This repository contains the code used in the paper "Representation of linguistic form and function in recurrent neural networks", the version published in Computational Linguistics.
The structure is the following:
doc
: Latex sources and figures for the papermodels
: RNN models trained on MS COCO (need to be downloaded from https://doi.org/10.5281/zenodo.825407)data
: data filessrc
: code for analysis- src/omission.py Omission scores
- src/depparse.py Dependency parse using Spacy
- src/mutual.py Mutual information scores
- src/example.py Examples
- src/ridge.py Ridge regression models
- src/analysis.R R code genarating figures
- src/analysis_mi.R R code genarating figures
The models were trained using the following software: https://github.com/gchrupala/reimaginet/releases/tag/v0.4