This repo contains code and data for "An information-theoretic characterization of morphological fusion" (at EMNLP 2021).

Contact neilrathi@gmail.com with any questions!

• code contains the code for creating fusion data for a language, as well as analysis code
• result_plots contains the plots used in the paper (main figure, paradigm size vs. fusion, frequency vs. fusion)
• langdata has data for
  • fusion by part-of-speech and language
  • paradigm size by part-of-speech and language (vs. fusion)
  • form frequency by feature and language (vs. fusion)

• R. We used version 4.0.3. Analyses and plot generation require tidyr, dplyr, ggplot2, and rPref.
• Python 3.8
• GPU TensorFlow. We used version 2.2.0.