I used to think AI would one day "master" Japanese.
Now I think it will never master it.
Not because it’s too hard.
Because it’s too human.
The beauty of Japanese isn’t in its kanji.
It’s in the spaces between.
The pause after ね.
The breath before か.
The way a sentence ends with ね and leaves the listener hanging—not because it’s incomplete, but because it’s shared.
An AI can’t replicate that.
It can generate the hiragana. It can count the い. It can even generate a sentence that sounds right.
But it can’t know why we say 見てね instead of 見てください.
It can’t know that the former is a mother’s whisper.
The latter is a bureaucrat’s demand.
And that’s the difference.
Not between languages.
Between humans and machines.
So can AI count hiragana in kanji readings?
Sure.
But should it?
Maybe not.
Maybe the real question isn’t whether AI can.
It’s whether we want it to.
Because if we do… we’re not asking for understanding.
We’re asking for imitation.
And imitation… is the loneliest kind of intelligence.
I’ve spent years studying how people use AI tools.
I’ve watched them rely on it to write emails.
To summarize meetings.
To translate Japanese.
And every time, they say the same thing: "It’s good enough."
But "good enough" isn’t understanding.
It’s compromise.
And compromise… is what we do when we stop asking the right questions.
So next time you ask an AI to count hiragana in a kanji reading… pause.
Ask yourself:
Why am I asking this?
And what am I really trying to learn?
Because the answer isn’t in the letters.
It’s in the silence.
Why Okurigana Breaks AI’s Assumptions
Let’s be clear: AI doesn’t "fail" at counting hiragana in kanji readings.
It doesn’t even try.
Because the question itself is a trap.
You see, Japanese doesn’t have "kanji readings" the way English has "word pronunciations."
There’s no dictionary entry for 見 that says "mi" or "ken."
Instead, the reading changes based on context, mood, and relationship.
That’s where okurigana comes in.
Okurigana are the hiragana that follow kanji—like the て in 見て, or the ます in 行きます.
They’re not decoration.
They’re grammar.
They tell you tense.
They tell you politeness.
They tell you whether you’re talking to your boss, your child, or your dog.
AI treats these like variables in a formula.
"If kanji = 見, then possible readings = mi, ken, miru, mite, etc."
But that’s not how Japanese works.
A native speaker doesn’t pick from a list.
They feel the weight of the sentence.
They sense the social gravity.
They don’t count hiragana.
They live them.
I once asked a Japanese colleague to explain why 見てね is different from 見てください.
She didn’t give me a linguistic breakdown.
She said: "One is a request. The other is a command."
Then she smiled.
"And one leaves room for the other person to say no."
That’s the moment AI dies.
Because AI can’t understand "leaving room."
It can’t understand silence as a form of respect.
It can’t understand that the difference between てね and てください isn’t in the letters.
It’s in the space between the speaker and the listener.
And that space? It’s not data.
It’s trust.
And trust… isn’t something you count.
It’s something you earn.
The Myth of the "Correct" Reading
I’ve seen AI-generated Japanese text that gets every kanji reading "right."
It’s terrifying.
Because it sounds perfect.
And it’s utterly, soullessly wrong.
A Japanese speaker reads 見る as "miru" in isolation.
But in a sentence like "今、見てね," the reading is still "mi," but the meaning isn’t "see."
It’s "pay attention."
It’s "don’t miss this."
It’s "I’m trusting you to notice."
The AI doesn’t know that.
It doesn’t know that 見てね is the most common phrase in Japanese parenting.
That it’s used to guide, not control.
That it’s the verbal equivalent of a hand on the shoulder.
I’ve watched Japanese mothers say 見てね a hundred times a day.
To toddlers.
To teenagers.
To husbands.
Always with the same soft tone.
Always with the same quiet expectation.
Never with the weight of obligation.
That’s not a grammar rule.
That’s a cultural algorithm.
And AI can’t learn cultural algorithms.
It can mimic them.
It can generate 10,000 variations of 見てね.
But it can’t feel the difference between one said by a tired mother at 10 p.m. and one said by a teacher at 8 a.m.
It can’t know that the same word, in the same context, with the same hiragana, can carry the weight of love… or the weight of disappointment.
I once saw a Japanese student cry because an AI translation turned her grandmother’s 見てね into "Look!"
"It sounded angry," she said.
"But it wasn’t."
"It was just… her."
And that’s the tragedy.
Not that AI gets it wrong.
That it gets it "right"—and still misses everything.
Because the "correct" reading isn’t a linguistic fact.
It’s a human moment.
And moments… aren’t countable.
The Cost of Mistaking AI for Understanding
I’ve worked with teams that use AI to translate Japanese customer service logs.
They think they’re saving time.
They’re not.
They’re creating a new kind of noise.
Because AI doesn’t understand the difference between 見てください and 見てね.
So it translates both as "Please look."
And then the customer gets a robotic reply.
"We have noted your request to look at the issue."
No one says that in Japanese.
No one wants to say that.
But AI doesn’t know that.
It doesn’t know that Japanese culture values implied meaning.
That "no" is often whispered.
That "yes" is sometimes silent.
That the most important part of a conversation isn’t the words.
It’s the pause after them.
I’ve seen companies lose customers because their AI chatbot responded to 見てね with a form letter.
"Thank you for your feedback. We will review your request."
The customer wrote back: "I didn’t ask you to review anything. I asked my mother to watch my child."
And then they left.
Not because the AI was wrong.
Because it was precise.
And precision… without empathy… is cruelty.
We’re not training AI to understand Japanese.
We’re training it to mimic the surface.
And that’s dangerous.
Because when you rely on AI to handle human communication… you stop learning how to do it yourself.
You stop listening.
You stop noticing.
You stop feeling the weight of a single hiragana.
And then… you become the machine.
That’s the real cost.
Not that AI can’t count hiragana.
That we’re letting it replace the part of us that should be doing the counting.
The part that remembers that language isn’t about accuracy.
It’s about connection.
What We’re Really Counting
So can AI count hiragana in kanji readings?
Technically? Yes.
Practically? No.
Because the question isn’t about counting.
It’s about control.
We want AI to count hiragana because we want to reduce Japanese to a set of rules.
We want to codify it.
To systematize it.
To make it predictable.
But Japanese isn’t meant to be predictable.
It’s meant to be alive.
It breathes.
It shifts.
It adapts.
It leaves space.
And the hiragana? They’re not letters.
They’re breaths.
They’re pauses.
They’re the sound of someone holding back.
The sound of someone choosing kindness over clarity.
The sound of someone saying "I’m here" without saying it.
AI can’t replicate that.
Because AI doesn’t have a heart.
It doesn’t have a mother.
It doesn’t have a memory of a voice saying 見てね at bedtime.
And maybe… that’s okay.
Maybe we don’t need AI to count hiragana.
Maybe we need AI to stop trying.
Maybe the real work isn’t in teaching machines to understand Japanese.
It’s in reminding humans why we still speak it.
Because language isn’t a problem to be solved.
It’s a gift to be held.
And some gifts… are meant to be kept sacred.
So next time you ask an AI to count hiragana in a kanji reading… don’t.
Just listen.
And remember:
The most important characters aren’t written.
They’re felt.