Work

Replication code for: Does a Bulgarian Language Model Simulate Bulgarians Better? A Pre-Registered Test of LLM Survey Respondents Against the European Social Survey

Year: 2026
Type: software
Abstract: Code and tests for a pre-registered study comparing BgGPT (Gemma 2 and Gemma 3), its base Gemma models, and Claude Sonnet 5 as silicon samplers of ESS Round 11 Bulgaria (N = 2,239, 25 questions). Includes the simulation pipeline, data-collection scripts, pre-registration calibration checks, every analysis script named in the paper, and the figures. The model outputs are archived separately (https://doi.org/10.5281/zenodo.23438036). The ESS data must be downloaded from the ESS Data Portal. See README.md for step-by-step reproduction.
Language: English
Open Access status: green
OpenAlex ID: W7222799034
Cites:
Cited by:
Related to: 10