Lower Bounds on the Expressivity of Recurrent Neural Language Models

Code accompanying the NAACL 2024 publication "Lower Bounds on the Expressivity of Recurrent Neural Language Models".

Caption (ChatGPT): "Here's an original illustration inspired by the theme of your NAACL 2024 paper, "Lower Bounds on the Expressivity of Recurrent Neural Language Models". It captures the essence of exploring the limits and challenges in understanding or generating language through recurrent neural networks."

Getting started with the code Clone the repository:

$ git clone [email protected]:rycolab/nondeterministic-rnns.git
$ cd nondeterministic-rnns

It may be beneficial to create a new Python virtual environment (e.g., using conda). We aim at Python 3.10 version and above.

Then you install the package in editable mode:

$ pip install -e .

We use pytest to unit test the code.