Humans can effortlessly talk about an infinite number of topics, from neuroscience to pink elephants, by combining words into sentences. This is thanks to compositionality : the ability to combine meaningful units into larger structures whose meaning is derived from the meaning of its units and the way they are combined.
For years, scientists believed that only humans extensively used compositionality. Animal communication was thought to be mostly a mere random assortment of calls, with only rare instances of compositionality. However, our new study , recently published in the journal Science, says otherwise.
By extensively researching the vocal communication of bonobos in their natural habitat, the Kokolopori Community Reserve in the Democratic Republic of the Congo, we found that vocal communication between bonobos – our closest living relatives, along with chimpanzees – relies extensively on compositionality, just like human language.
A BONOBO DICTIONARY Investigating compositionality in animals first requires a strong understanding of what single calls and their combinations mean. This has long presented a challenge, since accessing the minds of animals and reliably decoding the meaning of their calls is difficult.
To remedy this, we developed a new way of reliably determining the meaning of bonobo vocalisations, and used it to determine the meaning of all of their single calls and combinations.
We assumed that a bonobo call can have different types of meaning. It can give an order (“Run”), announce future actions (“I will travel”), express the internal states (“I am afraid”) or refer to external events (“There is a predator”).
To reliably understand the meaning of each vocalisation while avoiding human bias, we described in great detail the context of emission of each vocalisation, using more than 300 contextual parameters. For example, we recorded the presence of external events (such as the presence of another bonobo or monkey group nearby) and the behaviour of the caller, like whether it was feeding, travelling, resting, and so on.
We also coded, for the two minutes after the production of a call, what the caller and the audience started doing, kept doing, and stopped doing. We used this highly detailed description of the context to attribute meaning to the calls , with the meaning being the contextual parameters associated with the emission of this vocalisation. For example, if the caller always starts travelling after a particular call is emitted, then the call likely means “I will travel”.
With this approach, we were able to create a complete list of bonobo calls and their associated meaning: a bonobo dictionary of sorts. This dictionary represents an important step in understanding animal communication, as it is the first time researchers have systematically determined the meaning of all the calls of an animal.
BONOBO COMPOSITIONALITY In the second step of our study, we developed a method to investigate whether animal combinations are compositional. We found numerous call combinations whose meaning was related to the meaning of their parts – a key hallmark of compositionality. Furthermore, some of these call combinations bore a striking resemblance to the more complex compositional structures in human language.
In human language, compositionality can take two forms. In its simple (or trivial) version, each element of the combination contributes to the meaning of the whole independently, and the combination is interpreted by the sum of its parts. For example, “blond dancer” refers to a person who is both blond and a dancer. If this person is also a doctor, we can infer that they are a blond doctor as well.
In complex (or nontrivial) syntax, the units in a combination do not contribute independent meanings, but instead interact so that one part of the combination modifies the other. For example, “bad dancer” does not refer to a bad person who is also a dancer. Indeed, if this person is also a doctor, we cannot infer that they are a bad doctor. Here, “bad” is linked to “dancer” only.
Previous studies in birds and primates have demonstrated that animals can form trivially compositional structures. However, there has previously been little clear evidence of nontrivial compositionality in animals, reinforcing the idea that this ability is uniquely human.
To determine whether bonobo calls were compositional, we borrowed an approach from linguistics that states that, in order to be considered compositional, a combination has to meet three criteria:
Each of its elements have different meanings.
The meaning of the combination is different from the meaning of its elements.
The meaning of the combination is derived from the meaning of its elements.
Additionally, we assessed whether a compositional combination was nontrivial by determining whether its meaning was more than the sum of the meaning of its parts. To do this, we built a semantic space – a multidimensional representation of bonobo call meanings – which allowed us to measure the relationships between the meaning of single calls and combinations.
We used a method derived from distributional semantics , a linguistic approach that maps words based on their meaning similarities, with the idea that words with close meanings are used in similar contexts. For example, the words “shark” and “animal” are often used alongside similar words, such as “fish” and “predator”, suggesting that they have related meanings. In contrast, “animal” and “bank” are used in different contexts, they have less related meanings. This approach allows to reliably represent and measure the relationship between different words’ meanings.
By applying this methodology to bonobo vocalisations, we mapped the meaning of calls and call combinations within a semantic space based on their context of use. This ultimately allowed us to determine which combinations met the three criteria for compositionality and, in addition, whether they displayed nontrivial compositionality.
We identified four call combinations whose meaning was related to the meaning of their single parts, a key hallmark of compositionality. Importantly, every call type appeared in at least one compositional combination, similarly to how every word can occur in a phrase in human language. This suggests that, like in human language, compositionality is a fundamental feature of bonobo communication.
Furthermore, three of the call combinations bore a striking resemblance to the more complex nontrivial compositional structures seen in human language. This suggests that the capacity to combine call types in complex ways is not as unique to humans as we once thought, hinting that this capacity may have deeper evolutionary roots than previously assumed.
A bonobo emits a subtle peep, meaning ‘I would like to…’, before the whistle, which means ‘Let’s stay together’. In tense social situations, this combination means something like ‘Let’s relax’
EVOLUTION OF LANGUAGE An important implication of this research is the insight it provides into the evolutionary roots of language’s compositional nature. If our bonobo cousins rely extensively on compositionality, just like we do, then our last common ancestor likely did as well.
This suggests that the ability to construct complex meanings from smaller vocal units was already present in our ancestors at least 7 million years ago, if not earlier. These new findings indicate that, far from being unique to human language, compositionality likely existed long before humans did.
NOTE – This article was originally published in ndtv and can be viewed here
