TartuNLP at EvaLatin 2024: Emotion Polarity Detection

TL;DR

This paper describes the TartuNLP team's approach to emotion polarity detection in historical Latin texts, utilizing heuristic and GPT-4 generated labels, and achieving first place in the EvaLatin 2024 shared task.

Contribution

The paper introduces a novel combination of heuristic and GPT-4 based annotation methods with parameter-efficient fine-tuning for Latin emotion polarity detection.

Findings

LLM-generated labels improve detection accuracy

Parameter-efficient fine-tuning is effective for Latin NLP tasks

Achieved first place in the EvaLatin 2024 shared task

Abstract

This paper presents the TartuNLP team submission to EvaLatin 2024 shared task of the emotion polarity detection for historical Latin texts. Our system relies on two distinct approaches to annotating training data for supervised learning: 1) creating heuristics-based labels by adopting the polarity lexicon provided by the organizers and 2) generating labels with GPT4. We employed parameter efficient fine-tuning using the adapters framework and experimented with both monolingual and cross-lingual knowledge transfer for training language and task adapters. Our submission with the LLM-generated labels achieved the overall first place in the emotion polarity detection task. Our results show that LLM-based annotations show promising results on texts in Latin.