/nlp-project-group-4

Primary LanguageJupyter Notebook

Archived as the course has concluded.

1. nlp-project-group-4

1.1. Table of Contents

1.2. General

This repository contains the code for an NLP project focussing on morphology. Specifically, plural inflection in German and Turkish. The project was made as part of a course at the University of Groningen.

We make use of a ByT5 character level model taken from Huggingface to gauge its inflectional capabilities.

The necessary requirements.txt can be found in the root folder.

1.3. The process

For both languages (separately) we do the following:

  1. Compare a pre-trained ByT5 model finetuned on language data with a ByT5 trained on language data from scratch.

    • Compare the learning curves of the models
    • Look into the types of errors

  2. Analyse the finetuned ByT5 models using feature attribution methods to see whether there are any patterns regarding the importance of input characters for the output characters.

    • For feature attribution we make use of the Inseq library.

1.4. Overview of the repository

./
├── byt5.ipynb
├── byt5_learning_curves_finetuning.ipynb
├── byt5_learning_curves_scratch.ipynb
├── byt5_model.py
├── create_plots.ipynb
├── data/
│   ├── deu.dev
│   ├── deu.gold
│   ├── deu.test
│   ├── deu_100.train
│   ├── deu_200.train
│   ├── deu_300.train
│   ├── deu_400.train
│   ├── deu_500.train
│   ├── deu_600.train
│   ├── tur.dev
│   ├── tur.gold
│   ├── tur.test
│   ├── tur_large.train
│   └── tur_small.train
├── error_analysis.ipynb
├── generated_words/
│   ├── generated_words_ger.csv
│   └── generated_words_tur.csv
├── inseq_german.ipynb
├── inseq_turkish.ipynb
├── plot_utils.py
├── README.md
└── requirements.txt

The repository directory structure consists of a root directory with a data/ directory.

Within the root we find the fundamental $CODE.py files for the larger codebase used within the $ANALYSIS.ipynb. The $ANALYSIS.ipynb files contain the base analyses for the (1) model comparison and generation (byt5_*), (2) learning curve comparison plots (3) error_analysis, and (4) Inseq analyses (inseq_*).

The data/ directory contains the datasets that are used for training (data/*.train) and generation (data/*.gold) for the analyses.

1.5. Dependencies

The package dependencies can be found in requirements.txt.