中文导读

我们常听说地球历史上发生过五次大灭绝，而且每一次都像是突然降临的灾难。不过，南京大学古生物学家沈树忠领导的团队提出了不同的看法。他们训练了一个名为 OneStrata 的人工智能模型，用约280万条化石记录重建了约6亿年的海洋生物多样性时间线。结果显示，只有奥陶纪末和二叠纪末两次仍像全球性的急剧崩溃，另外三次（包括终结恐龙时代的白垩纪末灭绝）看起来更弱或在时间上更分散。这项成果于10月12日在美国地质学会会议上发布，《自然》于10月20日报道，但尚未正式发表；其他专家认为它可能是最详细的全球记录，却未必会彻底改写结论，还需与地区记录对照。文章按“常见说法—新发现—老问题—研究方法—结果—不同意见”展开。阅读时可留意 may、suggests、appear 等词如何表达结论的不确定性，以及 however、but 如何引出方法上的困难和不同看法。

正文

Most students learn that life on Earth has suffered five great mass extinctions, known as the “Big Five”. In each one, a large share of species seems to have disappeared in a short time. A new computer model now suggests that this familiar story may be too simple. According to its results, three of the five may have been slower declines rather than sudden disasters.

本段译文

大多数学生都学过，地球上的生命曾经历五次大规模物种灭绝，被称为“五大灭绝”。在每一次灭绝中，大量物种似乎都在短时间内消失了。一个新的计算机模型现在表明，这个人们熟悉的说法可能过于简单。根据它的结果，其中三次可能是较缓慢的衰退，而不是突然发生的灾难。

The model, called OneStrata, was developed by a team led by Shuzhong Shen, a fossil expert at Nanjing University in China. He presented the work at a meeting of the Geological Society of America in Denver on 12 October, and the journal Nature reported on it on 20 October. The study has not yet been published in a journal.

本段译文

这个名为OneStrata的模型由南京大学化石专家沈树忠领导的团队开发。他于10月12日在美国丹佛举行的美国地质学会会议上介绍了这项工作，《自然》杂志于10月20日对此进行了报道。这项研究尚未在期刊上正式发表。

The problem the team wanted to solve is an old one. Fossils are found in layers of rock, and deeper layers are usually older. But each site holds only part of the story, and it is very hard to join records from thousands of places into one timeline. To make the job easier, scientists often group fossils into periods of millions of years. This gives a blurry picture, and quick changes can be hidden.

本段译文

研究团队想解决的是一个老问题。化石存在于岩层中，较深的岩层通常更古老。但每个地点只保存了部分历史，要把来自成千上万个地方的记录拼接成一条时间线非常困难。为了简化工作，科学家常把化石归入以数百万年为单位的时期。这样得到的图景比较模糊，快速发生的变化可能被掩盖。

Shen’s team first tried a different method in 2020. Using a supercomputer, they built a detailed timeline for 11,000 sea species over about 300 million years. Adding more species, however, needed far too much computing power. So the team trained an AI model that learns patterns from fossil records, in a similar way to how chatbots learn patterns from words. It used about 2.8 million fossil records of more than 180,000 species from over 31,000 rock sections.

本段译文

沈树忠团队在2020年首先尝试了另一种方法。他们借助超级计算机，为约3亿年间的11000个海洋物种建立了详细的时间线。然而，再加入更多物种需要的计算能力太大了。于是，团队训练了一个人工智能模型，让它从化石记录中学习规律，方式类似于聊天机器人从词语中学习规律。该模型使用了来自3.1万多个岩石剖面、18万多个物种的约280万条化石记录。

The new timeline covers about 600 million years of sea life. In it, only two events, at the end of the Ordovician and the end of the Permian, still look like sharp global collapses. The other three, including the one that ended the age of the dinosaurs, appear weaker or more spread out over time.

本段译文

新的时间线涵盖了约6亿年的海洋生命史。在这条时间线中，只有奥陶纪末和二叠纪末的两次事件仍然像是全球范围的急剧崩溃。另外三次，包括终结恐龙时代的那一次，看起来规模较弱，或在时间上分布得更为分散。

Shen called the model “a real revolution“ for the field. Other experts are more careful. They say it seems to be the most detailed global record so far, but scientists already have sharp local records for many extinctions. “I don’t know if it would be Earth shattering,” said Matthew Clapham of the University of California, Santa Cruz. Researchers will now need to compare the model’s results with that local evidence.

本段译文

沈树忠称这一模型对该领域来说是“一场真正的革命”。其他专家则更为谨慎。他们表示，它似乎是迄今最详细的全球记录，但科学家已经掌握了许多灭绝事件的高精度地区记录。“我不知道它是否会带来翻天覆地的变化，”加州大学圣克鲁兹分校的马修·克拉帕姆说。研究人员接下来需要把模型的结果与这些地区证据进行比较。

新闻来源

• Nature · Vast new fossil record challenges story of Earth’s ‘big five’ mass extinctions (20 Oct 2026)
• Geological Society of America Connects 2026 · OneStrata conference abstract

✳更多文章 →