2024 | Francesco Periti, Haim Dubossarsky, Nina Tahmasebi
This paper evaluates the capabilities of (Chat)GPT 3.5 and BERT in detecting semantic change, specifically through two diachronic extensions of the Word-in-Context (WiC) task: TempoWiC and HistoWiC. The study aims to assess whether (Chat)GPT can effectively handle historical and diachronic text, which is a unique challenge for NLP models. The authors compare the performance of (Chat)GPT and BERT in detecting both short-term and long-term semantic changes, finding that while BERT consistently outperforms (Chat)GPT, the latter shows potential in detecting long-term changes. The results highlight the limitations of (Chat)GPT in handling temporal and historical text, suggesting that BERT remains the superior choice for semantic change detection tasks. The paper also discusses the impact of different prompting strategies, temperature settings, and the use of the OpenAI API versus the web interface on the performance of (Chat)GPT. Overall, the study provides valuable insights into the current capabilities and limitations of (Chat)GPT in the context of semantic change detection.This paper evaluates the capabilities of (Chat)GPT 3.5 and BERT in detecting semantic change, specifically through two diachronic extensions of the Word-in-Context (WiC) task: TempoWiC and HistoWiC. The study aims to assess whether (Chat)GPT can effectively handle historical and diachronic text, which is a unique challenge for NLP models. The authors compare the performance of (Chat)GPT and BERT in detecting both short-term and long-term semantic changes, finding that while BERT consistently outperforms (Chat)GPT, the latter shows potential in detecting long-term changes. The results highlight the limitations of (Chat)GPT in handling temporal and historical text, suggesting that BERT remains the superior choice for semantic change detection tasks. The paper also discusses the impact of different prompting strategies, temperature settings, and the use of the OpenAI API versus the web interface on the performance of (Chat)GPT. Overall, the study provides valuable insights into the current capabilities and limitations of (Chat)GPT in the context of semantic change detection.