ohnorobo / gibberish

make up statistically plausible words for a language

A gibberish word generator using markov chains

In action;

https://word-gibberish.onrender.com/

This program learns the low-level characteristics of a language through ngrams, then uses a markov chain to invent new plausible sounding words for that language.

to get started run

./gibberish en train 3
./gibberish en make 10

Sample words generated from a dictionary and trigrams

English

napachirs
visques
blent
huffurloussne
thed
overe
haws
unracces
ants
formalting
calcuoutic

German

bezo
grauszeltem
gargangsverfrerg
webriten
thuntehrigtenkh¦ng
vor
um
porbartrungverrobomense
dritatierf
haubirchmeufwer

French

éraus
déchabe
votesiess
ruiouit
céntez
pépontionitas

Russian

отменило
изащитнеом
ноенчаемому
гулом
ра
зандую
взловалаземен
проигразем
взберсов
бенстому

Eventual features

picking the language of your choice from wikipedia to learn from

gibberish's Issues

better file reading

read and analyze a file word by word instead of reading the whole thing in as a string and splitting it.

pull data from wikipedia

given an ISO language code (de, en, ru) find X* random articles on wikipedia in that language (how to deal with language specific urls for wikipedia:random?)

get those files into lang/dirty
strip the useful text of those files into lang/clean

learn from those files instead of dictionaries to get more realistic ngrams (dictionaries overrepresent patterns from rare words)

** likely dependant on how high the n in your ngram is.

unicode support