Evaluating Large Language Model Capability in Vietnamese Fact-Checking Data Generation

TL;DR

The use of LLMs for automatic data generation for the Vietnamese fact-checking task, which faces significant data limitations, is explored and the quality of the generated data cannot match the data quality produced by humans.

Abstract

Large Language Models (LLMs), with gradually improving reading comprehension and reasoning capabilities, are being applied to a range of complex language tasks, including the automatic generation of language data for various purposes. However, research on applying LLMs for automatic data generation in low-resource languages like Vietnamese is still underdeveloped and lacks comprehensive evaluation. In this paper, we explore the use of LLMs for automatic data generation for the Vietnamese fact-checking task, which faces significant data limitations. Specifically, we focus on fact-checking data where claims are synthesized from multiple evidence sentences to assess the information synthesis capabilities of LLMs. We develop an automatic data construction process using simple prompt techniques on LLMs and explore several methods to improve the quality of the generated data. To evaluate the quality of the data generated by LLMs, we conduct both manual quality assessments and performance evaluations using language models. Experimental results and manual evaluations illustrate that while the quality of the generated data has significantly improved through fine-tuning techniques, LLMs still cannot match the data quality produced by humans.

Figures (2)

• Figure 1: Large language model data generation process. Begin standard prompt construction and evidence selection from Wikipedia (see Section \ref{['subsec:Evidence Selection']}). In the automatic data generation phase through LLM, we conduct three stages of generation, including using the standard prompt in Stage 1 - Uncalibrated (see Section \ref{['subsec:Prompting Experimental Result']}), Stage 2 - Calibration with modified prompt and different flow (see Section \ref{['subsubsec:Why do we need to fine-tune LLMs in the generated task?']}), finally stage 3 - alignment we are fine-tuning LLM and using the standard prompt for generation (see Section \ref{['subsubsec:Results Of Fine-Tuning Stage']}). After data generation, we evaluate LLM dataset quality through language model (see Section \ref{['sec:AUTOMATIC EVALUATION']}) and human evaluation (see Section \ref{['subsec:MANUAL EVALUATION']}).
• Figure 2: Human evaluation in three stages of data generation (%).