Seminar: Graduate Seminar
Metaphor Detection in Medieval Hebrew
Date:
July,29,2025
Start Time:
13:00 - 14:00
Location:
1061, Meyer Building
Add to:
Lecturer:
Adir Cohen
Research Areas:
| Metaphor detection is a classification task in natural language processing (NLP) in which one trains a model that has to identify metaphors in a text. The task has relatively many challenges because the model has to learn a deep understanding of vocabulary and grammar in order to grasp the idea of metaphors, which even makes it difficult for a native speaker sometimes to understand metaphors. In this work, we will discuss the metaphor detection task in Medieval Hebrew poems and writings. The data we use contains poems and text written in the 5th to 8th century CE and were find in a synagogue in the 19th century. Our main goal is to label the metaphors in the writings. We will focus on two main approaches to predict the existence of metaphoric language: word-level prediction and sentence-level prediction. Working with a Medieval Hebrew dataset poses challenges. The limited dataset makes training a large language model impossible, as the latest models require datasets containing billions of sentences. The language itself is very different from the Modern Hebrew spoken today, which makes the expert manual tagging work itself ambiguous. In this study, we aim to address the various challenges associated with metaphor detection in general and with our specific dataset in particular. We will analyze the model architecture, loss function, and application of word embeddings. We will also describe the experiments conducted and their corresponding results. We will start with a simple baseline model and improve it section by section. We will see the improvement in performance for each modification we apply. Lastly, we will discuss the work done and possible future directions.
M.Sc. student under the supervision of Prof. Benny Kimelfeld.
|