语言正在消失

「两周一门」是推算出来的，数出来是三个月一门

1992 年那篇文章估计只有约一成语言长期安全，多达一半可能已经没有小孩在学。作者本人没说过「两周一门」——后来的人把「一百年里掉九成」摊成了速率，一个关于分布的估计被压成了听起来很精确的节拍。

另一条路是直接数：濒危语言编目（Catalogue of Endangered Languages）统计近几十年有明确记录、确实不再有人使用的语言，得到三个月一门左右。同一份目录里仍在使用但被判为濒危的有 3116 门，约占现存语言的 45%，另一套独立数据库用自己的量表算出约 44%——这在本领域算难得的一致。不过分母本身就是判断：一段话算独立的一门还是某门语言底下的一支，没有干净判据，这正是「谁的话算『标准』」那期的结论。

更有信息量的是 2022 年的一项研究：6511 门语言配上 51 个预测变量，推出本世纪末可能有约 1500 门不再有人说。两个结果跟常识拧着：跟别的语言接触本身不是驱动因素，多语共存能稳定维持很久；而受教育年限更长的地方濒危程度反而更高——不是教育有害，是学校通常只用一门主导语言授课，在孩子这一代抢走了一个关键场合。证据强度：大样本相关分析，不是因果实验。

EN The familiar "a language dies every two weeks" is not a count. Krauss (1992) estimated that only about 10% of languages were safe long-term and up to half might already have no child learners; the rate was later derived by spreading a 90%-by-2100 projection across a century. The Catalogue of Endangered Languages instead counted documented recent extinctions and arrived at roughly one language every three months, about four a year. It lists 3,116 still-spoken endangered languages, about 45% of those remaining; Ethnologue's independent EGIDS-based figure is about 44%. The denominator is itself a judgement call, as the standard-language topic argued. Bromham et al. (2022) modelled 6,511 languages against 51 predictors and projected about 1,500 could cease to be spoken by 2100; two findings cut against intuition — contact with other languages is not itself a driver, and more years of schooling predicts greater endangerment, because schooling usually runs in one dominant language. Evidence strength: large-sample correlational, not causal.

分界线不是最后一个说话人，是最后一个学会它的孩子

1991 年 Fishman 提出一个分级量表，其中最要紧的是一级：家庭、邻里、社区之间的日常传递。他的判断很硬——这一级不满足，其余所有措施至多是拖延时间。所以说话人数会骗人：五万个成年人会说但没有一个孩子在学，比五百人里孩子还在学更接近终点，前者的曲线还能平着走四十年再一起掉下去。通行的「濒危」定义正是按这个下的——育龄那一代不再把它传给孩子。

而且这不是某天突然发生的。语言按场合一个个退：先退出公文和法庭，再退出工作和买卖，再退出学校，然后是朋友之间，最后才是家里的饭桌。最后守住的那个场合最迟丢，也最容易让人误判——它还在，于是看上去整件事还在。

好在有更早的信号。Dorian 从 1963 年起追踪苏格兰东萨瑟兰几个渔村的盖尔语，跟了四十年，1981 年那本书提出「半个说话人」（semi-speaker）：听得懂也能开口，但产出的形式变了。这套方言的名词复数本来有好几种做法——换词干里的元音、加词尾、甚至把词尾的音去掉；最年轻的那批人几乎一律改成加同一个后缀，或者干脆不标。形态上的这种收缩比人口曲线早得多，而且不依赖人口统计就看得见。

EN Fishman's 1991 graded scale singles out one level as decisive: everyday transmission in home, neighbourhood and community. If that level fails, he argued, everything else amounts to biding time. So speaker counts mislead — fifty thousand adults with no child learners is closer to the end than five hundred speakers whose children are still learning; the first curve stays flat for forty years and then drops together. Shift proceeds domain by domain: official writing, then work and trade, then school, then friends, and the kitchen table last, which is why the surviving domain is the most deceptive. There is an earlier signal that does not depend on demography. Dorian tracked East Sutherland Gaelic in three Scottish fishing villages from 1963 for four decades and named the "semi-speaker": people who understand and speak but whose forms have changed. The dialect built noun plurals several ways — vowel change in the stem, suffixes, even subtraction of a final segment — and the youngest speakers levelled almost everything to one suffix, or marked no plural at all. Morphological contraction arrives well before the population curve turns.

消失的到底是什么：三个问题，三个不一样的答案

第一个问题是这门学科的。1963 年 Greenberg 从三十门语言里归纳出一批共性，头一条是：主语几乎总在宾语前面。六十年代末，Derbyshire 到亚马逊描写此前没人写过的 Hixkaryána（巴西的加勒比语系语言，说话人几百），发现它的基本语序把宾语放最前：「美洲豹抓住了那个人」说成 toto yahosïye kamara——先出口 toto（那个人），然后 yahosïye（抓住了他），最后才是 kamara（美洲豹）。跨 1377 门语言的样本里，宾语开头的只有十一门。所以那些「所有语言都……」的句子成不成立，取决于最边缘的几十门还在不在——恰好是说话人最少、最先消失的那批。

第二个问题是社群的，也最容易被讲坏。常见讲法是「一个世界观没了」。但「沃尔夫假说的兴衰」那期交代过，强版本没站住，这个论证撑不起它要撑的东西——而且它不需要。人们自己报告失去的是具体的：能不能跟上一辈把话说完整、仪式还办不办得下来、地名还认不认得出。另有一组常被引用的证据：2007 年一项研究按不列颠哥伦比亚省的原住民族群统计，一半以上成员报告能用本族语言对话的社区，青年自杀率接近零，不到一半的高出约六倍。证据强度：社区层面的相关，不是个体层面，也推不出因果——原作者本来就是把它和另外六个文化延续指标一起比的。

第三个问题是可逆性。Wampanoag 这门语言十九世纪以后没有说话人了，但留下一批文献：1663 年那部完整的圣经译本，还有一叠地契、遗嘱和请愿书，都是当年 Wampanoag 人自己用这门语言写的。1993 年 jessie little doe baird 发起 Wôpanâak 语言复原计划，从这批材料里把语言重建出来，她 2004 年出生的女儿成为一百多年来第一个母语者；现在部落办着一所全日沉浸式学校 Mukayuhsak Weekuw（「孩子的房子」）。文献不能让一门语言活着，但它决定了「还有没有可能」的答案。

EN Three questions, three answers. For the field: Greenberg's 1963 survey of thirty languages proposed that subjects almost always precede objects. In the late 1960s Derbyshire described Hixkaryána, a previously undescribed Cariban language of Brazil with a few hundred speakers, and found object-initial basic order — toto yahosïye kamara, literally "the man / it-grabbed-him / the jaguar". In a 1,377-language sample only eleven languages are object-verb-subject, clustered in one part of Amazonia. Our "all languages …" claims therefore depend on the few dozen languages most likely to go first. For communities: the common claim that a worldview dies overstates what the Whorf topic showed survives, and it is not needed. What people report losing is concrete — talking with elders, ceremony, place names. Hallett, Chandler and Lalonde (2007) found that among First Nations bands in British Columbia, youth suicide rates approached zero where over half of members reported conversational knowledge of their own language, and were about six times higher below that threshold; evidence strength: band-level correlation, not individual-level and not causal, compared alongside six other continuity measures. For reversibility: documentation and use are different things. Wampanoag lost its speakers in the nineteenth century but left a complete 1663 Bible translation plus deeds, wills and petitions written by Wampanoag people. The Wôpanâak Language Reclamation Project, begun in 1993 by jessie little doe baird, rebuilt the language from them; her daughter, born in 2004, became the first native speaker in over a century, and the tribe now runs the immersion school Mukayuhsak Weekuw, "the children's house".

复活成功过，但成功的从来不是「恢复原样」

唯一一个从「没有母语者」变回「一个社群日常第一语言」的案例是希伯来语。但前提很不普通：它从未停止被书写和诵读，使用者是一批彼此没有共同口语的移民，后面还接上了国家力量。结果本身也有争论——Zuckermann 主张现代以色列的这门语言是双源混合体，词汇来自希伯来语，语法底子带着复兴者的母语意第绪语。这场争论本身反倒说明了「复活」是什么意思。

更可复制的是「语言巢」（language nest）。1982 年毛利人办起第一所 kōhanga reo：学龄前的孩子整天待在那里全程只用目标语言，带他们的是还在说这门语言的老人。1983 年夏威夷照这个模式办起 Pūnana Leo；当时清点，全州十八岁以下能流利说夏威夷语的不到五十人，基本都在 Niʻihau 岛上，现在是数千。毛利语这边，2023 年人口普查有 213,849 人报告能用毛利语对话，比 2018 年多约 15%——这是自报能力，不等于日常使用。

还有一个小到好记的例子。1974 年 Ned Maddrell 去世，他是最后一位在说曼语的社群里长大的人。2009 年联合国教科文组织把曼语列为已灭绝；那所曼语小学的学生写信去问——用曼语写的——如果我们的语言已经灭绝了，那我们现在写信用的是什么？分类后来被改成极度濒危。

摆在一起看，成功的项目做的是同一件事：把语言塞回一个每天必然发生的场合，而那通常就是照看小孩的那几个小时。失败的往往不是没投钱，是钱只投在课时、教材和拼写规范上——那都在量表更靠外的级别，够不到家里那一级。这正是「分界线不是最后一个说话人」那节的反向验证。至于复活出来的语言和原来不完全一样：那从来不是可以要求的标准，没有哪门语言在三代之后还和自己一样。

EN Hebrew is the one case where a language with no native speakers became a community's everyday first language, but its preconditions were unusual: continuous liturgical and written use, a population of immigrants with no shared spoken language, and later state backing. Even the outcome is disputed — Zuckermann argues that the modern Israeli language is a hybrid whose grammar carries the revivalists' native Yiddish beneath a Hebrew lexicon, though its speakers believe they speak revived Biblical Hebrew. The reproducible method is the language nest: the first kōhanga reo opened in 1982, with preschoolers spending whole days in the target language alone, taught by elders who still spoke it; Hawai'i followed with Pūnana Leo in 1983. A 1983 count found fewer than fifty fluent speakers of Hawaiian under eighteen statewide, almost all on Ni'ihau; today the figure is in the thousands. New Zealand's 2023 census recorded 213,849 people reporting conversational Māori, up about 15% from 2018 — self-reported ability, which everywhere exceeds daily use. Manx offers the sharpest miniature: Ned Maddrell, the last person raised in a Manx-speaking community, died in 1974; UNESCO listed the language as extinct in 2009, pupils at the Manx-medium primary school wrote to ask, in Manx, what language they were writing in, and the classification was revised to critically endangered. What the successes share is putting the language back into a daily, unavoidable occasion — usually the hours spent looking after small children. Funding spent only on class hours, textbooks and orthography touches the outer levels of the scale and never reaches that one.