TAGtical Insertion

Contains all code and documentation of the final project of the course Computational Semantics, taught at the University of Groningen.

The main purpose of TAGtical Insertion is to tag sentences using semantic tags. These semantic tags are described in detail in Abzianidze and Bos (2017). The data used in this application originates from the Parallel Meaning Bank, also known as the PMB (Abzianidze et al., 2017).

This repository also contains our related publication An ATM for Meaning? A spaCy-based Approach to the Semantic Tagging of the Parallel Meaning Bank with TAGtical Insertion. You can find the paper in the doc folder.

Getting started

Installing additional packages

As the scripts in this repository may rely on external functionality, it might be wise to install the required software beforehand. Open your favourite command line interface, create a new venv (or install the required packages system-wide if that is your preference 😀) and enter the following line.

pip install -r requirements.txt

When you have installed the required packages, you are ready to further explore the functionalities of TAGtical Insertion!

Obtaining the PMB data

Before you can train and/or test the tagger, you need to download the data from the Parallel Meaning Bank. We use the .conll files from Rik van Noord's GitHub repository, as these files are already preprocessed and standardised for easier use (Abzianidze et al., 2017).

You can easily download these files from the repository by entering the following lines of code in your Python command line interface.

>>> from tagtical_insertion.pipeline.information import DownloadManager
>>> DownloadManager.download()

The files will pop up in the data directory. Note that this directory will be created automatically if it was not present before the download.

Preparing for training

We cannot directly feed the data from the .conll files into the training script for the semantic tagger. Using the functionality in the package tagtical_insertion.preprocessing.extraction, we convert the data to the format that suits spaCy. In the code sample below, you can see how you can perform the conversion from your Python command line interface.

>>> from tagtical_insertion.preprocessing.extraction import extract_from_conll, to_spacy_format
>>> corpus = extract_from_conll("data/train.conll")
>>> training_data = to_spacy_format(corpus)

We also added functionality to store the training data in a pickle, so you can quickly re-use the data at a later moment.

>>> from tagtical_insertion.preprocessing.extraction import store
>>> store(training_data, "data/training_data.pickle")

Using pickle can be convenient when you want to use other data to train the tagger. Overall, the preprocessing steps should not take long when using the data from Rik van Noord's repository.

Training the tagger

As mentioned in the last subsection, we use spaCy for the construction of our semantic tagger (Montani et al., 2020). spaCy offers functionality to create custom NER-like taggers. We can apply the same techniques to perform semantic tagging, using the tags mentioned in Abzianidze and Bos (2017) as the labels for the tokens.

The script model_creator.py offers the functionality to create a tagger model. This tagger model essentially resembles a spaCy processing pipeline. The main difference with a default pipeline is that our pipeline only contains the NER step to perform the semantic tagging. More information on using processing pipelines can be found here.

A brand new model can be created by running model_creator.py from the module tagtical_insertion.pipeline. The script performs the extraction and preprocessing of the data from the .conll training file. The model is conveniently stored in the model directory, part of the same tagtical_insertion.pipeline module as mentioned above.

References

Lasha Abzianidze, Johan Bos (2017): Towards Universal Semantic Tagging. Proceedings of the 12th International Conference on Computational Semantics (IWCS 2017) -- Short Papers, pp 1–6, Montpellier, France.

Lasha Abzianidze, Johannes Bjerva, Kilian Evang, Hessel Haagsma, Rik van Noord, Pierre Ludmann, Duc-Duy Nguyen, Johan Bos (2017): The Parallel Meaning Bank: Towards a Multilingual Corpus of Translations Annotated with Compositional Meaning Representations. Proceedings of the 15th Conference of the European Chapter of the Association for Computational Linguistics (EACL), pp 242–247, Valencia, Spain.

Abzianidze, L., Bjerva, J., Evang, K., Haagsma, H., van Noord, R., Ludmann, P., Nguyen, D. & Bos, J. The Parallel Meaning Bank: Towards a Multilingual Corpus of Translations Annotated with Compositional Meaning Representations, EACL 2017

Ines Montani, Matthew Honnibal, Matthew Honnibal, Sofie Van Landeghem, Adriane Boyd, Henning Peters, … Avadh Patel. (2020, December 11). explosion/spaCy: v2.3.5: Bug fixes and simpler source installs (Version v2.3.5). Zenodo. http://doi.org/10.5281/zenodo.4317367