Proverb Generator

Depends on

• the python library nltk : http://www.nltk.org/
• the C++ library lttoolbox : http://wiki.apertium.org/wiki/Lttoolbox

Building

Execute the build.sh script to build the libltpy.so library, which will be used to interface our python program with lttoolbox.

It may require some adjustments to the location of your lttoolbox installation.

Using

All functions are defined in the main.py file. To use them, start a python shell in this folder and import them with from main import *.

Building the background graph

We cannot provide the background graph as it is quite heavy (the fie is around 100MB). To build it, we provide two functions :

add_file_to_background_graph(graph_file, input_file)

This function inputs the whole input_file into the graph stored in graph_file. If the graph file doesn't exist, it is created.

add_words_to_background_graph(graph_file, words)

This function add a list of words to the background graph in graph_file. The word list is expected to be in the same format as the output of nltk.word_tokenize(). If the graph file doesn't exist, it is created.

In order to build the same graph as the one we used, run :

add_file_to_background_graph("proverbs.bgraph", "proverbsList.txt")
import nltk
add_words_to_background_graph("proverbs.bgraph", nltk.corpus.brown.words())
add_words_to_background_graph("proverbs.bgraph", nltk.corpus.gutenberg.words())

Using the generator

Once the background graph is created, you can make a generator object using it :

g = ProverbGenerator("proverbs.bgraph")

This will take some time to create it : the graph is normalized during the loading.

Then, you can generate proverbs by giving the generator an input word :

g.generate("darkness")

It will output the proverb used as a gramatical basis, as well as the generated proverb.

Note : if the input word is not in the graph, or is linked to too few words, the generated proverb will likely be the same as the one used as a basis.