Merge pull request #1592 from watkinsm/do_lower_case

Consider do_lower_case in PreTrainedTokenizer

transformers/tests/tokenization_tests_commons.py

@@ -110,6 +110,36 @@ class CommonTestCases:
 
             self.assertListEqual(subwords, subwords_loaded)
 
+        def test_added_tokens_do_lower_case(self):
+            tokenizer = self.get_tokenizer(do_lower_case=True)
+
+            text = "aaaaa bbbbbb low cccccccccdddddddd l"
+            text2 = "AAAAA BBBBBB low CCCCCCCCCDDDDDDDD l"
+
+            toks0 = tokenizer.tokenize(text)  # toks before adding new_toks
+
+            new_toks = ["aaaaa bbbbbb", "cccccccccdddddddd", 'AAAAA BBBBBB', 'CCCCCCCCCDDDDDDDD']
+            added = tokenizer.add_tokens(new_toks)
+            self.assertEqual(added, 2)
+
+            toks = tokenizer.tokenize(text)
+            toks2 = tokenizer.tokenize(text2)
+
+            self.assertEqual(len(toks), len(toks2))
+            self.assertNotEqual(len(toks), len(toks0))  # toks0 should be longer
+            self.assertListEqual(toks, toks2)
+
+            tokenizer = self.get_tokenizer(do_lower_case=False)
+
+            added = tokenizer.add_tokens(new_toks)
+            self.assertEqual(added, 4)
+
+            toks = tokenizer.tokenize(text)
+            toks2 = tokenizer.tokenize(text2)
+
+            self.assertEqual(len(toks), len(toks2))  # Length should still be the same
+            self.assertNotEqual(len(toks), len(toks0))
+            self.assertNotEqual(toks[0], toks2[0])  # But at least the first tokens should differ
 
         def test_add_tokens_tokenizer(self):
             tokenizer = self.get_tokenizer()

transformers/tokenization_utils.py

@@ -513,6 +513,8 @@ class PreTrainedTokenizer(object):
         to_add_tokens = []
         for token in new_tokens:
             assert isinstance(token, str) or (six.PY2 and isinstance(token, unicode))
+            if self.init_kwargs.get('do_lower_case', False):
+                token = token.lower()
             if token != self.unk_token and \
                     self.convert_tokens_to_ids(token) == self.convert_tokens_to_ids(self.unk_token) and \
                     token not in to_add_tokens:
@@ -606,6 +608,9 @@ class PreTrainedTokenizer(object):
 
             Take care of added tokens.
         """
+        if self.init_kwargs.get('do_lower_case', False):
+            text = text.lower()
+
         def split_on_token(tok, text):
             result = []
             split_text = text.split(tok)