A Neural-Network-Based Approach to Token-Level Metaphor Detection

This repository contains the source code for my bachelor thesis on metaphor detection using Neural Networks. It is based on the NAACL 2018 Shared Task for Metaphor Detection but did not compete in the task.

Contents:

• A Jupyter Notebook with a Keras implementation and detailed descriptions
• Standalone Keras implementations for training and evaluation
• Utils and Helper modules for data preprocessing
• Conversion and download tools for the VUAMC (training data)

For further details on the Shared Task and the training data, visit: https://github.com/EducationalTestingService/metaphor

Hint: the original data might not be available, but you (hopefully) can find a version here: https://web.archive.org/web/2019*/http://ota.ahds.ac.uk/headers/2541.xml

Prerequisites

The following prerequisites could not be included in the repository and need to be downloaded.

Word Embeddings

Download Word Embeddings for encoding lexical items (Gensim KeyedVectors, or pymagnitude) into the source/ directory.

cd source/
curl -O http://magnitude.plasticity.ai/fasttext+subword/wiki-news-300d-1M.magnitude
curl -O http://magnitude.plasticity.ai/word2vec+subword/GoogleNews-vectors-negative300.magnitude

• https://github.com/plasticityai/magnitude
• https://code.google.com/archive/p/word2vec/

VUAMC

Download the VUAM Corpus as XML (can't be included due to its license) into the starterkits/ directory.

cd starterkits/
curl -O http://ota.ahds.ac.uk/headers/2541.xml

# Or use the Python functions provided in the utils module
python3 -i utils.py
download_vuamc_xml()

The VUAMC needs to be converted into a CSV file and placed into the source/ directory. This is done using the starterkits scripts provided by the NAACL, which are included in the repository, or a Python function.

cd starterkits/
python3 vua_xml_parser.py
python3 vua_xml_parser_test.py

# Or use the Python functions provided in the utils module
python3 -i utils.py
generate_vuamc_csv()

Setup

Install the Python 3 requirements using pip:

# Optional venv
python3 -m venv .venv
source .venv/bin/activate

# Install requirements
pip3 install -r requirements.txt

Start

The implementation can either run as a Jupyter Notebook - in this case just start the Notebook:

# Start Notebook
jupyter notebook

or, as a standalone Python script which can be executed as such:

# Train the model
python3 naacl_train.py

# Evaluation
python3 naacl_evaluate.py