语言像河流一样变

掉的不是一个词的 k，是所有 kn- 的 k

先看证据。1679 年的一本英语词典把 knave 和 nave、knight 和 night、knead 和 need 列成同音词——说明到那时候这几对已经读得一模一样了。整个十七世纪里，k 和 g 在 n 前面陆续消失，拼写留在原地，读音走了。

关键不在"掉了"，在"一起掉"。没有哪个说话人坐下来一个词一个词地决定。变的是舌头在"词首 + 鼻音"这个位置上怎么动，而这个位置在每个词里都是同一个。所以音变看起来才这么整齐：它作用的对象是音，不是词。

最有名的一批对应就是这么来的。拉丁语保留着老样子的地方，日耳曼语这一支整批换了：pater 对 father，piscis 对 fish（p 变 f）；tres 对 three（t 变 th）；cornu 对 horn（k 变 h）。这不是四个巧合，是三条规律各跑过几千个词。十九世纪七十年代的一批德国学者把它推到极端：音变没有例外——有例外就说明条件还没找到。

这个口号差点被证伪，又被救了回来。上面那批对应确实有一批不听话的词。1875 年 Verner 发现它们取决于更早的重音落在哪个音节上：重音在前的保持清音，不在的被浊化了。英语里还留着一个现成的化石——同一个动词的 was 和 were，一个是 s，一个是 r。那个 r 是从被浊化的 s 变来的。例外被收编进规律，规律反而更硬了。

这事不限于印欧语。中古汉语的音节可以用 -p、-t、-k 收尾，粤语今天还带着：六 luk6、国 gwok3，尾巴上那个塞音只是把气流一堵、不放开。官话这一支整批丢掉了这个位置上的塞音，同样两个字成了 liù、guó。不是某几个字掉了，是这个位置塌了。（两边都不存在"保留得好"或"退化"——它们只是各自走了不同的路。）

但地基没那么平。王士元等人在六十年代末研究汉语方言时提出另一种图景：变化可能是一次性的，却一个词一个词地扩散过去，扩到一半就停下，留下"残余"——潮州话里有一类字分成了两种读法，找不到任何语音条件能解释这个分岔。Labov 1981 年调停说两种机制都在跑，各管一摊。证据强度：整批对应靠的是几千个词级别的系统对应，非常硬；词汇扩散靠的是方言调查里那些说不通的残余，也是真数据。今天没人只要一边。

EN A 1679 English dictionary lists knave/nave, knight/night and knead/need as homophones: by then the k had gone. Through the seventeenth century, k and g dropped before n across the whole vocabulary — knee, know, knife, knock — while the spelling stayed put. The point is that they went together. Nobody decides word by word; what drifts is how the tongue handles a position, and that position is the same in every word. Hence the classic correspondences: Latin pater/piscis against English father/fish, tres against three, cornu against horn — three regularities running over thousands of words, not four coincidences. The Neogrammarians pushed this to a slogan: sound change has no exceptions, and an apparent exception means you have not found the conditioning yet. Verner rescued the slogan in 1875 by showing that the leftover cases depended on where the older accent fell; English was/were is a surviving fossil of that voicing, the r descending from a voiced s. The same regularity shows up outside Indo-European: Middle Chinese syllables could end in -p, -t, -k, still audible in Cantonese 六 (luk6, "six") and 國 (gwok3, "country"), and lost wholesale in Mandarin (liù, guó) — a position collapsing, not a handful of words. Against this, Wang and colleagues working on Chinese dialects proposed lexical diffusion: a change can be abrupt but spread word by word and stall, leaving residue with no phonetic conditioning, as in a Chaozhou tone split. Labov's 1981 settlement: both mechanisms operate. Regular correspondence rests on thousands of words; diffusion rests on dialect residue. Both are real data.

"going to" 正在被磨成一个语法零件，而且磨到一半

说 I'm going to London（我要去伦敦）的时候，真的有人在移动。说 I'm going to tell you（我打算告诉你）的时候，没有任何人在走路。第二种用法大约从十五世纪末开始出现，1483 年的文献里已经有例子。一个表示"走"的动词，被征用成了表示"将来"的标记。

三件事同时发生：意思被漂白，位移的部分没了；形式被磨短，going to 缩成 gonna；位置被固化，它不再自由地当动词用。整个过程叫语法化（grammaticalization，实词变成语法成分）。

有个干净的检验，英语母语者自己就能验：gonna 后面只能接动词。I'm gonna tell you 可以，I'm gonna the store 不行——可 I'm going to the store（我要去那家店）完全正常。同一个来源已分成两条路，被磨短的那条没有继承"走"的用法。磨损标记出了分叉点。

汉语里同一条路走得更远。"了"本来是动词，读 liǎo，意思是完结，十世纪的白话文献里还这么用；后来它先被重新分析成动词后面表结果的成分，再被磨成体标记，读轻声的 le，不能单说、也不能带宾语。"把"本来是动词，意思是握住、拿；从唐代后期起虚化成宾语标记——"把门关上"里的"把"已经没人在握任何东西了。"给"走的也是同一条路：从"给予"这个动作，滑到"被"和"替"的位置上。

这不是碰巧。Bybee、Perkins 和 Pagliuca 1994 年查了 76 种语言、25 个语系，发现"位移动词 → 将来标记"这条路在毫无亲缘关系的语言里反复出现（法语说 je vais partir，字面也是"我去 + 离开"）。他们还提出了单向性：实词变虚词是常态，虚词变回实词极少。

单向性的强度要标清楚：它是很强的统计倾向，不是定律。反例存在但稀少——英语的词缀 -ism 被拽出来当名词用（all those isms），介词 up 被拽出来当动词用（up the ante，加注）。这类例子被专门收集过，数量远不足以推翻趋势。

引擎是频率。只有被说得足够多的组合才会被磨：going to 缩成 gonna，结构一样的 swimming to 不会。这也顺带解释了一件你早就见过、没想过的事——为什么各种语言里的语法零件几乎都很短。它们是被使用次数磨短的。

EN In I'm going to London somebody is actually moving; in I'm going to tell you nobody is. The second use appears from the late fifteenth century, with examples by 1483: a motion verb conscripted as a future marker. Three things happen at once — the meaning bleaches (the motion is gone), the form erodes (going to becomes gonna), and the slot hardens. That whole process is grammaticalization. A clean diagnostic any English speaker can run: gonna takes only a verb. I'm gonna tell you is fine; I'm gonna the store is not, although I'm going to the store is perfectly normal. The eroded form did not inherit the motion use, so erosion marks the fork. Chinese has gone further down the same road. 了 was a verb read liǎo, "to finish", still used that way in tenth-century vernacular texts; it was reanalysed as a resultative element after the verb, then ground down into the perfective particle le, toneless, unable to stand alone. 把 was a verb meaning "to grasp"; from the late Tang it bleached into an object marker — in 把门关上 (bǎ mén guān shàng, "close the door") nobody is holding anything. 给 (gěi, "to give") slid the same way toward passive and benefactive uses. None of this is coincidence: Bybee, Perkins and Pagliuca surveyed 76 languages in 25 phyla in 1994 and found the motion-verb-to-future path recurring in unrelated languages, French je vais partir among them. They also proposed unidirectionality — lexical to grammatical, rarely back. Strength: a strong statistical tendency, not a law. Counterexamples exist and are rare (the suffix -ism pulled out as a noun, up used as a verb in up the ante). The engine is frequency: going to erodes to gonna, swimming to does not. Which is why grammatical elements across languages tend to be short — usage ground them down.

没有一场变化是全体同时改的，它总是从某些人身上开始

开头那份罗马清单有名字，叫 Appendix Probi，227 条"要 A 不要 B"。其中一条是 speculum non speclum：要写 speculum（镜子），别写 speclum。那个被判错的 speclum，正是西班牙语 espejo 和意大利语 specchio 的直系祖先。清单抄写者看见的不是错误，是正在发生的变化——词中间那个不重读的元音被挤掉了。

变化在时间上有固定的形状。Kroch 1989 年用 Ellegård 整理的历史语料看英语里的 do：Did you see him? 里那个不表任何意思的 do 不是自古就有的，它从十三世纪的零星用例走到近代英语的强制使用，走了几百年。把使用比例对时间画出来是一条 S 形：开头很慢，中间陡升，末尾又慢下来。

真正有意思的是他接着发现的事。同一场变化会在好几种句法环境里同时进行——疑问句、否定句、强调句——这些环境的起点差得很远，有的早就开始了，有的还在零附近；但它们上升的斜率是一样的。他把这叫恒定速率效应。这句话的分量在于：上升的快慢不是由环境决定的，各个环境只是上车时间不同。

那么是谁先上的车？Labov 1963 年去了玛莎葡萄园岛。岛上两个元音的读法本来在向大陆靠拢，他发现一部分人反着走，把它拉回更"岛上"的读法，而这部分人集中在打算留在岛上、对夏季游客有意见的年轻渔民身上。发音在这里成了身份标记，虽然没人是有意识这么用的。1966 年他又在纽约几家百货公司数店员发不发词尾的 r：不同店比例不同，年轻人更常发。

这给了历史语言学一个省事的办法，叫表观时间：在同一个时间点比较不同年龄的人，当作变化过程的快照。前提得摆出来——假设人成年后发音基本不动，而这只是近似。

所以"语言在变坏"说的其实是另一件事：你把自己学会的那一版当成了尺子。河道一直在挪，没有哪一段是它本来该在的地方。

EN The Roman list has a name: the Appendix Probi, 227 entries of the form "A not B". One reads speculum non speclum — write speculum ("mirror"), not speclum. The condemned form is the direct ancestor of Spanish espejo and Italian specchio. What the scribe saw as an error was a change in progress: an unstressed medial vowel being squeezed out. Change also has a shape in time. Kroch, in 1989, used Ellegård's historical corpus on English do: the meaningless do in Did you see him? was not always there, taking centuries to go from scattered thirteenth-century uses to obligatory in modern English. Plot proportion against time and you get an S: slow, then steep, then slow. The interesting part came next. The same change runs simultaneously in several syntactic environments — questions, negations, emphatics — whose starting points differ widely, yet whose slopes are the same. Kroch called this the constant rate effect: the environments board at different times, but the rate is not set by the environment. As for who boards first: Labov went to Martha's Vineyard in 1963 and found that two vowels drifting toward mainland pronunciation were being pulled back toward the island variant, concentrated among younger fishermen who meant to stay and had views about the summer tourists. Pronunciation had become an identity marker, with nobody using it consciously. In 1966 he counted whether New York department-store clerks pronounced a final r, and found the proportion differed by store and was higher among the young. That gave historical linguistics apparent time: compare age groups at one moment as a snapshot of change. Its premise should be stated — that pronunciation is mostly fixed after adulthood, which is only an approximation. So "the language is decaying" is really a statement about your own yardstick: the version you learned. The river keeps moving its bed, and no stretch of it is where it was supposed to be.