[master] support some phrasal prepositions

example/demo.py

@@ -32,6 +32,7 @@ def main():
     demo('A woman is playing the space craft at NASA.')
     demo('A woman is playing with a space craft at NASA.')
     demo('A woman next to a piano.')
+    demo('A woman in front of a piano.')
     demo('A woman standing next to a piano.')
     demo('The woman is a pianist.')
     demo('A giraffe grazing a tree in the wildness with other wildlife.')

sng_parser/_data/phrasal-preps.txt

@@ -0,0 +1,11 @@
+in addition to
+in front of
+in reference to
+in regard to
+in spite of
+on account of
+on top of
+on side of
+on the side of
+with regard to
+

sng_parser/backends/spacy_parser.py

@@ -113,6 +113,7 @@ def parse(self, sentence):
 
         # Step 3: determine the relations.
         relations = list()
+        fake_noun_marks = set()
         for entity in doc.noun_chunks:
             # Again, the subjects and the objects are represented by their position.
             relation = None
@@ -154,6 +155,18 @@ def parse(self, sentence):
                         'relation': entity.root.head.text,
                         'lemma_relation': entity.root.head.lemma_
                     }
+                # E.g., A [woman] in front of a [piano].
+                elif (
+                        entity.root.head.head.dep_ == 'pobj' and
+                        database.is_phrasal_prep(doc[entity.root.head.head.head.i:entity.root.head.i + 1].lower_)
+                ):
+                    fake_noun_marks.add(entity.root.head.head.i)
+                    relation = {
+                        'subject': entity.root.head.head.head.head.i,
+                        'object': entity.root.i,
+                        'relation': doc[entity.root.head.head.head.i:entity.root.head.i + 1].text,
+                        'lemma_relation': doc[entity.root.head.head.head.i:entity.root.head.i].lemma_
+                    }
                 # E.g., A [piano] in the [room].
                 elif entity.root.head.head.pos_ == 'NOUN':
                     relation = {
@@ -197,13 +210,17 @@ def parse(self, sentence):
                     'lemma_relation': entity.root.head.lemma_
                 }
 
-            if relation is not None:
-                # Use a helper function to map the subj/obj represented by the position
-                # back to one of the entity nodes.
-                relation['subject'] = self.__locate_noun(entity_chunks, relation['subject'])
-                relation['object'] = self.__locate_noun(entity_chunks, relation['object'])
-                if relation['subject'] != None and relation['object'] != None:
-                    relations.append(relation)
+        # Apply the `fake_noun_marks`.
+        entities = [e for e, ec in zip(entities, entity_chunks) if ec.root.i not in fake_noun_marks]
+        entity_chunks = [ec for ec in entity_chunks if ec.root.i not in fake_noun_marks]
+
+        if relation is not None:
+            # Use a helper function to map the subj/obj represented by the position
+            # back to one of the entity nodes.
+            relation['subject'] = self.__locate_noun(entity_chunks, relation['subject'])
+            relation['object'] = self.__locate_noun(entity_chunks, relation['object'])
+            if relation['subject'] != None and relation['object'] != None:
+                relations.append(relation)
 
         return {'entities': entities, 'relations': relations}
 

sng_parser/database.py

@@ -30,6 +30,10 @@ def is_phrasal_verb(verb):
     return verb in load_list('phrasal-verbs.txt')
 
 
+def is_phrasal_prep(prep):
+    return prep in load_list('phrasal-preps.txt')
+
+
 def is_scene_noun(noun):
     head = noun.split(' ')[-1]
     s = load_list('scene-nouns.txt')