Repository for the EMNLP 2020 paper 'Refer, Reuse, Reduce: Generating Subsequent References in Visual and Conversational Contexts' by Ece Takmaz, Mario Giulianelli, Sandro Pezzelle, Arabella Sinclair and Raquel Fernández.
Check out https://dmg-photobook.github.io/ for more work and details on The PhotoBook Task and Dataset.
Please look at the README files of each individual subfolder in this repository for more details.
In chain-extraction, you can find the code for extracting utterance-based referring expressions from the PhotoBook dataset. We then use these expressions that are provided in dataset to develop our models in this work.
In models, we provide the model architectures, training and evaluation scripts. You can also find the links to the pretrained models there.
The code for the linguistic analysis of the utterances generated by the models can be found under the analysis folder.