Skip to content

clp-research/sempix

Repository files navigation

David Schlangen, 2019-04-07

Semantics with Pictures

This repository started out as a companion to my IWCS 2019 paper "Natural Language Semantics with Pictures: Some Language & Vision Datasets and Potential Uses for Computational Semantics" [pdf]. It documents the experiments reported there, but goes beyond the material in that paper. The repository also collects the main material for my summer 2019 Potsdam class on "computational semantics with pictures". Finally, the notebooks here rely on the output of the preprocessing of the various image and image annotation corpora discussed here, the code for which is collected in the clp-vision repository. In that sense, this repository can also be seen as a companion to that code.

What you can find here

Start reading in 01_SemPics, either with the paper or with the notebook.

You will find an overview of the image corpora that we have preprocessed, and an illustration of the preprocessing format, in 02_ImageCorpora/image_corpora.ipynb. (A technical overview without text is in all_preprocessed.ipynb.)

The additional annotation that links the image corpora with natural language expressions of all kinds, and tasks that can be defined with it, are shown in 03_Tasks. Perhaps start reading with denotations.ipynb. A draft of a paper that tries to make a bit of sense of the strategy of defining tasks and games and stuff to make progress is there as well, Language Tasks and Language Games.

Some technical notes

If you want to execute the notebooks, you need to have set an environment variable VISCONF that points to a config file (format explained / illustrated in the clp-vision repo), and you also need to have access to the output of the preprocessing done in that repo, and the image data. And you need to have a ton of dependencies installed.

The notebooks here look best (and as intended) when the Jupyter extensions latex_env, toc2, and codefolding are installed; which can be done easily via NB extensions.

Citation

If you make use of any material in here, please cite

David Schlangen, Natural Language Semantics with Pictures: Some Language & Vision Datasets and Potential Uses for Computational Semantics, Proceedings of the International Conference on Computational Semantics (IWCS), 2019, Gothenburg, May

About

Semantics with Pictures: Image Corpora as Semantic Annotation of Natural Language Expressions

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages