How well ChatGPT understand Malaysian English? An Evaluation on Named Entity Recognition and Relation Extraction

TL;DR

This work evaluates ChatGPT on named entity recognition and relation extraction for Malaysian English (ME) using a ME-specific dataset (MEN-Dataset) and a novel educate-predict-evaluate methodology. Across 18 prompt settings, ME NER achieves a modest average $F_1$ (~0.488) with a peak of ~0.497, while morphosyntactic adaptations (loanwords, compound blends, derived forms) hinder entity extraction but do not similarly affect relation extraction, which benefits more from context and DocRED-style labeling. The authors compare ME performance to Standard English (DocRED) and find higher NER accuracy in Standard English, whereas RE performance remains comparatively similar, suggesting ME-specific linguistic features primarily challenge NER. The study emphasizes the impact of annotation guidelines, few-shot context, and prompts on performance and highlights limitations such as co-reference handling and abbreviations. Future work includes expanding the dataset, exploring additional prompting strategies, and evaluating newer models (e.g., GPT-4, Llama 2) on ME NER and RE tasks.

Abstract

Recently, ChatGPT has attracted a lot of interest from both researchers and the general public. While the performance of ChatGPT in named entity recognition and relation extraction from Standard English texts is satisfactory, it remains to be seen if it can perform similarly for Malaysian English. Malaysian English is unique as it exhibits morphosyntactic and semantical adaptation from local contexts. In this study, we assess ChatGPT's capability in extracting entities and relations from the Malaysian English News (MEN) dataset. We propose a three-step methodology referred to as \textbf{\textit{educate-predict-evaluate}}. The performance of ChatGPT is assessed using F1-Score across 18 unique prompt settings, which were carefully engineered for a comprehensive review. From our evaluation, we found that ChatGPT does not perform well in extracting entities from Malaysian English news articles, with the highest F1-Score of 0.497. Further analysis shows that the morphosyntactic adaptation in Malaysian English caused the limitation. However, interestingly, this morphosyntactic adaptation does not impact the performance of ChatGPT for relation extraction.

Figures (7)

• Figure 1: Detailed steps in the proposed educate-predict-evaluate methodology
• Figure 2: F1-Scores based on entities extracted by ChatGPT for Malaysian English news articles.
• Figure 3: Performance of ChatGPT in classifying relations based on relation labels adapted from DocRED and ACE05
• Figure 4: Prompt template used to provide entity annotation guideline as separate chunks
• Figure 5: A few examples of manually annotated entities along with explanations for why they have been annotated.