Answers are grounded in this paper and cite their sources.
Evaluates MLLM performance on ancient oracle bone inscriptions, suggesting significant capability differences.
Although general Multimodal Large Language Models (MLLMs) benchmarks are prevalent, systematic evaluation for oracle bone inscriptions remains underdeveloped. Oracle bone glyph, as cultural heritage artifacts, present severe degradation, stylistic variation, and cross-tablet heterogeneity, posing challenges to evaluating multifaceted MLLM capabilities. To address this, we propose the Dual-Alignment Oracle Character Reasoning Benchmark (DAOCR-Bench), which evaluates structural perception, semantic grounding, reconstruction, and consistency judgment across four competency levels. Experiments on 18 MLLMs reveal substantial performance differences under standardized and degraded oracle bone conditions, with GPT-4o achieving the highest comprehensive score in our evaluation. This benchmark provides a standardized framework for oracle bone analysis and advances computational paleography.
No takes yet. Share an insight, caveat, or question.
Wong et al. (2026) studied this question.
Synapse has enriched 8 closely related papers on similar clinical questions. Consider them for comparative context: