1. Removed two languages (original Lang 1 and Lang 2) because there was overlap.
After presentation of the study at AMP 2022 in October 2022, I realized that two of the 68 languages initially reported (Lang 1 and Lang 2) were not the ones randomly sampled with Praat, but ones I created separately at an earlier stage of the project and mistakenly mixed into the bunch. Since they happened to be identical with Lang 14 and Lang 8 respectively, I excluded Lang 1 and Lang 2 and recalculated the mean performance and mean difference in number of grammar changes for a total of 66 languages.
2. Re-labeled Lang 67 and Lang 68 as the new "Lang 1" and "Lang 2" respectively.
I also re-labeled Lang 67 and Lang 68, the last two languages, as the new Lang 1 and 2 in the .xlsx and .csv files in the results folder and its subdirectories. This makes the very last label in the word list "Lang 66", hence avoiding confusion about the total number of languages -- especially in the proceedings paper. (Results from original Lang 1 and original Lang 2 are still preserved, but the directories are clearly labeled "EXCLUDED".)
This update does not affect the crucial findings about Lang 55 and lang 3, nor does it change the general conclusion of this study. But I believe it is a more accurate reflection of the nature of the project.
The older repository is preserved here: https://github.com/EunsunJou/Economy_RIPGLA_Old
This repository contains code and results files from my work on economy-based amendment of Robust Interpretive Parsing. Specifically, it contains:
- A python implementation of the RIP/OT-GLA, available as the module gla.py
- An economy-based amendment to the algorithm (the RIP/ERC-GLA), also available in the gla.py module
- A set of 66 randomly sampled abstract metrical stress languages, available in languages
(These are called Lang 1, Lang 2, ... in the paper but the files and folders say "hypo01", "hypo02", ...) - Files resulting from learning trials using the RIP/GLA and its amended version, available in results
gla.py is intended to be loaded as a module in a separate python script. workbench.py is an example script that loads the module and actually does the learning.