OpenGrm Pynini: Finite-state grammar development in Python

Version 2.1.4 is now available for download.

Pynini is now available on conda-forge. Linux (x86) and MacOS users who already have Conda can install Pynini and all dependencies using the following command: conda install -c conda-forge pynini

Linux (x86) users can now install precompiled manylinux wheels for Pynini directly from PyPI. Unlike the conda-forge installation method mentioned above, this does not install OpenFst command-line tools, does not install the Graphviz command-line tools Jupyter uses to generate state diagrams, and does not work on MacOS. To install a Pynini wheel use the following command: pip install pynini

When using Pynini from a Jupyter or Colab notebook, logging and error messages sent to standard error are lost. To recover them, install wurlitzer and then add the following at the top of your notebook: %load_ext wurlitzer

OpenGrm Pynini, like Thrax, compiles grammars expressed as strings, regular expressions, and context-dependent rewrite rules into weighted finite-state transducers. It uses the OpenFst library and its Python extension to create, access and manipulate compiled grammars. Pynini is embedded in a Python module, allowing users to write Thrax-like grammars using Python's flexible syntax (including imperative programming constructs not available in Thrax) and powerful toolchain, including an interactive development ("REPL") environment.

It is (loosely) named after Pāṇini (पाणिनि), the renowned Sanskrit grammarian.

• Download
• Documentation
• C++ API documentation (for advanced users):
  • Optimization
  • Path iteration
  • Specialty operators
  • String (de)compilation

For more information about Pynini, see:

• K. Gorman & R. Sproat. 2021. Finite-State Text Processing . Morgan & Claypool.
• K. Gorman & R. Sproat. 2016. How to get superior text processing in Python with Pynini. O'Reilly Ideas blog, accessed 11/16/16.
• K. Gorman. 2016. Pynini: A Python library for weighted finite-state grammar compilation. In Proceedings of the ACL Workshop on Statistical NLP and Weighted Automata, pages 75-80.

Other third-party information:

• Installing Pynini on Windows 10