@furongjun1999/dsh-memory 0.4.6 → 0.4.7

package/md_cg/lexicon/build_cedict_en_zh.py

@@ -0,0 +1,320 @@
+#!/usr/bin/env python3
+# -*- coding: utf-8 -*-
+"""build_cedict_en_zh.py · 词级 en→zh 反查表构建器（CC-CEDICT 词级，补字级反查缺口）
+
+背景（2026-09-14 裁定路径）：u2c 归因臂实证「语义路机制+doc 侧上界 hit@10=99.2%」，
+瓶颈 100% 在 query 侧 en→zh 归一覆盖（u2=23.8 vs u2c=99.2）。字级 char_atoms_clean
+反查对剩余 OOV 词次覆盖仅 ~15%（expand_en_zh analyze 实证），且字级产物与 doc 侧
+整词原子形态不齐（music→单字 vs 摘要「音乐」整词原子）。本脚本从 CC-CEDICT 词级
+词典反查英文词→中文词，产出 md_cg/lexicon/cedict_en_zh.json（独立数据文件）——
+许可 CC-BY-SA 4.0，故**不内联**进 MIT 的 en_normalizer.EN_ZH，独立文件+署名。
+
+清理规则（与 expand_en_zh.py 同纪律，机械可审计，不做语义发明）：
+  · 释义列剥括号注（(Taiwan pr. ...) 等）→ 剥 "to /a /an " 前缀 → 按非字母拆词
+  · 纯字母 fullmatch [a-z]+；min_len>=3（2 字母缩写域碰撞）
+  · STOPWORDS 排除；与手工 EN_ZH 重复 → 手工优先（build 时写入 excluded）
+  · 同一英文词反查到多个中文词 → 冲突跳过并计数（宁缺勿滥）
+
+用法：
+    python md_cg/lexicon/build_cedict_en_zh.py download   # 拉 CEDICT 原始 gz（gitignored）
+    python md_cg/lexicon/build_cedict_en_zh.py build      # 反查 → cedict_en_zh.json
+    python md_cg/lexicon/build_cedict_en_zh.py analyze    # 500 题 OOV 覆盖上界取证
+
+产物许可：本仓 cedict_en_zh.json 为 CC-CEDICT 派生数据，依 CC-BY-SA 4.0 署名：
+    CC-CEDICT (https://www.mdbg.net/chinese/dictionary, CC BY-SA 4.0)
+"""
+import collections
+import gzip
+import io
+import json
+import os
+import re
+import sys
+import time
+import urllib.request
+
+HERE = os.path.dirname(os.path.dirname(os.path.dirname(os.path.abspath(__file__))))
+for _p in (HERE, os.path.join(HERE, "md_cg")):
+    if _p not in sys.path:
+        sys.path.insert(0, _p)
+
+from md_cg.semantic.en_normalizer import EN_ZH, STOPWORDS  # noqa: E402
+
+RAW_DIR = os.path.join(HERE, "data", "external", "cedict")
+RAW_GZ = os.path.join(RAW_DIR, "cedict_1_0_ts_utf-8_mdbg.txt.gz")
+RAW_TXT = os.path.join(RAW_DIR, "cedict.txt")
+OUT_JSON = os.path.join(HERE, "md_cg", "lexicon", "cedict_en_zh.json")
+QUESTIONS = os.path.join(HERE, "data", "external", "locomo_zh",
+                         "raw_questions.jsonl")
+CEDICT_URL = "https://www.mdbg.net/chinese/export/cedict/cedict_1_0_ts_utf-8_mdbg.txt.gz"
+LINE_RE = re.compile(r"^(\S+)\s+(\S+)\s+\[[^\]]+\]\s+/(.+)/\s*$")
+
+
+def download():
+    os.makedirs(RAW_DIR, exist_ok=True)
+    req = urllib.request.Request(
+        CEDICT_URL, headers={"User-Agent": "Mozilla/5.0 (dsh-memory lexicon builder)"})
+    with urllib.request.urlopen(req, timeout=120) as resp:
+        blob = resp.read()
+    with io.open(RAW_GZ, "wb") as f:
+        f.write(blob)
+    with gzip.open(RAW_GZ, "rb") as f:
+        raw = f.read()
+    with io.open(RAW_TXT, "w", encoding="utf-8", newline="\n") as f:
+        f.write(raw.decode("utf-8", errors="replace"))
+    print("[download] %d bytes → %s（%d 行）"
+          % (len(blob), RAW_TXT, raw.count(b"\n")))
+    return 0
+
+
+def iter_entries():
+    """CEDICT 行 → (简体, [英文释义段])。注释/空行跳过。"""
+    with io.open(RAW_TXT, encoding="utf-8") as f:
+        for line in f:
+            line = line.strip()
+            if not line or line.startswith("#"):
+                continue
+            m = LINE_RE.match(line)
+            if not m:
+                continue
+            yield m.group(2), m.group(3)
+
+
+def reverse_map(min_len=3):
+    """词级反查 → (候选 {en: zh}, 冲突 {en: [zh...]}, 统计 dict)。
+
+    条目级子段反查：释义按 / 段 → 括号注剥离 → 按 ;, 切子段 → 子段清洗后
+    **恰为单个英文词**才收（"music album" 等复合短语被纯字母 fullmatch 滤除，
+    不再以复合词条身份污染反查——首轮实证 4.4% 覆盖的根因）。
+    """
+    rev = collections.defaultdict(set)
+    stat = {"entries": 0, "glosses": 0, "bracket_stripped": 0,
+            "prefix_stripped": 0, "stopword": 0, "too_short": 0,
+            "phrase_skipped": 0}
+    for zh, gloss in iter_entries():
+        stat["entries"] += 1
+        for seg in gloss.split("/"):
+            seg = re.sub(r"\([^)]*\)", "", seg)          # 剥括号注
+            if seg != gloss:
+                stat["bracket_stripped"] += 1
+            for sub in re.split(r"[;,]", seg):           # 子段
+                sub = sub.strip().lower()
+                sub = re.sub(r"^(to|a|an) ", "", sub)
+                if sub != sub.strip().lower():
+                    stat["prefix_stripped"] += 1
+                if not re.fullmatch(r"[a-z]+", sub):
+                    stat["phrase_skipped"] += 1
+                    continue
+                if sub in STOPWORDS:
+                    stat["stopword"] += 1
+                    continue
+                if len(sub) < min_len:
+                    stat["too_short"] += 1
+                    continue
+                stat["glosses"] += 1
+                rev[sub].add(zh)
+    cand, conflicts = {}, {}
+    for en, zhs in rev.items():
+        if len(zhs) == 1:
+            cand[en] = next(iter(zhs))
+        else:
+            conflicts[en] = sorted(zhs)
+    return cand, conflicts, stat
+
+
+def _corpus_text():
+    """领域语料全文（zh 正文+五槽摘要）——冲突集频次消歧用。
+
+    消歧本质=doc 侧自监督：query 词要匹配的对象就是 doc 侧原子，
+    「公园」在语料高频、「园囿」零出现——频次直接裁决，与匹配目标自洽。
+    """
+    from md_cg import bench6_common as b6
+    buf = []
+    with io.open(b6.CORPUS567, encoding="utf-8") as f:
+        for line in f:
+            if not line.strip():
+                continue
+            c = json.loads(line)
+            parts = [c.get("zh") or ""]
+            for v in (c.get("zh_fields") or {}).values():
+                if isinstance(v, str):
+                    parts.append(v)
+            buf.append("".join(parts))
+    return "\n".join(buf)
+
+
+def _atoms_texts():
+    """doc 侧同构消歧空间（2026-09-14 消歧口径升级）。
+
+    取证：tattoo 冲突集（刺字/刺花/刺青/文身/纹…）整词形在语料 count 全零
+    → 旧口径整词丢弃；而真实用词「纹身」在摘要 fm.semantic 落盘形态是
+    切分碎片「纹 身」——消歧 count 的字符串空间与匹配面形态从未对齐，
+    11257 个英文词（高频多义词重灾区）因此被整词丢弃。
+
+    返回 (摘要原子串, 正文原子串)：候选词经 semantic_atoms 切分后以
+    「原子 空格 原子」形态计数——与检索打分面（pair_hits 的 doc 侧输入）
+    完全同构。摘要优先（fm.semantic 是语义路唯一 doc 侧），正文兜底
+    （词法路匹配面）。
+    """
+    from md_cg import bench6_common as b6
+    from md_cg.semantic.canonical import semantic_atoms
+    sums, bodys = [], []
+    with io.open(b6.CORPUS567, encoding="utf-8") as f:
+        for line in f:
+            if not line.strip():
+                continue
+            c = json.loads(line)
+            s = str((c.get("zh_fields") or {}).get("summary") or "").strip()
+            if not s:
+                s = str(c.get("zh") or "").strip()
+            if s:
+                sums.append(" ".join(semantic_atoms(s)))
+            body = str(c.get("zh") or "").strip()
+            if body:
+                bodys.append(" ".join(semantic_atoms(body)))
+    return "\n".join(sums), "\n".join(bodys)
+
+
+def _disambiguate(en, zhs, texts):
+    """单英文词冲突集三级消歧：摘要碎片串 → 正文碎片串 → 碎片字级和。
+
+    返回 (胜出中文词, 级别) 或 (None, -1)（全零丢弃，宁缺勿滥——不发明词形）。
+    切分形态计数与 doc 侧匹配面同构（纹身→「纹 身」对「纹 身」count）；
+    字级兜底只解决「词条整词与语料词形微差」（文身 vs 纹身 异形词——
+    「文」「身」碎片分别可数）。候选恒限定词典集合内，不做语义发明。
+    """
+    from md_cg.semantic.canonical import semantic_atoms
+    sums, bodys = texts
+    scored = []
+    for z in sorted(zhs):
+        if not 2 <= len(z) <= 4:
+            continue
+        frag = " ".join(semantic_atoms(z))
+        n = sums.count(frag)
+        if n > 0:
+            scored.append((n, 0, len(z), z))
+    if not scored:
+        for z in sorted(zhs):
+            if not 2 <= len(z) <= 4:
+                continue
+            frag = " ".join(semantic_atoms(z))
+            n = bodys.count(frag)
+            if n > 0:
+                scored.append((n, 1, len(z), z))
+    if not scored:
+        for z in sorted(zhs):
+            if not 2 <= len(z) <= 4:
+                continue
+            atoms = semantic_atoms(z)
+            n = sum(sums.count(" ".join(atoms[i:i + 1])) for i in range(len(atoms)))
+            if n > 0:  # 字级兜底门槛试验（>=4）实测 u2 逐位一致：字级回收词
+                       # 为开放域词，locomo 题面零出现，无噪声代价，保留全覆盖
+                scored.append((n, 2, len(z), z))
+    if not scored:
+        return None, -1
+    scored.sort(key=lambda t: (-t[0], t[1], t[2], t[3]))
+    return scored[0][3], scored[0][1]
+
+
+def build(min_len=3, domain=True):
+    cand, conf, stat = reverse_map(min_len)
+    stat["disambiguated"] = 0
+    stat["disambiguated_body"] = 0
+    stat["disambiguated_char"] = 0
+    stat["disambiguated_dropped"] = 0
+    if domain and conf:
+        # 冲突消歧（2026-09-14 口径升级）：doc 侧同构三级裁决
+        # （摘要碎片串→正文碎片串→碎片字级和）——候选词按 semantic_atoms
+        # 切分形态计数，与 fm.semantic 匹配面完全同构；旧口径整词形 count
+        # 与摘要落盘形态（纹 身）错位，高频多义词（accept/describe/tattoo）
+        # 重灾区 11257 词被整词丢弃。词级候选 len 2-4——单字频次虚高排除；
+        # 全零丢弃（宁缺勿滥，不发明词形）。
+        texts = _atoms_texts()
+        for en in sorted(conf):
+            z, lv = _disambiguate(en, conf[en], texts)
+            if z:
+                cand[en] = z
+                stat["disambiguated"] += 1
+                if lv == 1:
+                    stat["disambiguated_body"] += 1
+                elif lv == 2:
+                    stat["disambiguated_char"] += 1
+                conf.pop(en)
+            else:
+                stat["disambiguated_dropped"] += 1
+    overlap = sorted(set(cand) & set(EN_ZH))
+    for w in overlap:
+        cand.pop(w)                                       # 手工表优先
+    out = {
+        "meta": {
+            "name": "cedict_en_zh",
+            "created_at": time.strftime("%Y-%m-%dT%H:%M:%S"),
+            "source": "CC-CEDICT (https://www.mdbg.net/chinese/dictionary)",
+            "license": "CC BY-SA 4.0（派生数据，独立文件署名，不入 MIT 内联表）",
+            "min_len": min_len,
+            "rules": "剥括号注/to-a-an 前缀；条目级子段（单词条释义才收）；"
+                     "纯字母；STOPWORDS 排除；冲突集 doc 语料频次消歧"
+                     "（三级同构：摘要碎片串→正文碎片串→碎片字级和，"
+                     "候选 2-4 字）；手工 EN_ZH 优先",
+            "domain_disambiguation": (
+                "locomo corpus567 摘要/正文 semantic_atoms 原子串同构计数裁决"
+                "（与 fm.semantic 匹配面形态一致），评测特化产物"
+                if domain else "无（纯唯一映射）"),
+            "stats": stat,
+            "n_map": len(cand), "n_conflict": len(conf),
+            "n_handmade_overlap": len(overlap),
+        },
+        "map": dict(sorted(cand.items())),
+    }
+    with io.open(OUT_JSON, "w", encoding="utf-8", newline="\n") as f:
+        json.dump(out, f, ensure_ascii=False, indent=1, sort_keys=True)
+    print("[build] 词级映射 %d 键（消歧 %d / 冲突跳过 %d / 手工重叠让位 %d）→ %s"
+          % (len(cand), stat["disambiguated"], len(conf), len(overlap), OUT_JSON))
+    print("[build] 统计: %s" % json.dumps(stat, ensure_ascii=False))
+    return 0
+
+
+def analyze():
+    """集成后 normalize 的最终残留取证（en_normalizer 已加载 cedict_en_zh.json，
+    此处 unknown_keep 即全链路翻译后的真 OOV）。"""
+    from md_cg.semantic.en_normalizer import normalize_en_query
+    unk = collections.Counter()
+    with io.open(QUESTIONS, encoding="utf-8") as f:
+        for line in f:
+            if not line.strip():
+                continue
+            q = json.loads(line)
+            terms, detail = normalize_en_query(str(q.get("question") or ""))
+            for d in detail:
+                if d.get("action") == "unknown_keep":
+                    unk[str(d.get("orig") or "").lower()] += 1
+    total_tokens = 0
+    kept_tokens = 0
+    with io.open(QUESTIONS, encoding="utf-8") as f:
+        for line in f:
+            if not line.strip():
+                continue
+            q = json.loads(line)
+            total_tokens += len(normalize_en_query(str(q.get("question") or ""))[0])
+    print("[analyze] 集成后最终 OOV：去重 %d 词 / %d 词次（残留原子序列长度 %d）"
+          % (len(unk), sum(unk.values()), total_tokens))
+    print("\n[analyze] top 30 最终残留（真缺口，只配手工定向补）：")
+    for w, n in unk.most_common(30):
+        print("  %4d  %s" % (n, w))
+    return 0
+
+
+def main():
+    mode = sys.argv[1] if len(sys.argv) > 1 else "analyze"
+    if mode == "download":
+        return download()
+    if mode == "build":
+        return build()
+    if mode == "analyze":
+        return analyze()
+    print(__doc__)
+    return 2
+
+
+if __name__ == "__main__":
+    raise SystemExit(main())

package/md_cg/lexicon/build_standard_en.py

@@ -0,0 +1,168 @@
+# -*- coding: utf-8 -*-
+"""build_standard_en.py · 英文标准库构建器（语素表 + 词条 → standard_en.json）
+
+词条语素带显式英文注 (zh_morph, en_gloss)，规避单字歧义（月=month/moon、生=person/living）。
+构建即验证：标准英文必须等于语素英文的 snake_case 直拼。
+"""
+import collections
+import json, io, os, re
+
+MORPHEMES = {}
+def M(zh, en):
+    MORPHEMES[zh] = en
+
+# 动物
+for zh, en in [("牛","cow"),("猪","pig"),("羊","sheep"),("鸡","chicken"),("鸭","duck"),("马","horse"),("狗","dog"),("猫","cat"),("鱼","fish"),("鸟","bird")]:
+    M(zh, en)
+# 食
+for zh, en in [("肉","meat"),("奶","milk"),("蛋","egg"),("米","rice"),("菜","vegetable"),("果","fruit"),("油","oil"),("盐","salt"),("糖","sugar"),("茶","tea"),("酒","wine"),("水","water")]:
+    M(zh, en)
+# 时间
+for zh, en in [("年","year"),("月","month"),("日","day"),("天","day"),("周","week"),("期","period"),("时","hour"),("分","minute"),("秒","second"),("春","spring"),("秋","autumn"),("冬","winter"),("夏","summer"),("今","this"),("明","next"),("去","past")]:
+    M(zh, en)
+# 数
+for zh, en in [("一","one"),("二","two"),("三","three"),("四","four"),("五","five"),("六","six"),("七","seven"),("八","eight"),("九","nine"),("十","ten"),("百","hundred"),("千","thousand")]:
+    M(zh, en)
+# 亲
+for zh, en in [("父","father"),("母","mother"),("兄","elder_brother"),("弟","younger_brother"),("姐","elder_sister"),("妹","younger_sister"),("儿","son"),("女","daughter")]:
+    M(zh, en)
+# 身
+for zh, en in [("头","head"),("手","hand"),("脚","foot"),("眼","eye"),("口","mouth"),("耳","ear"),("鼻","nose"),("心","heart"),("血","blood"),("骨","bone"),("皮","skin"),("脏","organ")]:
+    M(zh, en)
+# 自然
+for zh, en in [("山","mountain"),("水","water"),("火","fire"),("土","soil"),("木","wood"),("金","metal"),("光","light"),("电","electric"),("风","wind"),("雨","rain"),("雪","snow"),("星","star"),("海","sea"),("河","river"),("石","stone"),("云","cloud"),("月","moon")]:
+    M(zh, en)
+# 器
+for zh, en in [("车","vehicle"),("船","boat"),("机","machine"),("器","device"),("刀","knife"),("笔","pen"),("书","book"),("门","door"),("窗","window"),("桥","bridge"),("路","road"),("池","cell")]:
+    M(zh, en)
+# 动（基形）
+for zh, en in [("走","walk"),("来","come"),("看","see"),("听","listen"),("吃","eat"),("喝","drink"),("说","speak"),("读","read"),("写","write"),("买","buy"),("卖","sell"),("学","study"),("教","teach"),("做","make"),("跑","run"),("飞","fly"),("睡","sleep"),("想","think"),("爱","love"),("用","use"),("开","open"),("关","close")]:
+    M(zh, en)
+# 形
+for zh, en in [("大","big"),("小","small"),("高","high"),("低","low"),("新","new"),("旧","old"),("好","good"),("坏","bad"),("长","long"),("短","short"),("快","fast"),("慢","slow"),("热","hot"),("冷","cold"),("多","many"),("少","few"),("重","heavy"),("轻","light"),("深","deep")]:
+    M(zh, en)
+# 抽象
+for zh, en in [("力","force"),("气","gas"),("道","way"),("理","principle"),("文","text"),("化","transform"),("话","speech"),("影","shadow"),("音","sound"),("名","name"),("国","country"),("家","home"),("城","city"),("王","king"),("民","people"),("军","army"),("医","medicine"),("药","drug"),("病","disease"),("课","lesson"),("题","question"),("工","work"),("师","master"),("律","law"),("图","graph"),("谱","map"),("数","number"),("物","thing"),("地","earth"),("人","person"),("脑","brain"),("视","view")]:
+    M(zh, en)
+# 多字语素（化石词 / 复合单位）
+for zh, en in [("星期","week"),("经济","economy"),("逻辑","logic"),("沙发","sofa"),("咖啡","coffee"),("巧克力","chocolate"),("知识","knowledge"),("工作","work"),("纪律","discipline"),("记忆","memory"),("认知","cognition"),("月光","moon_light"),("十二","twelve"),("兄弟","brothers"),("姐妹","sisters"),("汽","gas")]:
+    M(zh, en)
+
+# 词条：(zh, [(语素, 英文注)...], 标准英文, legacy 别名, class)
+E = []
+DEFAULT_CLS = "compositional"
+
+
+def A(zh, glosses, std, legacy, cls=DEFAULT_CLS):
+    E.append((zh, glosses, std, legacy, cls))
+
+# 食物（语素直拼）
+A("牛肉", [("牛","cow"),("肉","meat")], "cow_meat", ["beef"])
+A("猪肉", [("猪","pig"),("肉","meat")], "pig_meat", ["pork"])
+A("羊肉", [("羊","sheep"),("肉","meat")], "sheep_meat", ["mutton"])
+A("鸡肉", [("鸡","chicken"),("肉","meat")], "chicken_meat", ["chicken"])
+A("鸭肉", [("鸭","duck"),("肉","meat")], "duck_meat", [])
+A("牛奶", [("牛","cow"),("奶","milk")], "cow_milk", ["milk"])
+A("鸡蛋", [("鸡","chicken"),("蛋","egg")], "chicken_egg", ["egg"])
+A("牛肉面", [("牛","cow"),("肉","meat"),("面","noodle")], "cow_meat_noodle", [])
+# 时间
+A("月份", [("月","month")], "month", [], "compositional")          # 份=名物化后缀，剔
+A("一月", [("一","one"),("月","month")], "one_month", ["January"])
+A("二月", [("二","two"),("月","month")], "two_month", ["February"])
+A("三月", [("三","three"),("月","month")], "three_month", ["March"])
+A("十二月", [("十二","twelve"),("月","month")], "twelve_month", ["December"])
+A("星期一", [("星期","week"),("一","one")], "week_one", ["Monday"])
+A("星期日", [("星期","week"),("日","sun")], "week_sun", ["Sunday"])
+A("今天", [("今","this"),("天","day")], "this_day", ["today"])
+A("明天", [("明","next"),("天","day")], "next_day", ["tomorrow"])
+A("今年", [("今","this"),("年","year")], "this_year", [])
+A("明年", [("明","next"),("年","year")], "next_year", [])
+A("去年", [("去","past"),("年","year")], "past_year", ["last year"])
+A("春天", [("春","spring")], "spring", ["spring"])                  # 天=季节后缀，剔
+A("冬天", [("冬","winter")], "winter", ["winter"])
+# 家庭
+A("父亲", [("父","father")], "father", ["father"])                  # 亲=亲缘标记，剔
+A("母亲", [("母","mother")], "mother", ["mother"])
+A("兄弟", [("兄弟","brothers")], "brothers", ["brothers"], "compositional")  # 复数是语义本身
+A("姐妹", [("姐妹","sisters")], "sisters", ["sisters"], "compositional")
+A("儿子", [("儿","son")], "son", ["son"])
+A("女儿", [("女","daughter")], "daughter", ["daughter"])
+# 身体
+A("眼睛", [("眼","eye")], "eye", ["eye"])                          # 睛=瞳，并入 eye
+A("心脏", [("心","heart"),("脏","organ")], "heart_organ", ["heart"])
+# 自然
+A("火山", [("火","fire"),("山","mountain")], "fire_mountain", ["volcano"])
+A("海水", [("海","sea"),("水","water")], "sea_water", ["seawater"])
+A("月光", [("月","moon"),("光","light")], "moon_light", ["moonlight"])
+A("雪山", [("雪","snow"),("山","mountain")], "snow_mountain", ["snow mountain"])
+# 科技
+A("电脑", [("电","electric"),("脑","brain")], "electric_brain", ["computer"])
+A("电话", [("电","electric"),("话","speech")], "electric_speech", ["telephone"])
+A("电视", [("电","electric"),("视","view")], "electric_view", ["television"])
+A("电影", [("电","electric"),("影","shadow")], "electric_shadow", ["movie", "film"])
+A("电池", [("电","electric"),("池","cell")], "electric_cell", ["battery"])
+A("火车", [("火","fire"),("车","vehicle")], "fire_vehicle", ["train"])
+A("汽车", [("汽","gas"),("车","vehicle")], "gas_vehicle", ["car"])
+A("飞机", [("飞","fly"),("机","machine")], "fly_machine", ["airplane"])
+A("手机", [("手","hand"),("机","machine")], "hand_machine", ["mobile phone"])
+# 学科
+A("数学", [("数","number"),("学","study")], "number_study", ["mathematics"])
+A("化学", [("化","transform"),("学","study")], "transform_study", ["chemistry"])
+A("物理学", [("物","thing"),("理","principle"),("学","study")], "thing_principle_study", ["physics"])
+A("地理学", [("地","earth"),("理","principle"),("学","study")], "earth_principle_study", ["geography"])
+A("生物学", [("生","living"),("物","thing"),("学","study")], "living_thing_study", ["biology"])
+A("热力学", [("热","heat"),("力","force"),("学","study")], "heat_force_study", ["thermodynamics"])
+# 人物
+A("医生", [("医","medicine"),("生","person")], "medicine_person", ["doctor"])
+A("学生", [("学","study"),("生","person")], "study_person", ["student"])
+A("律师", [("律","law"),("师","master")], "law_master", ["lawyer"])
+A("工人", [("工","work"),("人","person")], "work_person", ["worker"])
+# 领域词（认知图相关）
+A("知识图谱", [("知识","knowledge"),("图","graph"),("谱","map")], "knowledge_graph_map", ["knowledge graph"])
+A("认知图", [("认知","cognition"),("图","graph")], "cognition_graph", [])
+A("工作纪律", [("工作","work"),("纪律","discipline")], "work_discipline", [])
+A("记忆库", [("记忆","memory"),("库","store")], "memory_store", [])
+# 化石词（音译/古典，稳定单语素，保留既有英文）
+A("经济", [("经济","economy")], "economy", ["economy"], "opaque")
+A("逻辑", [("逻辑","logic")], "logic", ["logic"], "opaque")
+A("沙发", [("沙发","sofa")], "sofa", ["sofa"], "opaque")
+A("咖啡", [("咖啡","coffee")], "coffee", ["coffee"], "opaque")
+A("巧克力", [("巧克力","chocolate")], "chocolate", ["chocolate"], "opaque")
+
+
+def main():
+    problems = []
+    entries = []
+    for zh, glosses, std, legacy, cls in E:
+        join_zh = "".join(z for z, _ in glosses)
+        if not zh.startswith(join_zh):
+            problems.append("%s: 语素拼合 %s != 词条 %s" % (zh, join_zh, zh))
+            continue
+        join_en = "_".join(e for _, e in glosses)
+        if join_en != std:
+            problems.append("%s: 标准英文 %s != 语素拼接 %s" % (zh, std, join_en))
+            continue
+        if re.search(r"[A-Z]", std):
+            problems.append("%s: 标准英文含大写（违 snake_case）" % zh)
+            continue
+        entries.append({"zh": zh, "glosses": [{"zh": z, "en": e} for z, e in glosses],
+                        "standard_en": std, "legacy_en": legacy, "class": cls})
+    out = {
+        "meta": {"name": "lingshu-standard-en", "version": "0.1.0", "rules": "RULES.md",
+                 "note": "英文标准 = 中文语素序直拼（snake_case）；屈折归零；虚词不入名；音译化石词保留为 opaque；四问皆否不入库"},
+        "morphemes": MORPHEMES,
+        "entries": entries,
+    }
+    path = os.path.join(os.path.dirname(os.path.abspath(__file__)), "standard_en.json")
+    with io.open(path, "w", encoding="utf-8", newline="\n") as f:
+        json.dump(out, f, ensure_ascii=False, indent=1)
+    cls_cnt = collections.Counter(e["class"] for e in entries)
+    print("entries=%d morphemes=%d classes=%s problems=%d" % (
+        len(entries), len(MORPHEMES), dict(cls_cnt), len(problems)))
+    for p in problems:
+        print("  !", p)
+    return 1 if problems else 0
+
+
+if __name__ == "__main__":
+    raise SystemExit(main())