This repository contains data and code used my Masters' Thesis on "Levels of representation in a recurrent neural model of visually grounded language learning".
Phoneme transcriptions for captions in MS COCO are in data/dataset.ipa.jsonl.gz.
Code for the experiments on representation of linguistic knowledge can be found in experiments, and relies on [Reimaginet] (https://github.com/gchrupala/reimaginet).