Can AI Translate Cat Meows? What Meow-Decoding Research Really Shows

Apps that listen to a meow and display I am hungry or Play with me have spread widely, and in 2025 news reports announced an AI that classifies 40 types of meow with 95% accuracy. Has cat language finally been decoded? This article checks, from peer-reviewed experiments, what humans and machines can each read from a cat's voice.

How well can people identify the situation behind a meow?

Nicastro and Owren (2003) recorded meows from 12 cats in five behavioral contexts and ran two experiments testing whether human listeners could classify which context each came from. Accuracy was significantly above chance, but only modestly. Bouts of calls were easier to classify than single calls, and listeners who had lived with cats and liked them did better.

The authors concluded that the meow is not a word that refers to a specific object but a general-purpose, somewhat negatively toned signal for attracting human attention. They also noted that with experience, listeners get better at picking out calls that signal a positive state.

A 225-person experiment: fewer than half chose the right context

Prato-Previde and colleagues (2020) played 225 adults six meows recorded from ten Maine Coon cats in three situations, waiting for food, isolation, and brushing, and asked them to name the context and the emotional direction, positive or negative.

Fewer than half could choose the correct context. The easiest was the waiting-for-food meow, at about 40% for the most representative recording. Cat owners, women, and people with higher empathy toward animals did somewhat better on the isolation and brushing meows. Even when the context was missed, emotional direction came through fairly well: the isolation meow was rated the most negative and the brushing meow the most positive.

So people are poor at reading what a cat wants from its voice but reasonably good at reading whether it is in a good or bad mood. That is the picture the experiments give.

AI can classify contexts, but that is not translation

Ntalampiras and colleagues (2019), from the same research group, ran a proof of concept in which machine learning automatically classified meows from Maine Coon and European Shorthair cats recorded in the same three contexts. They extracted mel-frequency cepstral coefficients and temporal modulation features and used a classifier that splits the problem in stages; it outperformed the alternative methods they compared.

What this shows is that acoustic features of meows contain context-specific differences that a machine can pick up. But the target is which of three contexts, not a sentence expressing what the cat wants to say. The 40 types and 95% figures in news reports are likewise classification accuracies against context labels defined by researchers, not translation accuracy for cat language.

How to read what a translator app shows

Commercial translator apps build on this kind of classification, assigning a meow to one of several to a dozen or so labels such as hungry or affectionate. Some are reported to have analyzed meows in the hundreds of millions, but the correctness of the labels depends on owners' self-reports and has not been validated in peer-reviewed research.

In practice, read the app's output as an estimated context and mood. Recording the situation and body signals that accompany each meow and comparing them with the app's estimate can help you learn your own cat's patterns.

Meows are tuned for humans

That cats meow at people is well established. Adult cats rarely meow to each other; the sound is used mainly in interactions with humans. The possibility that cats tune pitch and length for people, and how they respond to human speech directed at them, are covered in How much do cats understand people? and Does cat-directed speech get through?.

Cats communicate a great deal through gaze and facial movement as well. Facial expression research is summarized in Do cats have 276 facial expressions?, and how cats use gaze when stuck in Do cats ask humans for help?. For the full picture of cat cognition research, see the cat IQ research guide.

Limits to keep in mind

All of the experiments described used a small number of breeds and voices recorded in a limited set of contexts. Individual and breed differences and the variety of interactions in real homes are not fully represented. There is also a fundamental difficulty: before talking about translation accuracy, no established method exists for objectively confirming what a cat was trying to convey.

The free NekoIQ IQ test, which measures human reasoning, has 20 questions and takes about 10 minutes. It does not assess your ability to communicate with cats, and the results are estimates.