From e5f7aa141201211183e0b2f2eb5d25d4ccb194ba Mon Sep 17 00:00:00 2001
From: Osma Suominen <osma.suominen@helsinki.fi>
Date: Fri, 26 Aug 2022 16:39:18 +0300
Subject: [PATCH] Use Lingua instead of pycld3 for language detection. Fixes
 #593

---
 annif/transform/__init__.py        |  2 +-
 annif/transform/langfilter.py      | 14 ++++++++++----
 pyproject.toml                     |  4 ++--
 tests/test_transform_langfilter.py |  1 -
 4 files changed, 13 insertions(+), 8 deletions(-)

diff --git a/annif/transform/__init__.py b/annif/transform/__init__.py
index 61b9a6c59..995fce172 100644
--- a/annif/transform/__init__.py
+++ b/annif/transform/__init__.py
@@ -48,4 +48,4 @@ def get_transform(transform_specs, project):
     _transforms.update({langfilter.LangFilter.name: langfilter.LangFilter})
 except ImportError:
     annif.logger.debug(
-        "pycld3 not available, not enabling filter_language transform")
+        "Lingua not available, not enabling filter_language transform")
diff --git a/annif/transform/langfilter.py b/annif/transform/langfilter.py
index f63f6f97a..68f6ef9ec 100644
--- a/annif/transform/langfilter.py
+++ b/annif/transform/langfilter.py
@@ -2,7 +2,7 @@
 different from the language of the project."""
 
 import annif
-import cld3
+import lingua
 from . import transform
 
 logger = annif.logger
@@ -16,14 +16,20 @@ def __init__(self, project, text_min_length=500, sentence_min_length=50):
         super().__init__(project)
         self.text_min_length = int(text_min_length)
         self.sentence_min_length = int(sentence_min_length)
+        self.detector = (
+            lingua.LanguageDetectorBuilder
+            .from_all_languages()
+            .with_low_accuracy_mode()
+            .build()
+        )
 
     def _detect_language(self, text):
         """Tries to detect the language of a text input. Outputs a BCP-47-style
         language code (e.g. 'en')."""
 
-        lan_info = cld3.get_language(text)
-        if lan_info is not None and lan_info.is_reliable:
-            return lan_info.language
+        lan_info = self.detector.detect_language_of(text)
+        if lan_info is not None:
+            return lan_info.iso_code_639_1.name.lower()
         else:
             return None
 
diff --git a/pyproject.toml b/pyproject.toml
index 52e1d4a5d..a93c7e2e3 100644
--- a/pyproject.toml
+++ b/pyproject.toml
@@ -56,7 +56,7 @@ tensorflow-cpu = {version = "2.9.1", optional = true}
 lmdb = {version = "1.3.0", optional = true}
 omikuji = {version = "0.5.*", optional = true}
 yake = {version = "0.4.5", optional = true}
-pycld3 = {version = "*", optional = true}
+lingua-language-detector = {version = "1.1.1", optional = true}
 spacy = {version = "3.3.*", optional = true}
 
 [tool.poetry.dev-dependencies]
@@ -79,7 +79,7 @@ voikko = ["voikko"]
 nn = ["tensorflow-cpu", "lmdb"]
 omikuji = ["omikuji"]
 yake = ["yake"]
-pycld3 = ["pycld3"]
+lingua = ["lingua-language-detector"]
 spacy = ["spacy"]
 
 [tool.poetry.scripts]
diff --git a/tests/test_transform_langfilter.py b/tests/test_transform_langfilter.py
index b3a2f2d90..3dade8c0b 100644
--- a/tests/test_transform_langfilter.py
+++ b/tests/test_transform_langfilter.py
@@ -32,7 +32,6 @@ def test_lang_filter(project):
         Kansalliskirjasto on kaikille avoin kulttuuriperintöorganisaatio, joka
         palvelee valtakunnallisesti kansalaisia, tiedeyhteisöjä ja muita
         yhteiskunnan toimijoita.
-        Abc defghij klmnopqr stuwxyz abc defghij klmnopqr stuwxyz.
         Turvaamme Suomessa julkaistun tai Suomea koskevan julkaistun
         kulttuuriperinnön saatavuuden sekä välittämme ja tuotamme
         tietosisältöjä tutkimukselle, opiskelulle, kansalaisille ja