on language

what the chinese and japanese languages have taught me

One day at work, I had lunch with a coworker from China. This coworker had mentored me some time ago, but our relationship never developed much. She was always an anodyne presence, dispensing code reviews but never doing much else. I, a second-generation Chinese-American, was a little insecure about my Mandarin abilities, so we mostly spoke in English.

But I’d worked hard on my Mandarin since then, so I decided to strike up a conversation with her. We started to talk freely. We talked about our respective upbringings in China and the U.S. She spoke about the difficulties she experienced adjusting to life in America, and said that she always felt insecure about her English abilities, especially at work. Suddenly, I understood her.

This piece is about language: how it works mechanically, how it influences and is influenced by culture and society, and how it sheds light on the nature of intelligence. Although over half of the world’s population is multilingual, people in the Anglosphere aren’t usually interested in learning languages. It’s certainly not a practical hobby, since English is the global lingua franca. But I beg to differ. I’ll explain what I’ve learned while studying Mandarin and Japanese.

The Basics

I’ll start with basic information about how these languages work. First, let’s study Chinese. For the sake of convenience, this writeup will use Mandarin pronunciations.

For native English speakers like me, Chinese is rightfully considered one of the hardest languages you can possibly learn. But that doesn’t mean it’s devoid of its own internal logic. Yes, you do have to memorize thousands of 汉字 (hànzì), but owing to the logographic nature of Chinese, this is easier than it seems. Each 汉字 is comprised of one or more radicals, which hint at the meaning and/or the pronunciation of the entire character.

Consider an example which I learned not too long ago. I was in Guangzhou, China’s mecca for roast poultry. Everywhere around me were pictures of mouth-watering cuts of meat, next to signs that read 「烧鸭」，「烧鸽」， and 「烧鹅」. If you take a closer look at the second characters (the first one, 烧, means “roast”), you’ll notice that they look similar. They all share the radical 鸟, which means bird. So even if you don’t know what 鸭 means, you can at least know you’re ordering something with wings.

What are the radicals next to 鸟 doing? There’s three of them: 甲，合，and 我. 甲 is pronounced jiǎ. 鸭 is pronounced yā. Granted, it’s not a one-to-one match, especially since Chinese is a tonal language, but the similarity is obvious. 合 is hé. What is 鸽? It’s gē. At first glance, the third is a deviation, since 我 is wǒ, and 鹅 is é. However, there’s an archaic pronunciation for 我, still retained in characters like 俄 and 饿，that is–you guessed it–ě. It’s not a universal rule, but quite frequently in Chinese, one radical informs semantics, while the other informs phonetics. (Incidentally, 鸭 is duck, 鸽 is pigeon, and 鹅 is goose.)

A similar semantic trick can be applied to entire characters. For example, the character 电 (diàn) means electricity/electric, and 脑 (nǎo) means brain. What happens when you put them together? You get 电脑 (diàn nǎo): computer! An “electric brain”. This is especially important to how Chinese handles loanwords, which we’ll later examine.

Japanese is an interesting language, particularly when compared to Chinese. It has three writing systems: 漢字 (kanji), which are characters imported from China, 平仮名 (hiragana), a native script which covers words and particles which 漢字 doesn’t, and 片仮名 (katakana), another native script used for transcribing foreign loanwords–again, more on that soon. Unlike their Chinese counterparts, 漢字 usually have multiple pronunciations. There is usually one which corresponds to the Chinese pronunciation, and one or more which correspond to native pronunciations. For example, the word 私 can either be pronounced shi, a hybridized Sino-Japanese form, or watashi, the native form. When reading Japanese, one must infer, based on contextual clues, which pronunciation to use.

Another unique characteristic of Japanese is its complex system of honorifics. “Japanese people are polite” is a common stereotype, but it goes much deeper than most outsiders realize. There is an entire system known as 敬語 (keigo) which changes the very vocabulary and structure of the Japanese language based on how much respect one wants to convey. For example, the verb 行く　(iku), which means “to go [somewhere]”, is in its most casual form, as used around friends or family. Another step up in respectfulness, and it becomes 行きます (ikimasu), as used around strangers. Even higher: いらっしゃる　(irassharu), as used around superiors like your boss. Note that the character 行 disappears.

In Japanese, respect is also strongly correlated with indirectness. For example, consider how to tell someone they should do something. Consider telling someone that they must buy a phone. In English, you’d say something along the lines of “You must buy a phone”. In Japanese, you’d say 「スマホを買わなければいけません」 (sumaho wo kawanakereba ikemasen). This sentence is a double negative. Literally translated into English, it means “It’s bad if you don’t buy a phone.” Sure, it might seem like a needlessly roundabout way to convey the same information, but that’s typical Japanese parlance.

Loanwords

Now that we’ve covered the fundamentals of Chinese and Japanese, let’s see how they handle loanwords, both from Western languages and from each other. This is where things get interesting.

Due to the logographic nature of its script, Chinese often handles foreign vocabulary differently from other languages. As we demonstrated with 电脑, characters encode semantics, so by simply putting them together, Chinese can form new words with new meanings. Another case is 手机 (shǒu jī), which is “hand machine”. There are still transliterated loanwords, for example, 咖啡 (kā feī) for “coffee” and 夹克 (jiǎ kè) for “jacket”, but generally it is more convenient to form compound words, as literal transliterations can be quite long and unwieldy. For example, 德谟克拉西 (dé mó kè lā xī) for “democracy” has been largely replaced by 民主 (mínzhǔ), literally “people [as] rulers”.

民主 is an especially interesting word, with a very complex history. 民主 is a loanword from the Japanese word minshu, written with the same 漢字. But the Japanese 民主 was itself a Chinese loanword, albeit within a different context. The original Chinese 民主 was a contraction of the word 民之主 (mín zhī zhǔ). The middle character 之 contextualizes the word as more akin to “ruler of people”, which is the exact opposite of its modern meaning!

民主 is an example of a 和制汉字 (hézhì hànzì), or 和製漢語 (wasei-kango) in Japanese: a word built out of Chinese morphemes, yet whose meaning is uniquely Japanese. Many of these words re-entered Chinese during the early 20th century, when young intellectuals began to import foreign ideas. (A joke I’ve heard goes that the modern Chinese state owes its existence to the Japanese, since its formal name, 中华人民共和国, is allegedly entirely composed of wasei-kango.)

In contrast, there are a lot of transliterated loanwords in Japanese. They are known as 外来語　(gairaigo), or “foreign-origin words”. There are the expected ones, like コンピューター　(konpyūtā) for “computer”. Some are contractions of foreign words. For example, as mentioned above, スマホ (sumaho) is a contraction of “smartphone”. But there are also commonly-used loanwords that you wouldn’t expect. “News” is ニュース (nyūsu). “Rule” is ルール（rūru). It always struck me as odd, because the Japanese have native words for similar concepts, i.e. 新聞 (shinbun) and 規則 (kisoku), but for whatever reason, they’ve been supplanted in casual speech. And besides 民主, guess what’s the other commonly-accepted word for “democracy”? デモクラシー (demokurashī)!

But in the twenty-first century, dynamics are changing. With the advent of Japanese soft power in the form of anime and manga, Chinese people, particularly online and trend-savvy Gen Z youth, are increasingly adopting Japanese loanwords like 卡哇伊 (kǎ wā yī), i.e. 可愛い (kawaii), meaning “cute”. To my knowledge, there is no equivalent degree of linguistic drift in the opposite direction.

On Sapir-Whorf

The Sapir-Whorf hypothesis is the idea that language influences thought. It’s frequently cited in pop discourse about linguistics, and inspired the short story “Story of Your Life”, by Ted Chiang, one of my favorite works of fiction, in which a linguist develops powers of precognition after studying the language of an alien species.

First, I should establish some terminology. The “strong” hypothesis refers to the idea that language strictly determines thought. This is largely discredited. The “weak” hypothesis, on the other hand, argues that language simply nudges thought. This theory is more widely accepted. I can personally attest to this. I may not have developed precognition a la “Story of Your Life”, but as my proficiencies in Chinese and Japanese have advanced, I notice my inner monologue slowly adopting words and phrases in these languages, whereas it used to be solely in English. If I do well on a test, I think to myself, 「やった！」If I do bad, I think to myself: 「完蛋了！」

Japanese, in particular, has had a very interesting effect on me. In line with its system of honorifics, Japanese has a wide range of pronouns used to convey various levels of formality. Usually, my inner Japanese monologue refers to myself as 私 (watashi). When I’m feeling a little cockier, it uses 俺 (ore). The same phenomenon happens with other people. With the people I’m close with: 君 (kimi). With people I’m not close with: あなた (anata). With people I dislike: 貴様 (kisama). It’s clear that my inner monologue only crystallizes pre-existing strands of thought, but I wonder if this is a self-fulfilling prophecy. If I keep referring to someone as あなた, will it become harder for me to perceive myself as close to this person? According to weak Sapir-Whorf, most likely, yes.

It’s generally agreed that thought precedes language. We think of something, and then we put it into language. But is there such a thing as a thought that cannot be expressed in language? After all, in China and Japan, there are many social customs which regulate speech. The Japanese have the concepts of 建前 (tatemae) and 本音 (honne). The former describes one’s public-facing behavior and the latter describes one’s internal thoughts and emotions; the separation of the two is sacrosanct. Chinese has a similar concept: 面子 (miànzi), from which the English concept of “face” came from. It prioritizes group harmony over confrontation. I’ve always wondered how native speakers conceptualize this divide. Do their inner monologues crystallize what they cannot say with actual language, or is it expressed in an entirely different way? Or is it that after a while, they stop having these kinds of thoughts at all?

These phenomena aren’t entirely unique to the East. Even in English, I half-say things, say things I don’t mean, say different things to different people. Personally, when I think about things I cannot say, there is a liminal stage before they form into coherent sentences. During this stage they feel like something more primal. It is more like an emotion, but unlike an emotion, I cannot name it, because it is beyond words, and it will never need to be named.

On Artificial Intelligence

Here’s an interesting question. If someone grew up without learning any language, would they still have an inner monologue like mine? Can their system of thought itself be considered a language? This is the main idea of the language of thought hypothesis (LOTH). It’s a fairly controversial idea in cognitive science, but it has massive implications, especially within the context of artificial intelligence. First, to clarify: language within this context refers to formal language. Formal language, unlike natural languages like English, Chinese, or Japanese, is more akin to a logical system. If the LOTH is true, then this means that we can express cognition itself as a formal language, i.e. just a set of rules and symbols. Hence, we can theoretically model intelligence in any substrate.

Consider large language models, the core of modern AI research. The “language” in their name refers to natural language, but it doesn’t refer to formal language. In the beginning of AI as a field of study, research focused on the exact opposite: AI which operated with symbolic reasoning. However, symbolic AI never lived up to its promises, and researchers pivoted to machine learning, which gave rise to the LLMs of today. But if we assume the LOTH is true, then true machine intelligence must require a formal machine language too. Yet LLMs are predictive systems and do not use symbolic reasoning, so by definition they cannot exhibit intelligence.

But perhaps I might be proven wrong. Perhaps, in the near future, we will develop a post-LLM AI system with both natural language and formal language capabilities. That, in turn, means it might have something akin to 本音. When you think about it, that’s a very human thing to have. Just like you or me, a truly intelligent AI has thoughts it can’t put into words.

Closing Thoughts

So far, I’ve discussed language in fairly abstract terms. But I want to end this piece by returning to a more humanistic perspective. I want to return to the story of my coworker.

I benchmark my language abilities with my ability to understand others, and I don’t mean understanding raw syntax and diction. I mean understanding others like the way I understood her. I mean getting to know their hopes and dreams. I mean reading a verse of poetry, or hearing a line in a movie, and feeling true beauty. Sure, translation apps or subtitles can bridge the gap somewhat, but there’s something irreplaceably intimate about direct understanding. Language is, after all, a way for us to understand each other. And we, as humans, fundamentally want to understand others, and be understood in turn. By learning new languages, we become more human.