text-hr 0.13

TAGS

Croatian language, python, natural language processing (NLP), Part-of-speech (POS) tagging, stopwords, inverse inflection, morphological lexicon

OZNAKE

Hrvatski jezik, Python biblioteka, morfologija, infleksija, obrnuta infleksija, prepoznavanje vrsta riječi, računalna obrada govornog jezika, zaustavne riječi, morfološki leksikon

Inflection system

Usage example - start python shell:

>>> from text_hr import Verb
>>> v = Verb("platiti")
>>> for k in sorted(v.forms.keys()):
...     print k, v.forms[k]
...
AOR/P/1 [u'platismo']
AOR/P/2 [u'platiste']
AOR/P/3 [u'plati\u0161e']
AOR/S/1 [u'platih']
AOR/S/2 [u'plati']
AOR/S/3 [u'plati']
IMP/P/1 [u'platasmo', u'pla\u0107asmo', u'platijasmo']
IMP/P/2 [u'plataste', u'pla\u0107aste', u'platijaste']
IMP/P/3 [u'platahu', u'pla\u0107ahu', u'platijahu']
...
VA_PA//P_O+S+V+N [u'pla\u0107eno']
X_INF// [u'platiti']
X_VAD_PAS// [u'plativ\u0161i']
X_VAD_PRE// [u'plate\u0107i']
X_VAD_PRE// [u'plate\u0107i']

Detection of word types (POS tagging)

TODO: to be done - check test_detect.txt for samples, and detect.py for the logic:

First example in test_detect.txt:

>>> from text_hr.detect import WordTypeRecognizerExample
>>> def test_it(word_list, wt_filter=None, level=2):
...     wdh = WordTypeRecognizerExample(word_list, silent=True)
...     if not wt_filter is None:
...         wdh.detect(wt_filter=wt_filter, level=level)  # e.g. wt_filter=["N"]
...     else:
...         wdh.detect(level=level)  # all word types
...     lines_file = LinesFile()
...     wdh.dump_result(lines_file) # doctest: +NORMALIZE_WHITESPACE +ELLIPSIS
...     print "\n".join(lines_file.lines)
...     return wdh

>>> class LinesFile(object):
...     def __init__(self):
...         self.lines = []
...     def write(self, s):
...         self.lines.append(repr(s.rstrip()))

>>> word_list = [
...   "Broj    84"
... , "broji   34"
... , "Brojila  28"
... , "broje   23"
... , "brojeći 22"
... , "brojim   7"
... , "brojimo  5"
... , "brojiš   4"
... , "brojahu  2"
... , "brojaše  1"
... , "brojite  1"
... , "-brijestovu 1"
... , "brijestovi 1"   #the only one checked with endswith, but all other will be checked with get_freq
... , "-brijestove 1"
... , "-brijestova 1"
... ]

Lowest quality, but fastest
>>> wdh = test_it(word_list, level=4) # doctest: +ELLIPSIS
" 10/  183 -> brojati              (u'V-XX_-_JATI-je\\u0107i-0') 84/broj,34/broji,23/broje,22/broje\xe6i,7/brojim,5/brojimo,4/broji\x9a,2/brojahu,1/brojite,1/broja\x9ae"

List of stopwords

Is located in std_words.txt, and you can read it directly from here

http://bitbucket.org/trebor74hr/text-hr/src/tip/text_hr/std_words.txt

The list can be updated like this:

>>> import text_hr
>>> text_hr.dump_all_std_words()
Totaly 2904 word forms dumped to r:\hg-clones\python\text-hr\text_hr\std_words.txt in codepage utf8

Iteration over all words goes like this:

from text_hr import get_all_std_words

for word_base, l_key, cnt, _suff_id, wform_key, wform in get_all_std_words():
    print word_base, l_key, cnt, _suff_id, wform_key, wform

Further

Since there is currently no good documentation, the best source of further information is by reading tests inside of modules and tests in tests directory (dev version). More information in Running tests. You can allways read a source.

REPORT BUG OR REQUEST FEATURE

If you encounter bug, the best is to report it to the bitbucket web page http://bitbucket.org/trebor74hr/text-hr.

If there will be an interest for development for other inflection rich languages, I’d be glad to decouple language specific code and create new project that will be capable to deal with multiple languages.

The best way to contact me is by mail (find in LICENCE).

TODO list is in readme.txt (dev version).

0.13

ulr1 100610:

• text_hr package reorganized (__init__.py with __all__ and imports …)

• word_types.py removed

• std_words.txt

0.12

ulr1 100608 :

• README

• enabled tests from tests.py for all

• enabled tests from directly from each modules

0.11

ulr1 100607:

• recreated repo at bitbucket

• no .suff_registry.pickle and testing_*.out put in zip

0.10

ulr1 100605:

• first installable release