奥地利科学院与法国人工智慧实验室 Mistral 及科技服务公司 Sail Reply 合作，推出了全球首个专为古希腊语设计的高级大型语言模型「Apollo」。该模型接受了约 6 亿个历史希腊词汇的训练，资料来源涵盖古代手稿、莎草纸残卷与铭文，并将透过聊天机器人介面免费提供给学术界使用，旨在协助研究人员更快速地检索文献并填补残缺文本中的字词。

修复残损的古代莎草纸向来是一项极度耗时且需要高深专业知识的工程，学者必须自行断词、判定年代、评估历史背景并查阅大量参考文献。Apollo 将这些专业知识融入演算法中，能根据荷马史诗或多立克方言等特定风格自动推断并提供最符合统计机率的填补建议，使学者能从繁琐的释读工作中解脱，进而专注于文献的历史意义与学术诠释。

尽管 Apollo 不太可能突然挖掘出大量失传的伟大文学经典，多数残卷记录的仅是日常信件与公文，但它能为理解古代生活细节提供关键佐证。同时，为避免机率模型可能带来的错误并维护历史记录的真实性，Apollo 设计为提供多个候选词供人类学者裁决，强调人机协作以确保学术专业判断的核心地位，未来该技术亦有望扩展至拉丁语或古埃及语等领域。

The Austrian Academy of Sciences, in partnership with French AI lab Mistral and technology services firm Sail Reply, has unveiled 'Apollo,' the world's first advanced large language model tailored for Ancient Greek. Trained on approximately 600 million historical Greek words from manuscripts, papyri, and inscriptions, the model is freely accessible to academics via a chatbot interface to accelerate the identification and restoration of damaged papyrus fragments. (Key numbers: 6)

Restoring damaged ancient papyri has traditionally been a painstaking task requiring rare scholarly expertise to discern word boundaries, establish dates, evaluate sociopolitical contexts, and consult reference materials. Apollo integrates this specialized knowledge, generating statistically probable completions adapted to specific registers such as Homeric Greek or Doric dialect, thereby allowing classicists to shift their focus from mechanical decipherment to broader historical analysis.

While Apollo is unlikely to uncover lost literary masterpieces—as many unread papyri document routine letters and legal contracts—it promises to enrich everyday historical understanding and confirm academic hypotheses. To mitigate the risk of probabilistic AI errors polluting the historical record, Apollo presents scholars with curated candidate words rather than definitive answers, ensuring human expertise remains indispensable while paving the way for similar tools in languages like Latin and Egyptian.