textblob-de 0.4.1

German language support for TextBlob.

German language support for TextBlob by Steven Loria.

This python package is being developed as a TextBlob Language Extension. See Extension Guidelines for details.

Features

• All directly accessible textblob_de classes (e.g. Sentence() or Word()) are initialized with default models for German

• Properties or methods that do not yet work for German raise a NotImplementedError

• German sentence boundary detection and tokenization (NLTKPunktTokenizer)

• Consistent use of specified tokenizer for all tools (NLTKPunktTokenizer or PatternTokenizer)

• Part-of-speech tagging (PatternTagger) with keyword include_punc=True (defaults to False)

• Parsing (PatternParser) with all pattern keywords, plus pprint=True (defaults to False)

• Noun Phrase Extraction (PatternParserNPExtractor)

• Lemmatization (PatternParserLemmatizer)

• Polarity detection (PatternAnalyzer) - Still EXPERIMENTAL, does not yet have information on subjectivity

• Full pattern.text.de API support on Python3

• Supports Python 2 and 3

• See working features overview for details

Installing/Upgrading

$ pip install -U textblob-de
$ python -m textblob.download_corpora

Or the latest development release (apparently this does not always work on Windows see issues #1744/5 for details):

$ pip install -U git+https://github.com/markuskiller/textblob-de.git@dev
$ python -m textblob.download_corpora

Note

TextBlob will be installed/upgraded automatically when running pip install. The second line (python -m textblob.download_corpora) downloads/updates nltk corpora and language models used in TextBlob.

Usage

>>> from textblob_de import TextBlobDE as TextBlob
>>> text = '''Heute ist der 3. Mai 2014 und Dr. Meier feiert seinen 43. Geburtstag.
Ich muss unbedingt daran denken, Mehl, usw. für einen Kuchen einzukaufen. Aber leider
habe ich nur noch EUR 3.50 in meiner Brieftasche.'''
>>> blob = TextBlob(text)
>>> blob.sentences
[Sentence("Heute ist der 3. Mai 2014 und Dr. Meier feiert seinen 43. Geburtstag."),
 Sentence("Ich muss unbedingt daran denken, Mehl, usw. für einen Kuchen einzukaufen."),
 Sentence("Aber leider habe ich nur noch EUR 3.50 in meiner Brieftasche.")]
>>> blob.tokens
WordList(['Heute', 'ist', 'der', '3.', 'Mai', ...]
>>> blob.tags
[('Heute', 'RB'), ('ist', 'VB'), ('der', 'DT'), ('3.', 'LS'), ('Mai', 'NN'),
('2014', 'CD'), ...]
# Default: Only noun_phrases that consist of two or more meaningful parts are displayed.
# Not perfect, but a start (relies heavily on parser accuracy)
>>> blob.noun_phrases
WordList(['Mai 2014', 'Dr. Meier', 'seinen 43. Geburtstag', 'Kuchen einzukaufen',
'meiner Brieftasche'])

>>> blob = TextBlob("Das Auto ist sehr schön.")
>>> blob.parse()
'Das/DT/B-NP/O Auto/NN/I-NP/O ist/VB/B-VP/O sehr/RB/B-ADJP/O schön/JJ/I-ADJP/O'
>>> from textblob_de import PatternParser
>>> blob = TextBlobDE("Das ist ein schönes Auto.", parser=PatternParser(pprint=True, lemmata=True))
>>> blob.parse()
      WORD   TAG    CHUNK   ROLE   ID     PNP    LEMMA

       Das   DT     -       -      -      -      das
       ist   VB     VP      -      -      -      sein
       ein   DT     NP      -      -      -      ein
   schönes   JJ     NP ^    -      -      -      schön
      Auto   NN     NP ^    -      -      -      auto
         .   .      -       -      -      -      .
>>> from textblob_de import PatternTagger
>>> blob = TextBlob(text, pos_tagger=PatternTagger(include_punc=True))
[('Das', 'DT'), ('Auto', 'NN'), ('ist', 'VB'), ('sehr', 'RB'), ('schön', 'JJ'), ('.', '.')]

>>> blob = TextBlob("Das Auto ist sehr schön.")
>>> blob.sentiment
Sentiment(polarity=1.0, subjectivity=0.0)
>>> blob = TextBlob("Das ist ein hässliches Auto.")
>>> blob.sentiment
Sentiment(polarity=-1.0, subjectivity=0.0)

Warning

WORK IN PROGRESS: The German polarity lexicon contains only uninflected forms and there are no subjectivity scores yet. As of version 0.2.3, lemmatized word forms are submitted to the PatternAnalyzer, increasing the accuracy of polarity values. New in version 0.2.7: return type of .sentiment is now adapted to the main TextBlob library (:rtype: namedtuple).

>>> blob.words.lemmatize()
WordList(['das', 'sein', 'ein', 'hässlich', 'Auto'])
>>> from textblob_de.lemmatizers import PatternParserLemmatizer
>>> _lemmatizer = PatternParserLemmatizer()
>>> _lemmatizer.lemmatize("Das ist ein hässliches Auto.")
[('das', 'DT'), ('sein', 'VB'), ('ein', 'DT'), ('hässlich', 'JJ'), ('Auto', 'NN')]

Note

Make sure that you use unicode strings on Python2 if your input contains non-ascii characters (e.g. word = u"schön").

Access to pattern API in Python3

>>> from textblob_de.packages import pattern_de as pd
>>> print(pd.attributive("neugierig", gender=pd.FEMALE, role=pd.INDIRECT, article="die"))
neugierigen

Note

Alternatively, the path to textblob_de/ext can be added to the PYTHONPATH, which allows the use of pattern.de in almost the same way as described in its Documentation. The only difference is that you will have to prepend an underscore: from _pattern.de import .... This is a precautionary measure in case the pattern library gets native Python3 support in the future.

Documentation and API Reference

• http://textblob-de.readthedocs.org/en/latest

Requirements

• Python >= 2.6 or >= 3.3

TODO

• Planned Extensions

• Additional PoS tagging options, e.g. NLTK tagging (NLTKTagger)

• Improve noun phrase extraction (e.g. based on RFTagger output)

• Improve sentiment analysis (find suitable subjectivity scores)

• Improve functionality of Sentence() and Word() objects

• Adapt more tests from the main TextBlob library (esp. for TextBlobDE() in test_blob.py)

License

MIT licensed. See the bundled LICENSE file for more details.

Thanks

Coded with Wing IDE 5.0 (free open source developer license)

Changelog

0.4.1 (03/10/2014)

• Removed dependency on nltk’s depricated PunktWordTokenizer and replaced it with TreebankWordTokenizer see nltk/nltk#746 (comment) for details

0.4.0 (17/09/2014)

• Fixed Issue #7 (restore textblob>=0.9.0 compatibility)

• Depend on nltk3. Vendorized nltk was removed in textblob>=0.9.0

• Fixed ImportError on Python2 (unicodecsv)

0.3.1 (29/08/2014)

• Improved PatternParserNPExtractor (less false positives in verb filter)

• Made sure that all keyword arguments with default None are checked with is not None

• Fixed shortcut to _pattern.de in vendorized library

• Added Makefile to facilitate development process

• Added docs and API reference

0.3.0 (14/08/2014)

• Fixed Issue #5 (text + space + period)

0.2.9 (14/08/2014)

• Fixed tokenization in PatternParser (if initialized manually, punctuation was not always separated from words)

• Improved handling of empty strings (Issue #3) and of strings containing single punctuation marks (Issue #4) in PatternTagger and PatternParser

• Added tests for empty strings and for strings containing single punctuation marks

0.2.8 (14/08/2014)

• Fixed Issue #3 (empty string)

• Fixed Issue #4 (space + punctuation)

0.2.7 (13/08/2014)

• Fixed Issue #1 lemmatization of strings containing a forward slash (/)

• Enhancement Issue #2 use the same rtype as textblob for sentiment detection.

• Fixed tokenization in PatternParserLemmatizer

0.2.6 (04/08/2014)

• Fixed MANIFEST.in for package data in sdist

0.2.5 (04/08/2014)

• sdist is non-functional as important files are missing due to a misconfiguration in MANIFEST.in - does not affect wheels

• Major internal refactoring (but no backwards-incompatible API changes) with the aim of restoring complete compatibility to original pattern>=2.6 library on Python2

• Separation of textblob and pattern code

• On Python2 the vendorized version of pattern.text.de is only used, if original is not installed (same as nltk)

• Made pattern.de.pprint function and all parser keywords accessible to customise parser output

• Access to complete pattern.text.de API on Python2 and Python3 from textblob_de.packages import pattern_de as pd

• tox passed on all major platforms (Win/Linux/OSX)

0.2.3 (26/07/2014)

• Lemmatizer: PatternParserLemmatizer() extracts lemmata from Parser output

• Improved polarity analysis through look-up of lemmatised word forms

0.2.2 (22/07/2014)

• Option: Include punctuation in tags/pos_tags properties (b = TextBlobDE(text, tagger=PatternTagger(include_punc=True)))

• Added BlobberDE() class initialized with German models

• TextBlobDE(), Sentence(), WordList() and Word() classes are now all initialized with German models

• Restored complete API compatibility with textblob.tokenizers module of the main TextBlob library

0.2.1 (20/07/2014)

• Noun Phrase Extraction: PatternParserNPExtractor() extracts NPs from Parser output

• Refactored the way TextBlobDE() passes on arguments and keyword arguments to individual tools

• Backwards-incompatible: Deprecate parser_show_lemmata=True keyword in TextBlob(). Use parser=PatternParser(lemmata=True) instead.

0.2.0 (18/07/2014)

• vastly improved tokenization (NLTKPunktTokenizer and PatternTokenizer with tests)

• consistent use of specified tokenizer for all tools

• TextBlobDE with initialized default models for German

• Parsing (PatternParser) plus test_parsers.py

• EXPERIMENTAL implementation of Polarity detection (PatternAnalyzer)

• first attempt at extracting German Polarity clues into de-sentiment.xml

• tox tests passing for py26, py27, py33 and py34

0.1.3 (09/07/2014)

• First release on PyPI

0.1.0 - 0.1.2 (09/07/2014)

• First release on github

• A number of experimental releases for testing purposes

• Adapted version badges, tests & travis-ci config

• Code adapted from sample extension textblob-fr

• Language specific linguistic resources copied from pattern-de