Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

GATech at AbjadGenEval Shared Task: Multilingual Embeddings for Arabic Machine-Generated Text Classification

Ahmed Khaled Khamis

TL;DR

The multilingual E5-large encoder for binary classification was fine-tuned, and several pooling strategies to pool token representations, including weighted layer pooling, multi-head attention pooling, and gated fusion failed to outperform simple mean pooling.

Abstract

We present our approach to the AbjadGenEval shared task on detecting AI-generated Arabic text. We fine-tuned the multilingual E5-large encoder for binary classification, and we explored several pooling strategies to pool token representations, including weighted layer pooling, multi-head attention pooling, and gated fusion. Interestingly, none of these outperformed simple mean pooling, which achieved an F1 of 0.75 on the test set. We believe this is because complex pooling methods introduce additional parameters that need more data to train properly, whereas mean pooling offers a stable baseline that generalizes well even with limited examples. We also observe a clear pattern in the data: human-written texts tend to be significantly longer than machine-generated ones.

GATech at AbjadGenEval Shared Task: Multilingual Embeddings for Arabic Machine-Generated Text Classification

TL;DR

The multilingual E5-large encoder for binary classification was fine-tuned, and several pooling strategies to pool token representations, including weighted layer pooling, multi-head attention pooling, and gated fusion failed to outperform simple mean pooling.

Abstract

We present our approach to the AbjadGenEval shared task on detecting AI-generated Arabic text. We fine-tuned the multilingual E5-large encoder for binary classification, and we explored several pooling strategies to pool token representations, including weighted layer pooling, multi-head attention pooling, and gated fusion. Interestingly, none of these outperformed simple mean pooling, which achieved an F1 of 0.75 on the test set. We believe this is because complex pooling methods introduce additional parameters that need more data to train properly, whereas mean pooling offers a stable baseline that generalizes well even with limited examples. We also observe a clear pattern in the data: human-written texts tend to be significantly longer than machine-generated ones.
Paper Structure (20 sections, 1 figure, 3 tables)

This paper contains 20 sections, 1 figure, 3 tables.

Table of Contents

  1. Introduction
  2. Background
  3. Task Setup
  4. Dataset
  5. Related Work
  6. System Overview
  7. Model Architecture
  8. Pooling Strategies
  9. Classification Head
  10. Loss Function
  11. Experimental Setup
  12. Data Split
  13. Hyperparameters
  14. Layer-wise Learning Rate Decay
  15. Other Implementation Details
  16. ...and 5 more sections

Figures (1)

  • Figure 1: Dataset analysis: (a) Balanced class distribution with 2,649 samples per class. (b) Word count distribution showing human texts are significantly longer.