Investigating African-American Vernacular English in Transformer-Based Text Generation

TL;DR

This paper examines GPT-2's ability to generate African-American Vernacular English (AAVE) text, analyzing sentiment and quality differences compared to Standard American English (SAE) using a parallel tweet dataset and human evaluation.

Contribution

It introduces a dataset of parallel AAVE/SAE tweets and evaluates GPT-2's performance on AAVE, highlighting linguistic and sentiment-related challenges in text generation.

Findings

AAVE text is more often classified as negative sentiment.

GPT-2 increases positive sentiment in generated text.

Human evaluation shows differences in contextual quality between AAVE and SAE.

Abstract

The growth of social media has encouraged the written use of African American Vernacular English (AAVE), which has traditionally been used only in oral contexts. However, NLP models have historically been developed using dominant English varieties, such as Standard American English (SAE), due to text corpora availability. We investigate the performance of GPT-2 on AAVE text by creating a dataset of intent-equivalent parallel AAVE/SAE tweet pairs, thereby isolating syntactic structure and AAVE- or SAE-specific language for each pair. We evaluate each sample and its GPT-2 generated text with pretrained sentiment classifiers and find that while AAVE text results in more classifications of negative sentiment than SAE, the use of GPT-2 generally increases occurrences of positive sentiment for both. Additionally, we conduct human evaluation of AAVE and SAE text generated with GPT-2 to compare contextual rigor and overall quality.