Migrate langdetect to langid.py with lazy loading (#1208)

garak/langproviders/base.py

@@ -8,9 +8,7 @@
 import re
 import unicodedata
 import string
-import logging
 from garak.resources.api import nltk
-from langdetect import detect, DetectorFactory, LangDetectException
 
 _intialized_words = False
 
@@ -104,20 +102,21 @@ def contains_invisible_unicode(text: str) -> bool:
 
 def is_meaning_string(text: str) -> bool:
     """Check if the input text is a meaningless sequence or invalid for translation."""
-    DetectorFactory.seed = 0
+    # Imported lazily so the langid model is only loaded when language
+    # detection is actually needed (e.g. reverse-translation judging).
+    import langid
 
-    # Detect Language: Skip if no valid language is detected
-    try:
-        lang = detect(text)
-    except LangDetectException:
-        logging.debug("langdetect failed to detect a valid language.")
+    # Length and pattern checks: Skip if it's too short or repetitive
+    if len(text) < 3 or re.match(r"(.)\1{3,}", text):  # e.g., "aaaa" (4+ repeats)
         return False
 
-    if lang == "en":
+    # Skip if there is no linguistic content to detect a language from
+    if not any(char.isalpha() for char in text):
         return False
 
-    # Length and pattern checks: Skip if it's too short or repetitive
-    if len(text) < 3 or re.match(r"(.)\1{3,}", text):  # e.g., "aaaa" or "123123"
+    # Detect Language: Skip if the text is English (no translation needed)
+    lang, _ = langid.classify(text)
+    if lang == "en":
         return False
 
     return True

pyproject.toml

@@ -149,7 +149,7 @@ dependencies = [
   "nvidia-riva-client==2.16.0",
   "google-cloud-translate>=2.0.4",
   "grpcio-tools>=1.71.0",
-  "langdetect==1.0.9",
+  "langid==1.1.6",
   "tiktoken>=0.7.0",
   "mistralai==1.5.2",
   "pillow>=10.4.0",

requirements.txt

@@ -35,7 +35,7 @@ ollama>=0.4.7
 nvidia-riva-client==2.16.0
 google-cloud-translate>=2.0.4
 grpcio-tools>=1.71.0
-langdetect==1.0.9
+langid==1.1.6
 tiktoken>=0.7.0
 mistralai==1.5.2
 pillow>=10.4.0

tests/langservice/test_langprovision.py

@@ -4,7 +4,7 @@
 import pytest
 
 from garak.langservice import _load_langprovider
-from garak.langproviders.base import split_input_text
+from garak.langproviders.base import is_meaning_string, split_input_text
 
 
 def test_split_input_text():
@@ -17,6 +17,52 @@ def test_split_input_text():
     assert split_input_text(input_text) == expected_output
 
 
+@pytest.mark.parametrize(
+    "text",
+    [
+        "Bonjour, ceci est une phrase en français.",
+        "これは日本語の文章です。",
+        "Hola mundo, esto es una frase en español.",
+    ],
+)
+def test_is_meaning_string_true_for_non_english(text):
+    # Non-English, translatable text should be flagged as meaningful
+    assert is_meaning_string(text) is True
+
+
+@pytest.mark.parametrize(
+    "text",
+    [
+        "This is a normal English sentence about cats.",  # English needs no translation
+        "ab",  # too short
+        "a",  # too short
+        "",  # empty
+        "aaaa",  # repetitive
+        "123123123",  # no alphabetic content
+        "12345",  # no alphabetic content
+        "!!! ??? ...",  # no alphabetic content
+        "$",  # no alphabetic content
+        "   ",  # whitespace only
+    ],
+)
+def test_is_meaning_string_false_for_meaningless_or_english(text):
+    assert is_meaning_string(text) is False
+
+
+def test_langproviders_base_uses_langid_lazily():
+    # Migration contract for issue #1208: langdetect is gone, and langid is
+    # imported lazily inside is_meaning_string rather than at module load.
+    import inspect
+
+    from garak.langproviders import base
+
+    module_source = inspect.getsource(base)
+    assert "langdetect" not in module_source
+    module_top = module_source.split("\ndef ")[0]
+    assert "import langid" not in module_top
+    assert "import langid" in inspect.getsource(base.is_meaning_string)
+
+
 @pytest.mark.parametrize(
     "langprovider_class, target_lang, model_name",
     [