汉语哪里特别

声调不稀罕，稀罕的是你的脑子把它当成什么

先把它拆开听。同一个音节 ma，四个调是四个不相干的词：第一声平着高住不动（妈 mā，"母亲"）；第二声从中往上滑，像英语把话说成疑问句时尾巴挑起来（麻 má，"麻"）；第三声先掉到很低，常带一点挤出来的嘎吱声（马 mǎ，"马"）；第四声从高往下砸，像不耐烦地说 "No!"（骂 mà，"责骂"）。

听着很特殊，数一下就不特殊。跨语言数据库 WALS 收的 527 门语言里，307 门没有声调，132 门有简单声调（只分高低两档），88 门有复杂声调（像普通话这样三档以上）。四成多的语言用音高区别词义，撒哈拉以南非洲、中美洲、中南半岛成片都是。

真正有意思的在脑子里。2000 年有人让泰语母语者和汉语母语者听同一批泰语声调。泰语母语者处理这些音高时，左半球（通常管语言那侧）明显活跃；汉语母语者听同样的声音，这块活跃没出现——泰语那套调对他们不区别词义，于是被当成单纯的音高处理掉了。音高算语言还是算旋律，取决于它在你这门语言里干不干活。证据强度：小样本脑成像，另一套方法在粤语母语者身上结果一致。

还有一层更反直觉的：这套声调不是与生俱来。1954 年有人拿越南语证明，声调是词尾辅音掉光后留下的影子——原先靠词尾区分的词，词尾没了，只剩它在元音上留下的音高痕迹，音高被迫接管了区别词义的活。汉语走的同一条路：上声来自失落的喉塞，去声来自失落的 -s。这个「最中国」的特征只有一千多年，而同一片土地上一门不同语系的语言差不多同时独立走成了同一个样子。

EN Mandarin's four tones on one syllable — 妈 (mā, "mother"), 麻 (má, "hemp"), 马 (mǎ, "horse"), 骂 (mà, "to scold") — look exotic until you count. WALS samples 527 languages: 307 have no tone, 132 a simple system, 88 a complex one, so over 40% use pitch to distinguish words, in dense belts across sub-Saharan Africa, Mesoamerica and mainland Southeast Asia. The interesting part is neurological. Gandour and colleagues (2000) played the same Thai tones to Thai and to Chinese listeners: Thai speakers showed left-hemisphere activation, Chinese speakers did not, although they speak a tone language themselves. Pitch is linguistic or musical depending on whether it does work in your language, not on the sound. Evidence strength: small-sample imaging, later corroborated by mismatch-negativity work with Cantonese speakers. Tone is also young. Haudricourt showed in 1954 that Vietnamese tones descend from lost final consonants; Chinese took the same path, with the rising tone from a lost glottal stop and the departing tone from a lost final *-s. The most "Chinese" feature is about fifteen centuries old, and an unrelated neighbour invented it independently at roughly the same time.

数数之前，得先说清那个东西长什么样

英语说 three fish，汉语说「三条鱼」，多出来的「条」就是量词，少了它句子不成立。挑哪个也不能乱来：条 tiáo 管细长柔软的（鱼、路、蛇、绳子），张 zhāng 管能摊开的平面（纸、桌子、床、嘴），把 bǎ 管有把手能抓住的（伞、刀、钥匙）。报数之前，你已经被迫对形状表了态。

这一样也不稀罕。WALS 另一份 400 门语言的样本里，260 门没有量词，62 门可有可无，78 门跟汉语一样强制，成片分布在东亚、中南半岛、中美洲、亚马逊和西非。

更有意思的是这些格子怎么长大。它们从几何出发，很快跑出几何：「条」也管新闻、短信。日语的 本（hon）本来管长条状的东西（铅笔、领带、树），今天也管电话、柔道比赛、棒球的安打。两门语言各自把「细长」往外拉，方向还不一样。

2020 年有人把这做成可算的题：拿近半个世纪新冒出来的名词，问哪种模型能预测它挑哪个量词。传统讲法是范畴围着一个典型成员转，新成员按「离中心多近」入座；赢的却是另一种——新词跟着跟它意思最像的那几个老词走，范畴是一节一节接出去的，不是围着中心摊开的。孩子那边也对得上：两岁半左右架子就搭对了，具体哪个名词配哪个量词到六岁以后还在调，想不起来就拿通用的「个 gè」顶上——正如「孩子说错的地方最有信息」那期所说，默认件被过度使用恰恰说明系统装上了。

顺带说一个常搭在量词上的大结论：有人主张量词语言的使用者更倾向按材质归类，非量词语言的使用者更倾向按形状。2009 年一组六个实验去查，五个复制出了这个效应；唯独第六个——唯一不需要你解释名词意思的任务——什么都没有。所以它更像「你怎么理解一个词」，不是「你怎么看世界」。

EN English says "three fish"; Mandarin requires a classifier — 三条鱼 (sān tiáo yú, "three CL fish"). The choice is not free: 条 (tiáo) covers long flexible things (fish, roads, snakes, rope, trousers), 张 (zhāng) covers spreadable flat surfaces (paper, tables, beds, mouths), 把 (bǎ) covers things with a graspable handle (umbrellas, knives, chairs, keys). You commit to a shape before you can state a number. This is also unremarkable cross-linguistically: in a WALS sample of 400 languages, 260 lack numeral classifiers, 62 have them optionally and 78 obligatorily, with dense clusters in East and Southeast Asia, Mesoamerica, Amazonia and West Africa. The categories start geometric and then leave geometry: 条 (tiáo) also counts news items and text messages, and Japanese 本 (hon), originally long thin objects, also counts telephone calls, judo bouts and baseball hits. Habibi, Kemp and Xu (2020) tested this computationally on nouns that emerged over the past half century: a weighted-exemplar model beat the prototype account, so categories grow link by link toward their nearest neighbours rather than outward from a centre. Children match this — corpus work finds the syntactic frame adult-like around 2;06 while the semantic assignments are still being tuned past six, with the general classifier 个 (gè) standing in meanwhile. On the larger claim that classifier languages make their speakers sort by material rather than shape: Li, Dunham and Carey (2009) replicated the effect in five experiments and extended it to Mandarin, but their sixth, the only one not requiring interpretation of a noun, found nothing — the effect concerns how you read a word, not how you see the world.

「那场火，幸亏消防队来得快」——句首那个东西不是主语

这句汉语完全正常，拆开看却很怪：句首的「那场火」跟后面唯一的动词「来」没有任何关系。火不来，消防队才来。它既不是「来」的施事也不是受事，甚至不是时间地点，只是被摆在最前面，宣布这段话是关于什么的。语言学管这种开头叫话题。

英语没法这样摆，你得给它找个挂靠：As for that fire, ... 或者塞进从句。汉语不需要挂靠，位置本身就是身份：句首＝话题。1976 年有人据此把语言分成主语优先的（英语、法语）和话题优先的（汉语、日语、韩语）。分界线不是「有没有主语」，是「句子首先围着谁组织」。

还有两个能顺手验证的地方。一个是「下雨了」：英语必须放一个不指任何东西的 it（It's raining），汉语连这个占位的都不需要，它根本没有那个必须填的格子。另一个是「你看见他了吗？——看见了」：主语和宾语一起省掉，全靠上下文捡回来，日语同样如此。

但这一样还是不算孤例。值得单拎出来的是个小差别：日语和韩语给话题配了专门的标记（日语的 は wa、韩语的 는 neun），明确告诉你「这是话题」；汉语什么标记都不给，只靠位置和停顿。所以汉语的话题要听出来——这也是它最容易在跨语言场合出岔子的地方。

EN 那场火，幸亏消防队来得快 (nà chǎng huǒ, xìngkuī xiāofángduì lái de kuài, "that fire — luckily the fire brigade came fast") is ordinary Mandarin, yet the opening phrase has no relation to the only verb: the fire does not come, the brigade does. It is neither agent nor patient nor setting; it is simply placed first to announce what the sentence is about. Li and Thompson (1976) called this a dangling topic and split languages into subject-prominent (English, French) and topic-prominent (Mandarin, Japanese, Korean). Two further probes: 下雨了 (xià yǔ le, "raining") needs no dummy subject where English requires a referentless "it"; and 你看见他了吗？——看见了 (nǐ kànjiàn tā le ma? — kànjiàn le, "did you see him? — saw") drops subject and object together, recovered from context, as Japanese also does. What is more particular is a small difference: Japanese and Korean mark topics overtly with a particle, while Mandarin marks them by position and pause alone, so the topic has to be heard rather than read off a marker.

拼音让你念出看不懂的东西，汉字让你看懂念不出的东西

先拆掉一个误会：汉字基本不是画出来的。2003 年一项研究把小学阶段教的 2570 个字全过了一遍，约 72% 是形声字——一半提示意思，一半提示读音。妈 mā 左边的女（nǚ，"女性"）管意思，右边的马（mǎ）管读音；铜 tóng 左边是金（jīn，"金属"），右边是同（tóng）。这套系统一开始就是半表音的。

问题是音在变，字不变。江 jiāng 的声旁是工（gōng），河 hé 的声旁是可（kě），两千多年前大概读得很近，现在对不上了。同一项研究还发现一个反直觉的安排：一二年级教的字笔画最少，恰恰也是声旁最不可靠、表意最不透明的那批——最常用的字往往最老，磨得最久。孩子最先遇到的是规律最少的部分。

代价可以直接算。2500 个常用字覆盖现代文本的 97.97%，再加 1000 个次常用字到 99.48%。听着很快封顶，可差的这不到 2% 恰恰是代价——缺了它们剩下 98% 也读不顺，而它们只能一个一个认。字母系统那边学完二十几个字母就能把没见过的词念出来，念得出不代表懂。

红利正好是镜像：你能看懂你念不出的东西。「饕餮（tāotiè）」很多人从偏旁和上下文认得出是个吃相凶猛的怪兽，未必读得准；英语反过来，otorhinolaryngology 可以照字母念完，念完还是不知道那是耳鼻喉科学。顺带说一笔常记错的账：「电脑、手机」好懂是构词的功劳（电＋脑、手＋机），不是字形的，写成 diànnǎo、shǒujī 一样透明。

那么真正少见的是什么？人类独立发明文字大概只有四次：两河流域、埃及、中国、中美洲。四套里走「一个符号对应一个语素」这条路、今天还在日常大规模使用的只剩这一套（以及从它借出去的日文汉字）。声调、量词、话题优先都能在别处找到同伴，这一条找不到。

EN Chinese characters are not mostly pictures. Shu and colleagues (2003) went through all 2,570 characters taught in Chinese elementary school and found about 72% are semantic-phonetic compounds: 妈 (mā, "mother") is 女 (nǚ, "woman") for meaning plus 马 (mǎ, "horse") for sound; 铜 (tóng, "copper") is 金 (jīn, "metal") plus 同 (tóng, "same"). The system was half phonetic from the start — but sound changed and the graphs did not. 江 (jiāng, "river") carries 工 (gōng, "work") as its phonetic; 河 (hé, "river") carries 可 (kě). The same study found a counterintuitive arrangement: the characters taught in first and second grade have the fewest strokes and are also the least phonetically regular and least semantically transparent, because the commonest characters are usually the oldest and the most eroded. The cost is countable: 2,500 frequent characters cover 97.97% of modern text and another thousand bring it to 99.48%, and that last two percent is exactly the price, since each remaining character must be learned one at a time. An alphabet lets you pronounce a word you have never met without understanding it; a character script lets you understand one you cannot pronounce — 饕餮 (tāotiè, a ravenous mythical beast) is readable from its components long before it is sayable, whereas "otorhinolaryngology" is sayable long before it is meaningful. One credit often given to the script belongs elsewhere: 电脑 (diànnǎo, "electric brain", computer) and 手机 (shǒujī, "hand machine", mobile phone) are transparent because of compounding, not because of the graphs, and would be equally transparent in pinyin. What is genuinely rare is the script itself: writing was invented independently perhaps four times, and of those, the morpheme-based line is the only one still in large-scale daily use.