Research case
GPT for multilingual psychological text analysis (Rathje et al., PNAS 2024)
Tested GPT-3.5, 4 and 4 Turbo on sentiment, emotions, offensiveness and moral foundations across 15 datasets (about 48,000 annotated tweets and headlines) in 12 languages.
What is reported
Simple prompts matched human annotators far better (r ≈ 0.59–0.77) than English dictionary methods (r ≈ 0.20–0.30); the authors share code and a video tutorial.
Before you use it
Correlation with annotators is not validity for your construct; replicate on a hand-coded sample in your own language and domain.
Sources and review
Source checked · Not tested by this site
Checked the published abstract and bibliographic record. The reported correlations are specific to the study’s datasets and model versions.
License and reuse
No license has been recorded. Check the source’s license before reusing code, models or data.
A license note is not a license audit. Open source does not guarantee free hosting, model access or commercial reuse.