Modernize Pred normalization functions.

Fixes #111.

CHANGELOG.md

@@ -16,6 +16,8 @@
 
 * Properly call `re.sub()` so the flags don't become the count (#108)
 * Include file size of gzipped tables in summary of `delphin mkprof` (#110)
+* `normalize_pred_string()` now strips `_rel` (#111) and lowercases
+* `is_valid_pred_string()` no longer requires `_rel` (#111)
 
 ### Deprecated
 

delphin/mrs/components.py

@@ -465,8 +465,7 @@ def short_form(self):
             >>> p.short_form()
             '_cat_n_1'
         """
-        s = self.string.strip('"').lstrip("'")
-        return re.sub(r'(.*)_rel$', r'\1', s, flags=re.U|re.I)
+        return normalize_pred_string(self.string)
 
     def is_quantifier(self):
         """
@@ -518,24 +517,23 @@ def is_valid_pred_string(predstr):
     Pred, `False` otherwise.
     """
     predstr = predstr.strip('"').lstrip("'")
-    if not predstr.endswith('_rel'):
-        return False
     # this is a stricter regex than in Pred, but doesn't check POS
-    return re.match(r'_?((?:[^_\\]|\\.)+_){1,3}rel', predstr) is not None
+    return re.match(
+        r'_([^ _\\]|\\.)+_[a-z](_([^ _\\]|\\.)+)?(_rel)?$'
+        r'|[^_]([^ \\]|\\.)+(_rel)?$',
+        predstr
+    ) is not None
 
 
 def normalize_pred_string(predstr):
     """
-    Make pred strings more consistent by removing quotes and using
-    the _rel suffix.
+    Make pred strings more consistent by removing quotes and the _rel
+    suffix, and by lowercasing them.
     """
-    tokens = []
+    tokens = [t for t in split_pred_string(predstr)[:3] if t is not None]
     if predstr.lstrip('\'"')[:1] == '_':
-        tokens.append('')
-    tokens.extend(t for t in split_pred_string(predstr) if t is not None)
-    if not tokens[-1] == 'rel':
-        tokens.append('rel')
-    return '_'.join(tokens)
+        tokens = [''] + tokens
+    return '_'.join(tokens).lower()
 
 
 class Node(

tests/mrs_components_test.py

@@ -464,13 +464,16 @@ def test_hash(self):
         assert spred('"_the_q_rel"') in s
         assert spred('_the_q_rel') in s
         assert spred('_the_q') in s
+        assert spred('_The_q_rel') in s
         assert spred('the_q_rel') not in s
         s.add(spred('_the_q_rel'))
         assert len(s) == 1
 
 
 def test_split_pred_string():
     sps = split_pred_string
+    # normalized
+    assert sps('dog_n_1') == ('dog', 'n', '1', None)
     # with rel
     assert sps('pron_rel') == ('pron', None, None, 'rel')
     # with pos
@@ -499,18 +502,23 @@ def test_is_valid_pred_string():
     assert ivps('_24/7_a_1_rel')
     assert ivps('_a+bit_q_rel')
     assert ivps('_A$_n_1_rel')
+    assert ivps('coord')
+    assert ivps('_dog_n_1')
+    assert ivps('_dog_n')
     # invalid
-    assert not ivps('coord')
-    assert not ivps('coord_relation')
+    assert not ivps('_dog_rel')
+    assert not ivps('_dog_1_rel')
     assert not ivps('_only_child_n_1_rel')
 
 
 def test_normalize_pred_string():
     nps = normalize_pred_string
-    assert nps('pron') == 'pron_rel'
-    assert nps('"udef_q_rel"') == 'udef_q_rel'
-    assert nps('\'udef_q_rel') == 'udef_q_rel'
-    assert nps('_dog_n_1_rel') == '_dog_n_1_rel'
+    assert nps('pron_rel') == 'pron'
+    assert nps('pron_rel_rel') == 'pron_rel'  # i hope nobody does this
+    assert nps('"udef_q_rel"') == 'udef_q'
+    assert nps('\'udef_q_rel') == 'udef_q'
+    assert nps('_dog_n_1_rel') == '_dog_n_1'
+    assert nps('_DELPH-IN_n_1') == '_delph-in_n_1'
 
 class TestNode():
     def test_construct(self):