> Is this really compositionality?
Usually it would be called "noncompositionality", the phenomenon that distinguishes words from phrases or sentences. As you might expect from that term, it's the opposite of compositionality. It also goes under the fancier name "the arbitrariness of the sign" or the even fancier "l'arbitraire du signe".
For example, hat, sat, bat, and flat are all semantically unlike each other, and flare, phlegm, flick, and flood are also all semantically unlike each other. It isn't the case that you can predict the meaning of flat by knowing what hat, sat, bat, flare, phlegm, flick, and flood mean.
The article draws an analogy to a general syntactic rule:
> But, so far, only “trivial compositionality” has been identified in non-human animals, whereby each unit adds independently to the meaning of the whole. For example, the phrase “blonde dancer” has two independent units: a blonde person who is also a dancer. Humans were thought to be unique in also having “non-trivial compositionality”, where the words in a combination means something different to what they mean individually. For example, the phrase “bad dancer” doesn’t mean a bad person who also dances.
Bad dancer is compositional in that you can know the meaning of the phrase by knowing the meanings of bad and dancer. So far so good; the article agrees that it is compositional.
They observe that the two words in the phrase aren't independent; a one-legged dancer is a dancer who has one leg, but a bad dancer isn't a dancer who is bad in general; it's a person who is bad at dancing.
The reason this is still compositional is that you can use any appropriate adjective this way: you can be a strong dancer, an energetic dancer, a timid dancer... so this is just a rule about the formation of certain English phrases, and knowing what bad means in general is sufficient to know what bad means in the phrase bad dancer.
They don't appear to be making a similar claim for the bonobos:
> For example, “high-hoot + low-hoot” combines the calls that seem to mean “pay attention to me” and “I am excited” to say “pay attention to me because I am in distress”
This example looks more like an arbitrary sign to me. But that might be an artifact of poor data management:
> they used a technique from linguistics to create a cloud of utterance types, placing vocalisations that occurred in similar circumstances closer together. “We kind of established this dictionary,” says Berthlet. “We have one vocalisation and one meaning.”
There is no reason to expect that one vocalisation should possess only one meaning.
They also have poor methodology:
> Once they had this semantic cloud, they could see whether the individual calls in a combination had distinct meanings, and found that the combinations were close to the units that they were made of, which would suggest compositionality. Using this approach, they identified four compositional calls
By comparison, throw, throw up, and throw away are all similar in meaning, which is why they have similar forms, but they are not compositional - knowing what one of them means won't help you with the others.