Normalize - Use language from Corpus

orangecontrib/text/language.py

@@ -41,7 +41,7 @@
     "ga": "Irish",
     "gl": "Galician",
     "got": "Gothic",
-    "grc": "Ancient greek",
+    "grc": "Ancient Greek",
     "gu": "Gujarati",
     "he": "Hebrew",
     "hi": "Hindi",

orangecontrib/text/preprocess/normalize.py

@@ -1,4 +1,5 @@
-from typing import List, Callable
+import warnings
+from typing import List, Callable, Optional
 import os
 import ufal.udpipe as udpipe
 from lemmagen3 import Lemmatizer
@@ -84,46 +85,44 @@ def __init__(self, language='en'):
         self.normalizer = stem.SnowballStemmer(ISO2LANG[language].lower()).stem
 
 
-def language_to_name(language):
-    return language.lower().replace(' ', '') + 'ud'
-
-
-def file_to_name(file):
-    return file.replace('-', '').replace('_', '')
-
-
-def file_to_language(file):
-    return file[:file.find('ud') - 1] \
-        .replace('-', ' ').replace('_', ' ').capitalize()
-
-
 class UDPipeModels:
     server_url = "https://file.biolab.si/files/udpipe/"
 
+    # some languages differ between udpipe and iso standard
+    UDPIPE2LANG = {"Norwegian Bokmaal": "Norwegian Bokmål"}
+
     def __init__(self):
         self.local_data = os.path.join(data_dir(versioned=False), 'udpipe/')
         self.serverfiles = serverfiles.ServerFiles(self.server_url)
         self.localfiles = serverfiles.LocalFiles(self.local_data,
                                                  serverfiles=self.serverfiles)
 
     def __getitem__(self, language):
-        file_name = self._find_file(language_to_name(language))
+        file_name = self._find_file(language)
         return self.localfiles.localpath_download(file_name)
 
     @property
     def model_files(self):
         try:
-            return self.serverfiles.listfiles()
+            files = self.serverfiles.listfiles()
         except ConnectionError:
-            return self.localfiles.listfiles()
+            files = self.localfiles.listfiles()
+        return self.__files_to_dict(files)
 
     def _find_file(self, language):
-        return next(filter(lambda f: file_to_name(f).startswith(language),
-                           map(lambda f: f[0], self.model_files)))
+        return self.model_files[language][1]
+
+    def __files_to_dict(self, files):
+        iso2lang = {}
+        for f in files:
+            language_name = self.__file_to_language(f[0])
+            iso = self.__lang2iso(language_name)
+            iso2lang[iso] = (language_name, f)
+        return iso2lang
 
     @property
     def supported_languages(self):
-        return list(map(lambda f: file_to_language(f[0]), self.model_files))
+        return [(name, iso) for iso, (name, _) in self.model_files.items()]
 
     @property
     def online(self):
@@ -133,6 +132,29 @@ def online(self):
         except ConnectionError:
             return False
 
+    # use _ since - is already used in iso standard
+    VARIATION_DELIMITER = "_"
+
+    # todo: improve
+    def __lang2iso(self, language):
+        if "(" in language:
+            language, model = language.split("(")
+            language = LANG2ISO[language.strip()]
+            return UDPipeModels.VARIATION_DELIMITER.join((language, model.strip(")")))
+        return LANG2ISO[language]
+
+    def __file_to_language(self, file):
+        lg = file[: file.find("ud") - 1].split("-")
+        # if filename includes "-" then variation is part of the name
+        lg, model_variation = lg if len(lg) == 2 else (lg[0], "")
+        # capitalize multi-word languages separated by _
+        lg = " ".join(map(lambda x: x.capitalize(), lg.split("_")))
+        # fix wrong spelling for Norwegian Bokmål
+        lg = self.UDPIPE2LANG.get(lg, lg)
+        if model_variation:
+            model_variation = f"({model_variation})"
+        return " ".join((lg, model_variation)).strip()
+
 
 class UDPipeStopIteration(StopIteration):
     pass
@@ -141,7 +163,7 @@ class UDPipeStopIteration(StopIteration):
 class UDPipeLemmatizer(BaseNormalizer):
     name = 'UDPipe Lemmatizer'
 
-    def __init__(self, language='English', use_tokenizer=False):
+    def __init__(self, language="en", use_tokenizer=False):
         super().__init__()
         self.__language = language
         self.__use_tokenizer = use_tokenizer

orangecontrib/text/tests/test_preprocess.py

@@ -20,12 +20,7 @@
     PreprocessorList,
     StopwordsFilter,
 )
-from orangecontrib.text.preprocess.normalize import (
-    file_to_language,
-    file_to_name,
-    language_to_name,
-    UDPipeModels,
-)
+from orangecontrib.text.preprocess.normalize import UDPipeModels
 
 
 SF_LIST = "orangecontrib.text.preprocess.normalize.serverfiles.ServerFiles.listfiles"
@@ -270,7 +265,7 @@ def test_call_word_net(self):
         self.assertEqual(len(corpus.used_preprocessor.preprocessors), 2)
 
     def test_call_UDPipe(self):
-        pp = preprocess.UDPipeLemmatizer(language="Lithuanian")
+        pp = preprocess.UDPipeLemmatizer(language="lt")
         self.assertFalse(self.corpus.has_tokens())
         corpus = pp(self.corpus)
         self.assertTrue(corpus.has_tokens())
@@ -304,7 +299,7 @@ def test_snowball_all_langs(self):
 
     def test_udpipe(self):
         """Test udpipe token lemmatization"""
-        normalizer = preprocess.UDPipeLemmatizer("Lithuanian")
+        normalizer = preprocess.UDPipeLemmatizer("lt")
         with self.corpus.unlocked():
             self.corpus.metas[0, 0] = "esu"
         corpus = normalizer(self.corpus)
@@ -313,20 +308,19 @@ def test_udpipe(self):
 
     def test_udpipe_doc(self):
         """Test udpipe lemmatization with its own tokenization"""
-        normalizer = preprocess.UDPipeLemmatizer("Lithuanian", True)
+        normalizer = preprocess.UDPipeLemmatizer("lt", True)
         with self.corpus.unlocked():
             self.corpus.metas[0, 0] = "Ant kalno dega namas"
         corpus = normalizer(self.corpus)
         self.assertListEqual(list(corpus.tokens[0]), ["ant", "kalno", "degas", "namas"])
         self.assertEqual(len(corpus.used_preprocessor.preprocessors), 1)
 
     def test_udpipe_pickle(self):
-        normalizer = preprocess.UDPipeLemmatizer("Lithuanian", True)
+        normalizer = preprocess.UDPipeLemmatizer("lt", True)
         # udpipe store model after first call - model is not picklable
         normalizer(self.corpus)
         loaded = pickle.loads(pickle.dumps(normalizer))
-        self.assertEqual(normalizer._UDPipeLemmatizer__language,
-                         loaded._UDPipeLemmatizer__language)
+        self.assertEqual(normalizer._language, loaded._language)
         self.assertEqual(normalizer._UDPipeLemmatizer__use_tokenizer,
                          loaded._UDPipeLemmatizer__use_tokenizer)
         with self.corpus.unlocked():
@@ -336,10 +330,9 @@ def test_udpipe_pickle(self):
         )
 
     def test_udpipe_deepcopy(self):
-        normalizer = preprocess.UDPipeLemmatizer("Lithuanian", True)
+        normalizer = preprocess.UDPipeLemmatizer("lt", True)
         copied = copy.deepcopy(normalizer)
-        self.assertEqual(normalizer._UDPipeLemmatizer__language,
-                         copied._UDPipeLemmatizer__language)
+        self.assertEqual(normalizer._language, copied._language)
         self.assertEqual(normalizer._UDPipeLemmatizer__use_tokenizer,
                          copied._UDPipeLemmatizer__use_tokenizer)
         with self.corpus.unlocked():
@@ -395,21 +388,27 @@ def test_cache(self):
 class UDPipeModelsTests(unittest.TestCase):
     def test_label_transform(self, _):
         """Test helper functions for label transformation"""
-        self.assertEqual(file_to_language('slovenian-sst-ud-2.0-170801.udpipe'),
-                         'Slovenian sst')
-        self.assertEqual(file_to_name('slovenian-sst-ud-2.0-170801.udpipe'),
-                         'sloveniansstud2.0170801.udpipe')
-        self.assertEqual(language_to_name('Slovenian sst'), 'sloveniansstud')
+        model = UDPipeModels()
+        self.assertEqual(
+            model.file_to_language("slovenian-sst-ud-2.0-170801.udpipe"),
+            "Slovenian (sst)"
+        )
+        self.assertEqual(model.iso_to_file("sl_sst"), "slovenian-sst-ud")
+        self.assertEqual(
+            model.file_to_language("norwegian_bokmaal-sst-ud-2.0-170801.udpipe"),
+            "Norwegian Bokmål (sst)",
+        )
+        self.assertEqual(model.iso_to_file("nb_sst"), "norwegian_bokmaal-sst-ud")
 
     @patch(SF_DOWNLOAD, download_patch)
     def test_udpipe_model(self, _):
         """Test udpipe models loading from server"""
         models = UDPipeModels()
-        self.assertIn("Lithuanian", models.supported_languages)
+        self.assertIn("lt", models.supported_languages_iso())
         self.assertEqual(7, len(models.supported_languages))
 
         local_file = os.path.join(models.local_data, "lithuanian-ud-2.0-170801.udpipe")
-        model = models["Lithuanian"]
+        model = models["lt"]
         self.assertEqual(model, local_file)
         self.assertTrue(os.path.isfile(local_file))
 
@@ -419,17 +418,26 @@ def test_udpipe_local_models(self, sf_mock):
         models = UDPipeModels()
         [models.localfiles.remove(f[0]) for f in models.localfiles.listfiles()]
         # use Uyghur, it is the smallest model, we can have it in the repository
-        _ = models["Lithuanian"]
+        _ = models["lt"]
         sf_mock.side_effect = ConnectionError()
+        self.assertIn("lt", UDPipeModels().supported_languages_iso())
         self.assertIn("Lithuanian", UDPipeModels().supported_languages)
-        self.assertEqual(1, len(UDPipeModels().supported_languages))
+        self.assertEqual(1, len(UDPipeModels().supported_languages_iso()))
 
     def test_udpipe_offline(self, sf_mock):
         """Test if UDPipe works offline"""
         self.assertTrue(UDPipeModels().online)
         sf_mock.side_effect = ConnectionError()
         self.assertFalse(UDPipeModels().online)
 
+    def test_language_to_iso(self, _):
+        self.assertEqual("en", UDPipeModels.language_to_iso("English"))
+        self.assertEqual("en_lines", UDPipeModels.language_to_iso("English (lines)"))
+
+    def test_iso_to_language(self, _):
+        self.assertEqual("English", UDPipeModels.iso_to_language("en"))
+        self.assertEqual("English (lines)", UDPipeModels.iso_to_language("en_lines"))
+
 
 class FilteringTests(unittest.TestCase):
     def setUp(self):