diff --git a/dan/datasets/extract/extract.py b/dan/datasets/extract/extract.py
index 88975536ce4a725b94e9ecafe3671d0d63d393f3..62a4c620c53d998e6436bce04d0fb8a3c837ac21 100644
--- a/dan/datasets/extract/extract.py
+++ b/dan/datasets/extract/extract.py
@@ -365,6 +365,7 @@ class ArkindexExtractor:
                 self.mapping.encode[token]
             ) if token in self.mapping.encode else self.language_tokens.append(token)
         self.language_tokens.append(self.mapping.ctc.encoded)
+        self.language_tokens.append(self.unknown_token)
 
         # Build LM corpus
         train_corpus = [
diff --git a/tests/test_extract.py b/tests/test_extract.py
index 120cd78838c20c756769fc8bf782289612749967..ca7b0e56daf22b7ab164293fae48a3c8f7b7023a 100644
--- a/tests/test_extract.py
+++ b/tests/test_extract.py
@@ -654,6 +654,7 @@ def test_extract(
         "▁" if t.isspace() else t for t in sorted(list(expected_charset))
     ]
     expected_language_tokens.append("◌")
+    expected_language_tokens.append("⁇")
     assert (output / "language_model" / "tokens.txt").read_text() == "\n".join(
         expected_language_tokens
     )