Pet translator apps translate the wrong thing
The problem with pet translator apps is not that the accuracy is poor. It is that a meow is a request aimed at you, not a report of an internal state — so the category is pointed at the wrong signal.
There are a lot of apps that claim to translate what your cat is saying. The category is enormous — the biggest one on the Chinese App Store has over 240,000 ratings. The reviews are a bloodbath. People feel cheated, and the usual explanation is that the accuracy just isn't there yet.
I think the accuracy is beside the point. The category is pointed at the wrong signal.
Cats don't meow at cats
Adult cats barely vocalise at each other. They hiss and yowl in conflict, and queens call to kittens, but the ordinary conversational meow is something cats developed for humans. Kittens meow at their mothers; most stop as they grow. House cats keep going, aimed at us, because it works.
That means a meow is not an expression of internal state. It is a request directed at a specific audience. "Open the door." "The bowl looks empty from where I'm standing." "You walked past me."
So when an app records a meow and returns "I am happy," it has performed a category error before the model even runs. It is reading a message addressed to you and reporting it as a feeling.
Dogs are the same story in a different key. A bark is contextual and highly audience-dependent. The dog barking at the delivery driver and the dog barking at you holding the leash are not in the same state, and the audio alone will not separate them reliably.
The state is in the body
Posture is the opposite kind of signal. It is continuous rather than event-based, it is not aimed at anyone, and much of it is not under voluntary control. An animal cannot decide to un-flatten its ears because you walked in.
Four places carry most of it.
Ears. Direction, and whether they track sound or stay pinned. Flattened out sideways is a different message from pinned flat back.
Eyes. Pupil size relative to the light in the room, blink rate, and whether the whites show at the side.
Tail. Height, stiffness, and how much of it is actually moving. A wag says the animal is aroused, nothing more — a low loose sweep and a high stiff tail with a flicking tip are close to opposites.
Posture. Where the weight sits, and how the head is carried relative to the shoulders. Weight on the back legs means ready to leave.
None of these means anything alone. That is the part most "one weird trick" listicles get wrong. Read together they are fairly legible, and read against the same animal last week they are more legible still — which is the single biggest thing amateur readings miss. A cat that has always been quiet tells you nothing by being quiet. A cat that used to greet you at the door and now does not is telling you something specific.
Where this breaks down
Photos are a snapshot of a continuous signal, so they lose motion — and the tail and the weight shift are largely motion. Short video is meaningfully better than a still, and the gap is bigger than people expect.
Lighting wrecks pupil readings. Every "dilated pupils mean fear" chart is useless if you do not know how bright the room was.
And posture reading cannot separate mood from pain. A hunched back, paws held forward rather than tucked, head carried below the shoulder line — that is a pain posture, and it is not the beginning of a conversation about mood. It is a reason to call a vet. Any tool in this space that will not say that plainly is selling you something.
What we built
We make Hachimo, a browser tool that reads those four cues from a photo or a short video and returns a mood score, an emotion breakdown, and — the part we care about most — the specific cues it saw, so you can check its work rather than take its word.
It does not listen to anything. That is deliberate, for the reason above.
It also says when it cannot read an image. A dark photo or a cropped one gets "not enough to go on" instead of an invented mood, which sounds like a small thing until you notice that the apps people feel cheated by are the ones that always produce an answer.
We wrote up the individual signals as separate guides — what ears back actually means across four different situations, how a comfortable loaf differs from a pain crouch, what a slow blink is doing. They are free to read; you do not need the tool to use any of it.
If you take one thing from this: stop asking what the noise means, and start watching what the body is doing when it makes the noise.
Made with 💛 and many head scratches. Hachimo is for fun and curiosity — it's not veterinary advice. If your pet seems unwell, please talk to a real-life vet. 🩺