Researchers in Budapest ran 20 dogs and 20 people through the same listening test, and the dogs did something previously recorded only in humans. Their brains pulled individual words out of continuous speech by leaning on the consonants.
Vowels are loud and easy. Consonants are quiet but they carry the skeleton of a word, which is why human brains start prioritizing them in infancy. No other species tested had shown the same bias, including primates far closer to us than dogs. So the team asked whether this needs special human wiring, or whether any brain could pick it up from hearing a lot of talking. The ELTE release walks through the design; the paper is in Science.
The dogs heard artificial speech with no pauses between words. In the streams where consonants repeated to form word skeletons, both human and dog brains locked onto the rhythm. A distinct response showed up about 400 milliseconds after each word began, which is the timing already known to signal word segmentation.
The detail that surprised me: dogs who spent their first three months in a shelter or on the street showed the bias just as strongly as dogs raised in living rooms.
So they are not only reading your face. They are grabbing chunks out of the stream. That is a fair argument for saying one thing and meaning it, and for keeping cues short. “Sit” is a word. “Okay so what we are going to do now is sit” is a paragraph.
There is one limit, though. The dogs tracked syllable rhythm but not the individual speech sounds inside a syllable, so they work at a coarser grain than we do. They get the word. They do not get the phonetics.