cfgen
Uses a combination of Markov chains and context-free-grammars to generate random sentences with features of both language models.
Created by William Gilpin, 2014-2016
Requirements and Installation
- Python 2
- pyStatparser
- NLTK
- language-check (optional)
You can install the code and basic dependencies by running these commands
$ git clone https://github.com/williamgilpin/cfgen
$ conda install nltk
$ pip install git+git://github.com/bendavis78/pyStatParser
For scoring grammar or automatically correcting the resulting sentences, install the Python Package language-check.
$ pip install 3to2
$ pip install language-check
Basic Usage
Point the tool to your corpus and set up the language model. Here we will use 2-grams of Mary Shelley's Frankenstein.
mycorp = clean_corpus('cfgen/full_books/frankenstein.txt')
tagged_corpus = tag_corpus(mycorp)
termrules_mycorp = make_terminal_rules(tagged_corpus)
my_kgram = make_kgram(mycorp, k=2)
Now generate an example sentence
example_sentence = make_sentence(mycorp, termrules_mycorp, my_kgram)
corrected_example_sentence = clean_output_text(example_sentence, use_language_tool=True)
print(example_sentence + '\n')
print(corrected_example_sentence)
A full workflow is given in the file demos.ipynb.
Sample output
From a model trained on Mary Shelley's Frankenstein:
these conversation possessed to his time
those sense all fulfillment, me of affairs a soul been of I.
The next kindness her black, light of it I wished as our perish certainty.
horrible wretch that memory, their lake I to me, which the glad unhappiness
of the paroxysm of a several engagement of Clerval in her apartment and all
mainland, and on a progress, me surprised so divine me.
From a model trained on the Book of Revelations:
Who said ever to no crown to was the listener and SECOND:
my man, and to the wheat,
and had eat in this angel.
Their pieces kill the sort the angel come up
to another translucent and weep any stone.
Her timeless will measure them to the day,
hold created with earth noises and hurled every nation.
There shown out upon the voice
It be in seventh which is to trample, I.
This tampering opened not for its time.
The land to their moment Who threw their glory to cherish that art.
The glory to the speaking, and at that white appearance, and say given
the thousand for the sake. And said show in myself. And it of no sweet victory
whose gateways enemies was loathe to the bowl
and it for them and worked out as my hast to every vision.
Their noise erase me.