M-fabrik 把滿文的識別、查詞與翻譯匯成一處工坊——爲奏摺、老檔與寫本里的每一行穆麟德轉寫，配上原形、詞義與譯文。brings Manchu recognition, lookup, and translation together in one workbench — pairing every line of Möllendorff transliteration from memorials, archives, and manuscripts with dictionary lemmas, glosses, and translations.
The Workbench
掃描頁進來，轉寫、查詞、譯文出去。三件工具各自能用，也能一鍵接力。Scanned pages go in; transliteration, lookups, and translations come out. Each tool works on its own, or chain them together in one click.
A Short Primer
滿文是清代的國書，豎寫、自左向右換行，字母源自蒙古文；學界通用穆麟德(Möllendorff)羅馬轉寫來書寫與檢索。Manchu was the Qing dynasty's national script — written vertically, with lines running left to right, its letters descended from Mongolian. Scholars conventionally use Möllendorff Latin transliteration for writing and searching it.
你在本站看到的 han i hese 這類拉丁串，就是滿文的穆麟德轉寫——一套一一對應滿文字母的羅馬化寫法。字母 ū 常寫作 v、š 寫作 x，方便鍵盤輸入與檢索。Latin strings like han i hese that you see on this site are Manchu's Möllendorff transliteration — a romanization system with a one-to-one correspondence to Manchu letters. The long vowel ū is often written as v, and š as x, for easier keyboard input and search.
滿文黏着：名詞有八個格(屬格 i、賓格 be、方位格 de…)，動詞以 -mbi 收尾並層層加綴。滿語解析會把屈折詞形還原到原形並標出每一層。Manchu is agglutinative: nouns have eight cases (genitive i, accusative be, dative-locative de…), and verbs end in -mbi with layers of suffixes stacked on. Language Query lemmatizes inflected forms and labels every layer.
下面幾個詞，拆開看就是一行奏摺的骨架——把它們貼進滿語解析即可逐詞展開。The words below, taken apart, are the skeleton of a memorial's opening line — paste them into Language Query to expand each one word by word.
Recognise
三步：上傳一頁滿文，拿到整頁轉寫，再一鍵接力查詞或翻譯。印刷、奏摺、行草都收。Three steps: upload a Manchu page, get a full-page transliteration, then chain it into lookup or translation with one click. Print, memorials, and cursive script are all covered.
Look up & Parse
滿文查中文、中文查滿文，雙向都行——從一整句，到一個詞，到術語與學術概念。Manchu-to-Chinese or Chinese-to-Manchu — both directions work, from a whole sentence down to a single word, terminology, and scholarly concepts.
正字是「牛祿章京」——這裏把章誤打成張，拼音摺疊照樣幫你找到 nirui janggin。The correct form is "牛祿章京" (nirui janggin) — here 章 was mistyped as 張, and pinyin folding still finds nirui janggin for you.
六部詞典打底，道、法這類概念詞都能查到對應滿文原形，供治史與譯經取詞。Backed by six merged dictionaries, concept words like dao and fa can all be traced to their Manchu dictionary lemmas — useful for historical research and scriptural translation.
In Practice
不止讀一頁——把一整批文書喂進來，得到可檢索、可翻譯的滿文語料。More than reading a single page — feed in a whole batch of documents and get a searchable, translatable Manchu corpus.
把一函奏摺整批喂進來，逐頁 OCR、逐詞解析、成段翻譯，產出一份滿漢對照、可全文檢索的語料——原本要逐字啃的一函文書，一趟走完。Feed in a whole case of memorials at once — page-by-page OCR, word-by-word parsing, and passage-level translation produce a full-text-searchable Manchu-Chinese parallel corpus. A case of documents that once took character-by-character reading is now done in one pass.
底氣所在：六部詞典合併 66,509 原形(滿漢大辭典 · 新滿漢 · 満漢和 · 語料 · 金啓孮人名)；識別頁級 CER 印刷 0.30% · 奏摺 1.59% · 行草 4.83%，多專家 Pareto 路由按難度擇優。What backs it: six dictionaries merged into 66,509 lemmas (Comprehensive Manchu-Chinese Dictionary · New Manchu-Chinese · Manchu-Chinese-Japanese · corpus · Jin Qicong's personal-name dictionary); recognition page-level CER is 0.30% for print · 1.59% for memorials · 4.83% for cursive script, with multi-expert Pareto routing choosing the best model for each difficulty.
Paper
本工坊背後的方法論文——多專家路由如何在低資源下兼顧印刷、奏摺、行草三種書體。The methodology behind this workshop — how multi-expert routing balances print, memorial, and cursive Manchu under low-resource constraints.