Words Are Assembled

You can pluralize a word that does not exist

First, a name for the parts. A morpheme is the smallest piece that still carries meaning. Unhappiness splits into three: un (negation), happy, and ness (the part that turns an adjective into a noun). Below that, nothing is left to split — the u in un means nothing at all.

The test for splitting is not "looks like a piece." In cranberry, the berry part is obviously a morpheme; what about cran? It has no independent meaning and appears in no other word. These are common enough that linguistics named them: cranberry morphemes. Their existence tells you what the criterion really is — a chunk counts as a part if it can be reused somewhere else.

Now the experiment. In 1958 Jean Berko drew a small blue creature, told children aged four to seven "this is a wug," turned the page to show two of them, and read the fragment "Now there are two ___" for the child to complete. The children said wugs. That drawing became the most reproduced picture in the field.

The trick is that wug does not exist: no dictionary, no parent, no exposure. If a child's head held a list of words, they would stall. They did not stall, which means what they hold is a rule that runs on any stem at all.

They also picked the right ending. The three pronunciations divide the work precisely: after a voiceless consonant you get the s sound (cats), after a voiced one you get z (dogs, wugs), and after hissing sounds like s, z or sh you add an entire syllable (glasses). Nobody states this division out loud, and four-year-olds obey it.

The unevenness carries the most information. Children are near ceiling on the first two branches and collapse on the one that needs an extra syllable: for Berko's invented word gutch (plural gutches), only 28% of four- and five-year-olds produced it, and only 38% of the five-and-a-half to seven group — while the same children were at 91% on the real word glasses. So what a child learns is not "the plural rule" but several sub-rules covering different territory, the frequent ones first and the rare ones years later.

Evidence strength: a single classic experiment, but replicated endlessly. When the design is exported to other languages, one precondition is easy to miss — the language must have an obligatory inflectional category to test in the first place. Japanese nouns do not obligatorily mark plurality, so the same materials transplanted directly leave nothing to measure. That is not a fact about Japanese; it is a limit on the ruler.

One kind of assembly makes new words, the other just changes clothes

Put teacher and teachers side by side. Teach plus er built a new word and even changed its class (verb to noun); it earns a dictionary entry. The s in teachers built nothing — it is the same word wearing different clothes for a different slot in the sentence, and no dictionary lists it separately.

The first is derivation, the second inflection. Telling them apart is not a matter of feel; there are four checkable properties.

First, whether identity changes. Happy to happiness changes word class, so it is derivation; happy to happier stays an adjective, so it is inflection. Second, order. Teach-er-s works, teach-s-er does not — inflection always sits on the outermost layer. Third, holes. Inflection has almost none: every English verb has a past tense, including one you invented a minute ago. Derivation is full of them: arrive gives arrival, but not arrivement, even though ment is plainly alive (government, movement); and th (warm to warmth, wide to width) retired long ago. Fourth, whether the meaning is predictable. Teachers is always "more than one teacher," without a single exception; transmission can mean sending something, or the gearbox in your car — a derived word grows its own meaning and then keeps it.

How these two kinds of assembly get packaged varies enormously across languages. The Turkish evlerimden on the left of the figure is a single word, one bead per meaning, in a fixed order, like stringing a necklace — that is agglutinative packaging. The Latin amō in the middle goes the other way: the ō at the end does four jobs at once, and you cannot slice it into a person part plus a tense part. That is fusional.

Inflection has one further property that matters far more than whether a language has it: you cannot opt out. In English, the moment you mention a countable noun you are forced to choose between singular and plural, even when the information is useless here, even when you do not know how many there are. Jakobson put it best in 1959: languages differ essentially in what they must convey, not in what they may convey. The next section is built entirely on that sentence. And to be clear: agglutinative, fusional and barely-packaged are not rungs on a ladder. All three convey the same content; they just put the complexity in different places — inside the word, or out in the word order and the function words.

When people say Mandarin has no morphology, which half is right

You have probably heard the claim. Half of it holds, and the half that gets dropped is the more interesting one.

The half that holds: inflection really is sparse. In 我昨天去 (wǒ zuótiān qù, "I went yesterday") and 他明天去 (tā míngtiān qù, "he goes tomorrow"), the verb 去 (qù, "go") is simply 去 (qù), unchanged. Nouns are not obliged to mark number either: the 书 (shū, "book") in 一本书 (yī běn shū, "one book") and in 三本书 (sān běn shū, "three books") is identical. The "cannot opt out" from the last section can be opted out of here — a Mandarin speaker can go a whole sentence without revealing how many, and an English speaker cannot.

But sparse is not absent. Mandarin has 们 (-men), as in 学生们 (xuéshengmen, "students") and 朋友们 (péngyoumen, "friends"). It comes with hard limits: it attaches essentially only to nouns denoting people, and it cannot combine with a numeral-classifier phrase — 三个学生们 (sān ge xuéshengmen, "three students-PL") strikes most native speakers as off. It is not obligatory inflection but a conditional, optional marker.

The half that gets dropped: word-building is extremely active. Compounding does most of the work — 电 (diàn, "electricity") plus 脑 (nǎo, "brain") gives 电脑 (diànnǎo, "computer"); 手 (shǒu, "hand") plus 机 (jī, "machine") gives 手机 (shǒujī, "mobile phone"). There is a set of half-bleached parts too: 性 (-xìng, roughly "-ity"), 化 (-huà, roughly "-ize"), 员 (-yuán, "staff member"). And there is reduplication — 看 (kàn, "look") becomes 看看 (kànkan, "have a quick look"), lighter and briefer. All of this is words being assembled; it just happens on the word-building layer rather than the inflectional one.

The category Mandarin genuinely will not let you skip sits elsewhere: the classifier. When you count things you cannot escape it. You cannot say 三书 (sān shū); you must say 三本书 (sān běn shū, "three books"), and you must pick the right one from a large set — 本 (běn, for bound volumes), 张 (zhāng, for flat things), 条 (tiáo, for long thin things), 只 (zhī, for many animals). This is exactly the "forces you to be explicit" from the last section, landing in a different place: English forces you to declare quantity, Mandarin forces you to declare shape or kind.

So Mandarin-speaking children have their own hurdle. They do not learn plural agreement; they learn classifiers, and the observed path looks like the wug story in reverse: first the general 个 (gè, the all-purpose classifier) is used for everything (三个鸟 sān ge niǎo, "three birds"; 三个纸 sān ge zhǐ, "three papers"), then it slowly differentiates. Studies report roughly a third of children using nothing but 个 (gè) throughout a production task, with individual classifiers still incomplete at ten and a dip-then-recovery U-shape along the way. This is the same thing as an English child saying goed — grab the cheapest general rule first, patch in the exceptions later.

Two things should be flagged honestly. First, whether 了 (le), 着 (zhe) and 过 (guò) count as affixes has been argued for a long time — they behave like inflection (fixed position after the verb) and like detachable function words, with no settled answer. Second, whether the unit "word" is even clear in Chinese is also contested: early small studies (Hoosain 1992, and the manual annotation comparisons Sproat and colleagues ran in 1996 while building a segmenter) reported substantial disagreement among native speakers about where words end; a large crowdsourced study in 2017 obtained proportionate agreement of 0.91 to 0.93 and concluded the opposite. Evidence strength: the later study has a far bigger sample, but its metric is per-position proportionate agreement, and since the vast majority of positions are obviously not boundaries, that number runs high by construction.

So the accurate version of the claim is: Mandarin has very little inflectional morphology. Information can be encoded in the shape of a word (that English s) or somewhere else entirely (Mandarin's classifiers, word order, time adverbs). That is a division of labour, not a ranking — every language forces you to be explicit about something, just in a different place.