Shanghai Municipal Police Files
1894–1949

English

中文

The records of the International Settlement's police force — half a century of criminal investigation, municipal administration, the city's foreign communities and its press, alongside political surveillance — have been publicly available in scanned form for years. In practice they remain very hard to use.

公共租界巡捕房近半个世纪的档案：刑事侦查、市政管理、 各国侨民与租界报刊，也包括政治监视。扫描件早已公开多年， 但实际用起来，仍然极难检索。

The originals are carbon copies struck on a typewriter, then photographed onto microfilm. The OCR accompanying them — produced by the Virtual Shanghai / bnAsie project with ABBYY FineReader, a conventional character recogniser, not a language model — copes only partly: measured across the whole collection, about 69% of the words it produces are recognisable English; it usually gets most letters of a word and misses one or two. That is enough to half-read a page and not nearly enough to search one. enguirics for enquiries, revesleu for revealed — a human guesses; a search index cannot. Errors also compound across a phrase: if roughly one word in three is mangled, a two-word phrase survives intact about half the time, a three-word name about a third. Chinese-language enclosures, which appear throughout the files, were never recognised at all.

原件是打字机复写纸经缩微胶卷翻拍。随之而来的那份 OCR——由 Virtual Shanghai / bnAsie 项目用 ABBYY FineReader 制作，是传统逐字识别、不是大模型——只能部分应付： 全库实测，它产出的词约有 69% 是可识别的英文—— 通常能认出一个词的大部分字母，错一两个。 这个水平够勉强读懂一页，却远远不够检索。 enguirics 之于 enquiries、revesleu 之于 revealed—— 人能猜出来，检索引擎不能。误差在词组上还会叠加：若每三个词错一个， 两词词组完整存活的概率约一半，三字人名只剩三分之一。 档案中大量夹带的中文原件更是从未被识别过。

English

中文

Every page is re-transcribed from the original scan at full resolution using a vision language model, producing an English text that can actually be searched. A parallel Chinese translation is generated alongside it, for readers who need to survey the collection quickly rather than read it closely.

每一页都以原生分辨率、用视觉大模型重新转录，产出真正可检索的英文文本； 并生成中文对照译文，供需要快速浏览而非精读的读者使用。

The existing machine transcription is retained beside the new one. Where the two disagree is precisely where a human should look — a free quality signal, kept rather than discarded.

原有的机器转录一并保留。两份不一致的地方，正是需要人工核对的地方—— 这是一个免费的质检信号，所以不丢掉。

Every page carries a direct link to its scan on Internet Archive. Nothing here replaces the document; it only helps you find it.

每一页都附有 Internet Archive 上原件影像的直链。本站不替代原件， 只帮你找到它。