_nsubj");
       add("simple");
     }});
+    put("csubj", new ArrayList<String>() {{
+      add("clone_dobj");
+      add("simple");
+    }});
   }});
 
   /**

diff --git a/src/edu/stanford/nlp/naturalli/OpenIE.java b/src/edu/stanford/nlp/naturalli/OpenIE.java
@@ -65,7 +65,7 @@ private static enum OutputFormat { REVERB, OLLIE, DEFAULT }
   private boolean noModel = false;
 
   @Execution.Option(name="splitter.threshold", gloss="The minimum threshold for accepting a clause.")
-  private double splitterThreshold = 0.5;
+  private double splitterThreshold = 0.1;
 
   @Execution.Option(name="max_entailments_per_clause", gloss="The maximum number of entailments allowed per sentence of input.")
   private int entailmentsPerSentence = 100;

diff --git a/src/edu/stanford/nlp/naturalli/RelationTripleSegmenter.java b/src/edu/stanford/nlp/naturalli/RelationTripleSegmenter.java
@@ -306,12 +306,14 @@ public List<RelationTriple> extract(SemanticGraph parse, List<CoreLabel> tokens)
   /** A set of valid arcs denoting a subject entity we are interested in */
   private final Set<String> VALID_SUBJECT_ARCS = Collections.unmodifiableSet(new HashSet<String>(){{
     add("amod"); add("compound"); add("aux"); add("nummod"); add("nmod:poss"); add("nmod:tmod"); add("expl");
+    add("nsubj");
   }});
 
   /** A set of valid arcs denoting an object entity we are interested in */
   private final Set<String> VALID_OBJECT_ARCS = Collections.unmodifiableSet(new HashSet<String>(){{
     add("amod"); add("compound"); add("aux"); add("nummod"); add("nmod"); add("nsubj"); add("nmod:*"); add("nmod:poss");
-    add("nmod:tmod"); add("conj:and"); add("advmod"); add("acl"); add("advcl");
+    add("nmod:tmod"); add("conj:and"); add("advmod"); add("acl");
+    // add("advcl"); // Born in Hawaii, Obama is a US citizen; citizen -advcl-> Born.
   }});
 
   /** A set of valid arcs denoting an entity we are interested in */
@@ -561,8 +563,21 @@ public Optional<RelationTriple> segment(SemanticGraph parse, Optional<Double> co
           objNoopArc = Optional.ofNullable(m.getRelnString("objIgnored"));
         }
 
+        // Find the subject
+        // By default, this is just the subject node; but, occasionally we want to follow a
+        // csubj clause to find the real subject.
+        IndexedWord subject = m.getNode("subject");
+        /*
+        if (parse.outDegree(subject) == 1) {
+          SemanticGraphEdge edge =  parse.outgoingEdgeIterator(subject).next();
+          if (edge.getRelation().toString().contains("subj")) {
+            subject = edge.getDependent();
+          }
+        }
+        */
+
         // Subject+Object
-        Optional<List<CoreLabel>> subjectSpan = getValidSubjectChunk(parse, m.getNode("subject"), subjNoopArc);
+        Optional<List<CoreLabel>> subjectSpan = getValidSubjectChunk(parse, subject, subjNoopArc);
         Optional<List<CoreLabel>> objectSpan = getValidObjectChunk(parse, object, objNoopArc);
         // Create relation
         if (subjectSpan.isPresent() && objectSpan.isPresent() &&

diff --git a/test/src/edu/stanford/nlp/naturalli/RelationTripleSegmenterTest.java b/test/src/edu/stanford/nlp/naturalli/RelationTripleSegmenterTest.java
@@ -4,7 +4,6 @@
 import edu.stanford.nlp.international.Language;
 import edu.stanford.nlp.ling.CoreLabel;
 import edu.stanford.nlp.ling.IndexedWord;
-import edu.stanford.nlp.naturalli.RelationTripleSegmenter;
 import edu.stanford.nlp.semgraph.SemanticGraph;
 import edu.stanford.nlp.trees.GrammaticalRelation;
 import junit.framework.TestCase;
@@ -221,6 +220,17 @@ public void testCatsAreCute() {
     assertEquals("1.0\tcats\tare\tcute", extraction.get().toString());
   }
 
+  public void testPropagateCSubj() {
+    Optional<RelationTriple> extraction = mkExtraction(
+        "1\ttruffles\t2\tnsubj\n" +
+        "2\tpicked\t4\tcsubj\n" +
+        "3\tare\t4\tcop\n" +
+        "4\ttasty\t0\troot\n"
+    );
+    assertTrue("No extraction for sentence!", extraction.isPresent());
+    assertEquals("1.0\ttruffles picked\tare\ttasty", extraction.get().toString());
+  }
+
   public void testHeWasInaugurated() {
     Optional<RelationTriple> extraction = mkExtraction(
             "1\the\t3\tnsubjpass\n" +