CFG-parser to identify semantic attributes in automatic image descriptions. Specifically, this parser analyses person-labels in descriptions generated by 9 different systems.
Note that this code is still being developed. If you want to reproduce the exact numbers from the LANTERN publication, please refer to the 1.0 release.
These are the software versions that we used. Other versions remain untested, and may give different results.
- Python 3.6.6
- NLTK 3.3
- SpaCy 2.0.4 with model 'en_core_web_sm'
example.py
shows the basic usage of this resource./Full-example
shows how to use this resource to assess your own system output./Resources
contains:- The
SystemOutputs
folder with scripts that were used for the LANTERN2020 workshop. This is not very clean (the code shouldn't be inResources/SystemOutputs/...
), and some day I may refactor the code. - The
Grammar
folder, which contains the grammar and the scripts used to build the grammar. This is further divided in: - The
Grammar/History
folder, which contains the previous versions of the lexicon, with the scripts that we used to compile the current version. - The
Grammar/Nouns
folder, which contains the set of nouns used in our grammar. The grammar does not provide full coverage of all nouns, but rather aims to provide an accurate analysis for the nouns that are given. - The
Grammar/Other
folder, which contains other terms that are essential for the grammar to work.
- The
Run python example.py
to see an analysis of the sentence "There is a young man in the garden with a tall woman."
Main reference for the current version:
@inproceedings{van-miltenburg-2020-image,
title = "How Do Image Description Systems Describe People? A Targeted Assessment of System Competence in the {PEOPLE}-domain",
author = "van Miltenburg, Emiel",
booktitle = "Proceedings of the Second Workshop on Beyond Vision and LANguage: inTEgrating Real-world kNowledge (LANTERN)",
month = dec,
year = "2020",
address = "Barcelona, Spain",
publisher = "Association for Computational Linguistics",
url = "https://www.aclweb.org/anthology/2020.lantern-1.4",
pages = "30--36"
}
This project is based on the lexica developed by Van Miltenburg et al. (2018):
@InProceedings{W18-6550,
author = "van Miltenburg, Emiel
and Elliott, Desmond
and Vossen, Piek",
title = "Talking about other people: an endless range of possibilities",
booktitle = "Proceedings of the 11th International Conference on Natural Language Generation",
year = "2018",
publisher = "Association for Computational Linguistics",
pages = "415--420",
location = "Tilburg University, The Netherlands",
url = "http://aclweb.org/anthology/W18-6550"
}
You can find the repository for this paper here.