# 改进了用于查词的Hunspell拼写法文件（日文、英文、德文、中文繁简通查）

**URL:** <https://forum.freemdict.com/t/topic/45074>\
**Category:** GoldenDict-ng\
**Tags:** hunspell\
**Created:** [2026 年9 月 14 日 18:13 UTC](https://forum.freemdict.com/t/topic/45074 "2026-09-14T18:13:15Z")\
**Posts on this page:** 6\
**Page:** 1

<div class="post-metadata">

**Author:** ![yx\_wh](https://forumcdn.freemdict.com/user_avatar/forum.freemdict.com/yx_wh/32/77200_2.png) [@yx\_wh](https://forum.freemdict.com/u/yx_wh)\
**Post date:** [2026 年9 月 14 日 18:13 UTC](https://forum.freemdict.com/t/topic/45074/1 "2026-09-14T18:13:15Z")

</div>

> **[GitHub - yx-wh/hunspell\_ja\_JP: Hunspell morphology dictionary for Japanese used...](https://github.com/yx-wh/hunspell_ja_JP)**
>
> Hunspell morphology dictionary for Japanese used in GoldenDict.

forked from [MrCorn0-0/hunspell\_ja\_JP](https://github.com/MrCorn0-0/hunspell_ja_JP)  
删掉了来自汉直输入法词库中的人名地名的混书。人名地名一般不太可能会出现混ぜ書き。  
增加了来自《三◯堂国语辞典》的语词的混书。大概有万条？  
《词悦》的作者实现了两次调用 Hunspell，于是把 ja\_JP.dic 写成这样：

```auto
引き籠もる/5A5I5T5E5OBDUYUBubKN
ひきこもる/5A5I5T5E5OBDUYUBubKN
引きこもる/5A5I5T5E5OBDUYUBubKN
引きこもる st:引き籠もる

```

和原来手动写出动词六活用形

```auto
引き籠もる/5A5I5T5E5OBDUYUBubKN
ひきこもる/5A5I5T5E5OBDUYUBubKN
引きこもっ st:引き籠もる
引きこもら st:引き籠もる
引きこもり st:引き籠もる
引きこもる st:引き籠もる
引きこもれ st:引き籠もる
引きこもろ st:引き籠もる

```

相比，可维护性更好。  
9月26日补充：  
原来Hunspell支持  
`引きこもる/5A5I5T5E5OBDUYUBubKN st:引き籠もる`  
这样写……对.dic全面修订中  
英、德文的Hunspell拼写法文件来自

> **[GoldenDict - Browse /better morphologies/1.0 at SourceForge.net](https://sourceforge.net/projects/goldendict/files/better%20morphologies/1.0/)**
>
> GoldenDict /better morphologies/1.0 files. Browse /better morphologies/1.0 files for GoldenDict

但是就我实际体验，英文有些重读闭音节转换有误（比如hoped的原形应该是hope，但是却也会显示hop），德文缺少某些可分动词的变化，故用DeepSeek Harness驱动DF4.1修正之。

* * *

AI总结的修改en\_US拼写法文件的内容：

> ## 2025 morphology cleanup (GoldenDict-ng)
> 
> The D/G/R/T affixes in this file do not double a final consonant, so a base  
> carrying them generated a wrong surface form that collided with another lemma  
> (hop/D → hoped, while “hoped st:hope” is the correct link). All such tags were  
> removed from the base; the doubled forms stay explicit (hopped st:hop).
> 
> - 212 base words lost 438 wrong D/G/R/T (and one Z) tags.
> - bar(red/ring), cub(bed/bing), gap(ped/ping), snip(ped/ping) got the missing  
> st: links; halved/halving now point at halve instead of half.
> - sky lost R/Z, ski gained R.
> - Derivational tags were added to 2001 words, only where the derived form is  
> already a headword (nation/a → national, hope/f → hopeful, …).
> - New suffixes in en\_US.aff: f -ful, l -less, o -ous, i -ish.
> - The new -ic flag is “Q”, NOT “c”: “ONLYINCOMPOUND c” already reserves lower  
> case c, which silently removed academy/acid/artist from the dictionary.
> - Added “googleable st:google” (google/B only generated “googlable”).
> 
> See REPORT.md and CHANGES.tsv for details.
> 
> ## 2025 ODE merge (GoldenDict-ng)
> 
> The Oxford Dictionary of English headword list (ODE.csv, 160 984 entries) was  
> merged into this dictionary. Derived words are not listed when a chain of at  
> most two suffixes derives them from a base that is itself an ODE word of at  
> least four letters (Hunspell’s two-affix limit); the flag goes on the base.
> 
> - 16 941 ODE words were compressed away, e.g.  
> reconstruct/V covers reconstructive, reconstructively, reconstructiveness  
> reconstruct/j covers reconstruction, reconstructional, reconstructionary,  
> reconstructionist  
> reconstruct/B/k/v cover reconstructable / reconstructible / reconstructor
> - 104 229 new headwords were added (including 3 508 hyphen parts such as  
> “headedness” needed by compounds like “hard-headedness”).
> - All 159 744 addable ODE words are accepted by hunspell; 0 regressions against  
> the previous dictionary.
> - New affixes: j -ion, k -ible, v -or, e -ist, r -ism, z -ize, x -ise,  
> s -ship, b -hood, d -dom, w -ward, y -ify, g -ate, h -ary, O non-, W anti-.
> - WORDCHARS now contains ‘-’ and ‘.’, so hyphenated entries and abbreviations  
> such as “a.m.” are matched as single words.
> - Compression only uses suffixes; prefixed ODE words (reconstruct, nonabrasive,  
> …) stay headwords so that they can be the base of their own derivatives.
> 
> See ODE\_compression.tsv for the full compressed list and REPORT.md for details.

支持Hunspell的辞典程序：

> **[Releases · xiaoyifang/goldendict-ng](https://github.com/xiaoyifang/goldendict-ng/releases)**
>
> The Next Generation GoldenDict. Contribute to xiaoyifang/goldendict-ng development by creating an account on GitHub.

> **[GoldenDict Mobile](http://goldendict.mobi/)**

> **[Releases · mumu-lhl/Ciyue](https://github.com/mumu-lhl/Ciyue/releases/)**
>
> A simple mdict dictionary with Android/Windows/Linux support - mumu-lhl/Ciyue

另外：  
Hunspell还能用来实现中文的简体繁体通查。

`zh_CN.aff` 不需要任何实质内容，指定编码为 UTF‑8 就行：

```auto
SET UTF-8
LANG zh
FLAG long
# 空壳

```

`zh_CN.dic` 的内容类似这样，第一行是词条数目，然后每行都是st:（本来是hunspell用来标记不规则动词原形的doge）：

```auto
12
簡體 st:简体
简体 st:簡體
繁體 st:繁体
繁体 st:繁體
发 st:髮
发 st:發
髮 st:发
發 st:发
復制 st:复制
複製 st:复制
复制 st:復制
复制 st:複製

```

在GoldenDict-ng的某本词典上按右键，然后选择「词典词条」，再选择「导出」。

 ![屏幕截图_20260915_093944](https://forumcdn.freemdict.com/uploads/default/original/3X/2/0/2082bf492f7ddbaa5b6b720aa2347d8c9994df63.png)  
 ![屏幕截图_20260915_094122](https://forumcdn.freemdict.com/uploads/default/original/3X/6/7/6718e3652918441f72f60d90e5e9009c67232e01.png)  
如果导出词条是传统汉字或者传统和简化写法并存的，那就可以直接用，如果是纯简化字，用OpenCC转换成传统字之后手动校对一对多的汉字。  
然后运行以下脚本：

```auto
#!/usr/bin/env python3
# -*- coding: utf-8 -*-
"""
gen_zh_dic.py

讀取一行一個詞的詞彙表，生成 Hunspell .dic，用於 Goldendict 簡繁通查。

依賴：
    pip install opencc-python-reimplemented

用法：
    python3 gen_zh_dic.py words.txt zh_CN.dic
"""

import sys
import os
from collections import OrderedDict, defaultdict
from opencc import OpenCC

# OpenCC t2s 漏掉的一簡對多繁異體字，逐字手動補齊。
CHAR_FIX = {
    "懽": "欢",
    "歓": "欢",
    "歡": "欢",
    #"讙": "欢",
}

def make_normalizer(cc):
    def normalize(word: str) -> str:
        s = cc.convert(word).strip()
        if not s:
            return s
        return "".join(CHAR_FIX.get(ch, ch) for ch in s)
    return normalize

def main():
    if len(sys.argv) != 3:
        print("用法: python3 gen_zh_dic.py 詞彙表.txt 輸出.dic")
        sys.exit(1)

    in_path = sys.argv[1]
    out_path = sys.argv[2]

    # 繁體 -> 簡體，作為規範形式
    cc = OpenCC("t2s")
    normalize = make_normalizer(cc)

    words = []
    seen = set()

    with open(in_path, "r", encoding="utf-8") as f:
        for line in f:
            w = line.strip()
            if not w or w.startswith("#"):
                continue
            if w not in seen:
                seen.add(w)
                words.append(w)

    std_of = OrderedDict() # 原詞 -> 規範形式
    variants = defaultdict(list) # 規範形式 -> [原詞, ...]

    for w in words:
        #std = cc.convert(w).strip()
        std = normalize(w)
        if not std:
            continue
        std_of[w] = std
        if w not in variants[std]:
            variants[std].append(w)

    lines = []
    emitted_std = set()

    for w in words:
        std = std_of.get(w)
        if std is None or std in emitted_std:
            continue
        emitted_std.add(std)

        others = [v for v in variants[std] if v != std]
        if not others:
            # 繁簡相同、無變體，對 Goldendict 詞幹還原無用，跳過
            continue

        for v in others:
            lines.append(f"{v} st:{std}")
        for v in others:
            lines.append(f"{std} st:{v}")

    with open(out_path, "w", encoding="utf-8", newline="\n") as f:
        f.write(str(len(lines)) + "\n")
        if lines:
            f.write("\n".join(lines) + "\n")

    # 順便生成空殼 .aff；如果不想自動生成，可刪掉下面這段
    aff_path = os.path.splitext(out_path)[0] + ".aff"
    with open(aff_path, "w", encoding="utf-8", newline="\n") as f:
        f.write("SET UTF-8\n")
        f.write("LANG zh\n")
        f.write("FLAG long\n")

    print(f"已生成: {out_path}")
    print(f"已生成: {aff_path}")

if __name__ == " __main__":
    main()

```

这比GoldenDict直接调用opencc没准还好，因为OpenCC的一个输入只能对应一个输出，而hunspell是可以一个输入对应多个输出的。挂载Hunspell之后，輸入「复制」，GoldenDict能同时查到「復制」和「複製」。

* * *

[de\_DE\_1.2.zip](https://forum.freemdict.com/uploads/short-url/99L1m1fbP46Xhh0cZY2aIuCO3kj.zip) (1.1 MB)  
[en\_US\_1.1.zip](https://forum.freemdict.com/uploads/short-url/7QRdG533MctE8SjsaqpYW3HW9Li.zip) (760.9 KB)

---

<div class="post-metadata">

**Author:** ![feiwu](https://forumcdn.freemdict.com/letter_avatar_proxy/v4/letter/f/f17d59/32.png) [@feiwu](https://forum.freemdict.com/u/feiwu)\
**Post date:** [2026 年9 月 15 日 11:50 UTC](https://forum.freemdict.com/t/topic/45074/2 "2026-09-15T11:50:55Z")

</div>

用汉语大词典构建的dic有40w条，8M大小。  
抛开体积不谈，这种做法完全抛弃了模糊匹配，正常他应该匹配「木红」开头的词，但hunspell做不到，他只能完全匹配

 ![image](https://forumcdn.freemdict.com/uploads/default/original/3X/e/3/e35d04d01a3748ec54ca312f75cf8c03725d2092.png)  
 ![image](https://forumcdn.freemdict.com/uploads/default/original/3X/e/a/ead9fb51e3f7ca43ae7bfc6b85b6d41e5747ff26.png)

---

<div class="post-metadata">

**Author:** ![yx\_wh](https://forumcdn.freemdict.com/user_avatar/forum.freemdict.com/yx_wh/32/77200_2.png) [@yx\_wh](https://forum.freemdict.com/u/yx_wh)\
**Post date:** [2026 年9 月 15 日 12:29 UTC](https://forum.freemdict.com/t/topic/45074/3 "2026-09-15T12:29:15Z")

</div>

hunspell的拼写建议是按照西文来的，所以他会觉得「红木」跟「木红」更接近啊。类似西文中正好把两个字母按反了，搜preisdent他会给你建议president。对中日文来说，他这个拼写建议确实比较鸡肋。

---

<div class="post-metadata">

**Author:** ![chigre3](https://forumcdn.freemdict.com/letter_avatar_proxy/v4/letter/c/47e85d/32.png) [@chigre3](https://forum.freemdict.com/u/chigre3)\
**Post date:** [2026 年9 月 15 日 14:14 UTC](https://forum.freemdict.com/t/topic/45074/4 "2026-09-15T14:14:29Z")

</div>

我还在想中文有没有Hunspell呢，网络上确实没有，好像只有你做了？哈哈

---

<div class="post-metadata">

**Author:** ![yx\_wh](https://forumcdn.freemdict.com/user_avatar/forum.freemdict.com/yx_wh/32/77200_2.png) [@yx\_wh](https://forum.freemdict.com/u/yx_wh)\
**Post date:** [2026 年9 月 15 日 15:50 UTC](https://forum.freemdict.com/t/topic/45074/5 "2026-09-15T15:50:42Z")

</div>

嗯，网上应该没有。我是想到似乎可行，试了一下还真的可行。不过由于中文不分词，不同辞典生成的zh\_CN.dic差距巨大，比如《大辞海》就有大量像「`奧伯斯佯謬 st:奥伯斯佯谬`」这样的条目，导致zh\_CN.dic无法像西文那样一个hunspell.dic适配绝大多数辞典，最好是根据自己实际使用的辞典条目生成。所以我就只放了AI生成的python代码没有放.dic本身。PC上最好跟GoldenDict-ng自带的OpenCC一起使用，hunspell.dic只负责一简对多繁。

---

<div class="post-metadata">

**Author:** ![yx\_wh](https://forumcdn.freemdict.com/user_avatar/forum.freemdict.com/yx_wh/32/77200_2.png) [@yx\_wh](https://forum.freemdict.com/u/yx_wh)\
**Post date:** [2026 年9 月 27 日 18:47 UTC](https://forum.freemdict.com/t/topic/45074/6 "2026-09-27T18:47:23Z")

</div>

查询supersized没查到supersize，遂打开Codebuddy蹬了一轮腾讯的免费智能体，把德文和英文的Hunspell文件修正了一番。已上传到主楼。