Remove NOCOREF entities e.g. from AnCora.

dan-zeman · dan-zeman · commit e84741a6e78a · 2023-02-17T22:06:56.000+01:00
diff --git a/udapi/block/corefud/removenocorefentities.py b/udapi/block/corefud/removenocorefentities.py
@@ -0,0 +1,21 @@
+from udapi.core.block import Block
+import udapi.core.coref
+import re
+import logging
+
+class RemoveNoCorefEntities(Block):
+    """
+    Some corpora (e.g., AnCora) include annotation of named entities that are
+    not annotated for coreference. To distinguish them, their cluster ID starts
+    with 'NOCOREF' (optionally followed by entity type, so that one cluster
+    still has just one type). We may want to remove such entities from datasets
+    that are used to train coreference resolves, to prevent the resolvers from
+    thinking that all members of a NOCOREF cluster are coreferential. That is
+    what this block does.
+    """
+
+    def process_document(self, doc):
+        entities = doc.coref_entities
+        if not entities:
+            return
+        doc.coref_entities = [e for e in entities if not re.match(r'^NOCOREF', e.eid)]
diff --git a/udapi/core/coref.py b/udapi/core/coref.py
@@ -300,7 +300,7 @@ def __init__(self, eid, etype=None):
         self.split_ante = []
 
     def __lt__(self, another):
-        """Does this CorefEntity precedes (word-order wise) `another` entity?
+        """Does this CorefEntity precede (word-order wise) `another` entity?
 
         This method defines a total ordering of all entities
         by the first mention of each entity (see `CorefMention.__lt__`).
