谁的话算「标准」

标准语是被挑出来的那一支，不是长得最好的那一支

先说意大利。1861 年统一时，这个新国家里能用意大利语的人少得惊人：De Mauro 算出 2.5%，Castellani 1982 年用更宽的口径重算得到约 9.5%。两个数差三倍多，至今没有定论，但指向同一件事——这门「国语」当时是极少数人的书面语。它的底子是十四世纪的佛罗伦萨话，但丁、彼特拉克、薄伽丘写作用的那一支。选中它的不是语音更清楚或语法更整齐，是这三个人的书被抄得最多。

再回到开头那场会。1913 年的读音统一会定出后来所谓的「老国音」：以北京音为底，保留入声（音节收得极短，末尾像被掐断），又区分尖团（「尖」和「坚」在这套音里声母不同）。南北都不吃亏，设计得很用心。麻烦在于它没有母语者——示范朗读的人得专门受训才发得准，教师自己都拿不稳。到 1932 年，教育部公布《国音常用字汇》，改按北平音定标准。

摆在一起，标准化的步骤就露出来了：先从现有变体里选一支，写下来（编词典、写语法书，让它有固定的样子），扩充用途（法律、教科书、新闻都改用它写），最后等人接受。这四步没有一步跟这支话本身的性质有关，只跟谁的书传得广、谁定课本、谁上电台有关。

而且第一步并非必需。挪威有两套官方书面标准，各市镇自选；口语则干脆没有标准，议会发言和晚间新闻里人们说各自家乡的话。所以「标准语」和「方言」的差别不是语言学上的差别：给你一段转写的话，你没法从语法里判断它标不标准，只能查它有没有被写进词典、教进学校。

EN Standard varieties are selected, not grown. Standard Italian rests on fourteenth-century Florentine because Dante, Petrarch and Boccaccio were copied most widely; at unification in 1861 De Mauro estimated 2.5% of the population could use it, while Castellani's 1982 recount put the figure near 9.5% — an unsettled dispute whose two answers point the same way. China's 1913 Commission on the Unification of Pronunciation fixed readings for over six thousand characters, keeping the entering tone and the sharp/rounded distinction; nobody spoke it natively, teachers could not hold it steady, and by 1932 the Ministry of Education switched the standard to Beijing pronunciation. The steps are always the same — select a variety, codify it, extend it to new functions, get it accepted — and none of them depends on properties of the variety itself. Nor is step one required: Norway maintains two written standards and no spoken standard at all, with dialects used in parliament and on the evening news.

那条界线画在国界上，不画在语言上

荷兰和德国之间的边界地带是个老例子。标准语普及之前，住在荷兰东部边境的人听懂界那边德国村子的话，比听懂本国西部的更容易，德国一侧也一样。方言是连续变化的：每往前几十公里变一点，相邻两处永远互通，走远了才听不懂。可两侧的人在学校里学的不是同一套书面语——这边往标准荷兰语靠，那边往标准德语靠。

反过来是斯堪的纳维亚。丹麦语、挪威语、瑞典语各有正字法、词典和管理机构，算三门；可它们同处一个连续体，三国的人各说各的大体能谈下去，而且互通还不对称——北欧做过的调查里，挪威人听懂邻居最容易，丹麦人和瑞典人互相最费劲。按能不能互通划，这三门算一门；按有没有各自的国家和正字法划，就是三门。没有哪条干净的标准能分开这两个名字——而把某一支叫「方言」不是说它简单或不完整，只是说它没走完上一节那四步。

EN Before standard languages spread, people on the Dutch side of the German border understood the villages just across it more easily than the far west of their own country, and the reverse held too. Dialects vary continuously: neighbours always understand each other, distant ends do not, and nowhere along the chain is there a break. What created "two languages" was schooling pulling each side toward Standard Dutch or Standard German. Scandinavia shows the mirror image: Danish, Norwegian and Swedish each have their own orthography, dictionaries and regulatory bodies, yet sit in one continuum and remain broadly mutually intelligible — asymmetrically so, with Norwegians understanding their neighbours most easily and Danes and Swedes struggling most with each other. Intelligibility would make them one language; states and orthographies make them three. There is no clean linguistic criterion separating language from dialect, and calling a variety a dialect says nothing about its complexity — only that it has not been through selection, codification, elaboration and acceptance.

被判成「错」的那些，多半是旧规则，或者某个人的偏好

英语里有条规矩：句子不能用介词结尾。源头查得到。1672 年，诗人 Dryden 挑剔前辈 Ben Jonson，说介词放句末是他常犯的毛病，没说为什么不好。九十年后，Lowth 主教 1762 年的语法书措辞其实很克制：这是我们这门语言强烈倾向的说法，日常对话里通行，也合乎亲切的文体，只是把介词放到前面更雅致、更配庄重的文体。有意思的是，他写这句话时用的正是介词收尾的格式。一条描写偏好的话，被后来的课本改造成了禁令。

「不定式中间不能插副词」更晚：1803 年被提出，1864 年经 Alford 推广开，背景是十九世纪一批语法书热衷于把拉丁语的规矩搬进英语——拉丁语的不定式是一个词，中间插不进东西。双重否定也一样：乔叟和莎士比亚笔下很常见，是十八世纪的语法书引入「两个负号相乘得正」这类算术论证才把它判出局。

中文这边，「的」和「地」的分工是二十世纪初定下的书面约定。白话文兴起、大量西文作品译进来的那几十年里，译者仿照英语法语区分形容词与副词词尾的办法，有意把这两个字分开用；当时还有第四个字「底」专管领属（「我底书」），后来没留住。口语里它们都读 de，说话时根本没这个区分——它从头就是写字的人之间的协调约定。

更要紧的是另一半：被判成「不规范」的说法，往往自己有一套严密的规则。非裔美国人英语里常省略系动词，He is nice 说成 He nice。Labov 在 1969 年指出，省略的位置一点也不随便：标准英语能把 is 缩成 's 的地方，这套系统就能省；不能缩的地方，它也不能省。He's nice 缩得成，于是 He nice 成立；而 I don't know where he is 句末的 is 缩不成，这套系统里也就不能说成 I don't know where he。这不是把规则放松了，是另一套规则——要知道哪里能省，你得同时握着两套系统。

EN The ban on ending a sentence with a preposition traces to Dryden, who in 1672 faulted Ben Jonson for it without saying why. Lowth's 1762 grammar was far more measured: the construction "is an Idiom which our language is strongly inclined to; it prevails in common conversation, and suits very well with the familiar style in writing" — and the sentence in which he wrote that itself ends with a preposition. A note on register became a prohibition in later textbooks. The split-infinitive rule arrived in 1803 and was popularised by Alford in 1864, part of a nineteenth-century habit of importing Latin constraints; Latin infinitives are single words. Double negatives were ordinary in Chaucer and Shakespeare until eighteenth-century grammarians applied arithmetic to them. In Chinese, the written division of labour between 的 (de, attributive) and 地 (de, adverbial) was settled in the early twentieth century in imitation of European adjective and adverb endings — a fourth character 底 (de, possessive, as in 我底书 wǒ de shū, "my book") did not survive — and none of it exists in speech, where all are pronounced alike. Meanwhile, stigmatised varieties turn out to be rule-governed. Labov showed in 1969 that African American English deletes the copula exactly where Standard English can contract it: He is nice contracts to He's nice, so He nice is well formed, while the final is in I don't know where he is cannot contract, and cannot be deleted either.

你听到的口音里，有一部分是你自己装进去的

1960 年前后在蒙特利尔做过一个实验：找一批英法双语者，每人录一段英语、一段法语，打乱顺序放给听众评价「说话的这些人」——听众不知道听到的是同一批人。结果英语那一版在聪明、可靠、有雄心这类评价上得分更高，而拉开差距的不只是英语母语的听众，法语母语的听众给法语版打得更低。这说的不是两门语言怎么样，是当时那座城市里谁占着经济位置；换个地方重做，方向跟着当地的权力关系变。

更尖锐的是 Rubin 1992 年那个设计。美国大学生听同一段录音——讲课的是美国英语母语者——一半人同时看到一张白人照片，另一半看到一张东亚面孔的照片，两张的外貌吸引力事先匹配过。看到东亚面孔那组报告自己「听到了外国口音」，随后的听力理解测验分数也更低。声音一模一样。

这个结果后来被仔细追查过。Zheng 与 Samuel 2017 年先用静态照片复现出同样的偏移；但把照片换成配音视频、掺进口音无歧义的样本、或者把两种面孔混着随机出现，偏移就减小、消失，甚至反过来。他们的判断是：变的是评分，不是听觉本身——你没有真的听见另一种声音，你是给同一个声音打了不同的分。这个区分反而让结论更该被认真对待：落到成绩单和绩效反馈上的正是评分那一端。

态度还会反过来改变人怎么说话。Labov 在纽约的分层调查里，场合越正式，人越往心目中那条线上靠；靠得最用力的却不是最上面那一层。念词表这种最正式的任务上，下中层发 r 的比例反超了上中层——他们最清楚哪种说法「体面」，也最在意，于是用过了头。

EN Around 1960 in Montreal, bilingual speakers each recorded a passage in English and one in French; listeners, told they were judging different people, rated the English guises higher on intelligence, dependability and ambition — and French-speaking listeners rated the French guises lowest of all. That result is about who held economic position in that city, not about the two languages; repeated elsewhere, its direction tracks local power relations. Rubin's 1992 study is sharper: American undergraduates heard one recording made by a native speaker of American English while seeing either a white or an East Asian face, matched for attractiveness. Those shown the Asian face reported hearing a foreign accent and scored lower on a comprehension test. Zheng and Samuel followed this up in 2017: they reproduced the shift with static photos, but it shrank, vanished or reversed once the photos became dubbed videos, once unambiguous items were mixed in, or once the two ethnicities were interleaved rather than blocked. Their reading is that expectation moves the judgement, not the percept — which matters more, not less, because judgement is what lands on a transcript or a performance review. Attitudes also reshape production: in Labov's New York survey the lower middle class overshot the upper middle class on r-pronunciation in the most formal task of all, reading word lists.