Skip to content
Generates sentences based on token dependency probabilities.
Java Mathematica
Branch: master
Clone or download
Fetching latest commit…
Cannot retrieve the latest commit at this time.
Type Name Latest commit message Commit time
Failed to load latest commit information.

Sentence Generator

A sentence generator using the probabilities of universal dependency relations.

Examples of auto-generated sentences:

  • yet that has more aboriginal toiletry.
  • we has abnormal blameworthiness.
  • screechy wallpaper brook Iraq.
  • Bush also roughhouse roost.
  • feckless comic input along dolorous wrongfulness.
  • limp tactician Al recommence salutary mid crackpot.
  • would that seine ruminate these chamberlain.
  • self-discipline sorrowfulness undefended save US.

###Build The project can be built with target "run" via ant run.


I developed this during a period of obsession over universal dependencies as a means of building language models, I wanted to see how capable they are to create coherent sentences. It is a Markov chain based algorithm using exclusively the two sets conditional probabilities, one for each pair of parts-of-speech (pos) and another for the possible relations originating from a given pos, to generate sentences.

The code heavily takes advantage of the object-oriented design of Java, and encapsulates each pos and grammar relation in classes and enums for modularity and logic clarity.

You can’t perform that action at this time.