On the logistical difficulties and findings of Jopara Sentiment Analysis

TL;DR

This study explores sentiment analysis for Jopara, a code-switching language, highlighting data collection challenges and comparing neural and traditional models, with transformers performing best despite limited Guarani data.

Contribution

It provides the first corpus of Guarani-dominant tweets and evaluates neural versus traditional models for low-resource Jopara sentiment analysis.

Findings

Transformers outperform traditional models in Jopara sentiment analysis.

Traditional machine learning models perform close to neural models in low-resource settings.

Data collection for low-resource languages remains a significant challenge.

Abstract

This paper addresses the problem of sentiment analysis for Jopara, a code-switching language between Guarani and Spanish. We first collect a corpus of Guarani-dominant tweets and discuss on the difficulties of finding quality data for even relatively easy-to-annotate tasks, such as sentiment analysis. Then, we train a set of neural models, including pre-trained language models, and explore whether they perform better than traditional machine learning ones in this low-resource setup. Transformer architectures obtain the best results, despite not considering Guarani during pre-training, but traditional machine learning models perform close due to the low-resource nature of the problem.