华中科技大学等联合团队推出了AlphaOracle人工智能系统，这是首个模拟专家工作流程的甲骨文辅助释读系统 [1]。该系统通过拓片解析、字形演变分析、语境对齐和文献佐证四个步骤，帮助专家破译未释甲骨文 [1]。目前已发现的甲骨碎片超过15万片，至少包含4500个不同字形，其中得到较可靠释读的约有1500个 [1]。系统整合了大规模甲骨拓片、摹本、传世典籍和两万余篇相关研究 [1]。
在86名专家的评估中，78.7%的参与者认为系统"有帮助"或"非常有帮助" [1]，平均可节省约64%的分析时间 [1]。在语境分析中，系统准确率为56.1%；从候选结果中选出符合上下文释读的准确率为95.2% [1]。盲评"人类竞技场"中，AlphaOracle获得45.2%的最高偏好率，DeepSeek为34.8%，Gemini 2.5 Pro为12.9%，GPT-5为7.1% [1]。该研究论文已被《The Innovation》接收 [1]。
A joint research team led by Huazhong University of Science and Technology has unveiled AlphaOracle, an artificial intelligence system designed to assist experts in interpreting undeciphered oracle bone script.[1] This marks the first system to replicate the workflow of human specialists in oracle bone text analysis.[1] The system processes undeciphered bones through four sequential steps: rubbing analysis, character evolution examination, contextual alignment, and textual corroboration.[1]
Among 86 expert evaluators, 78.7 percent found the system helpful or very helpful, with the capability to reduce analysis time by an average of 64 percent.[1] In contextual analysis tasks, the system achieved 56.1 percent accuracy in initial candidate generation, and when selecting the most contextually appropriate interpretation from candidate results, accuracy reached 95.2 percent.[1] In a blind preference evaluation comparing AlphaOracle with other major AI models, the system received the highest preference rate at 45.2 percent, followed by DeepSeek at 34.8 percent, Gemini 2.5 Pro at 12.9 percent, and GPT-5 at 7.1 percent.[1]
The research addresses a significant challenge in paleography: over 150,000 oracle bone fragments have been discovered containing at least 4,500 distinct character forms, yet only approximately 1,500 have achieved reliable interpretation.[1] AlphaOracle integrates a large-scale dataset comprising oracle bone rubbings, copies, transmitted classical texts, and over 20,000 related research papers.[1] The research has been accepted for publication in The Innovation.[1]
