A very simple way to train a model to generate random sentences given a corpus of text is to use a Markov Chain. In python, we can use markovify to build such models.

pip install markovify

Assuming that the training corpus is a collection of files, we first create a Markov Chain for each file as follows:

import os
import markovify

PATH = '...'

chains = []
for filename in os.listdir(PATH):
  content = open(f'{PATH}{filename}', 'r').readlines()
  markov_chain = markovify.Text(content)
  chains.append(markov_chain)

Then we combine the different models into one larger Markov Chain as follows:

model = markovify.combine(chains)

Finally, we can start generating random text:

model.make_sentence()