This Python module can be used to quickly retrieve absolute word frequency for English language, as used in Wikipedia articles as of year 2012.

Author: Paweł Foremski <[email protected]>, IITiS PAN www.iitis.pl

>>> import wikiwords
>>> wikiwords.freq("monty")
6.348454761413523e-06
>>> wikiwords.occ("python")
18972
>>> wikiwords.freq("no such word", lambda x: 1./len(x))
0.08333333333333333

$ sudo pip install wikiwords

Wikipedia files were downloaded on 23.11.2012

The corpus was filtered:

1. all words containing non-latin letters were removed (A-Z)
2. in single files (see below), words with frequency < 5e-7 were removed
3. in the final file (28 files merged), words with f < 5e-8 were removed
4. all words shorter than 2 characters were removed (see below)

The final corpus contains over 100,000 words with over 2 billion occurances.

For copyright information on data, see

http://en.wikipedia.org/wiki/Wikipedia:Copyrights

The Python source code is licensed to you under the GNU GPL v3

See "the making of/" subdirectory for details on how the data was created.

The single-letter words were removed by the gensim parser. In order to address this, relevant frequencies and number of occurances were artifically injected from the Google Web Trillion Word Corpus, described by Thorsten Brants and Alex Franz in 2006 [1]. For more information -- and for an example of possible application of wikiwords.py -- see Peter Norvig ngrams site at [2].

1. http://googleresearch.blogspot.com/2006/08/all-our-n-gram-are-belong-to-you.html
2. http://norvig.com/ngrams/