Language is compositional: we systematically combine smaller language units—words into phrases, and phrases into sentences—to form more complex ideas. The meaning of each sentence depends on the meanings of its words and how we put them together. Consider the example, “John hugs Mary.” Each word has a specific meaning: “John” refers to one person, “Mary” to another, and “hugs” describes an interaction between them. The word order (Subject-Verb-Object) tells us that John is the hugger and Mary is the one being hugged. If we understand “John hugs Mary,” we can naturally understand “Mary hugs John” as well, since both sentences use the same words and structure. Switching the word order changes who hugs whom. One question scientists want to answer is when and how babies learn about these kinds of properties of language: What is the origin of linguistic compositionality? Does it come from learning language from scratch, or do babies mix-and-match ideas like this before they know a lot of words?
Children start putting meanings together early on. One classic study showed that 16-month-old babies learning English could use word order to understand sentences. For example, when they heard “Big Bird is tickling Cookie Monster,” they preferred to look at Big Bird acting upon Cookie Monster, but not the reverse. In a more recent study, 12-month-old babies learning Hungarian could combine new quantity words with familiar nouns. Researchers first taught them nonsense words for “one” (mize) and “two” (padu) while showing them pictures of one vs. two apples. After the short learning period, the babies could already apply these new quantity words to other objects they know. For example, they preferred to look at one duck when they heard mize duck, and two ducks when they heard padu duck. These findings show that even babies can combine words in their mind to understand complex meanings.
It turns out other animals might also understand some combinations of symbols that resemble those we use in language, at least after extensive training by humans. Recent research on baboons shows they can learn a symbol for “not” after extensive training. For example, given three shapes {A, B, C}, researchers taught baboons that shape A means “only A.” However, when they added a “not” symbol, it meant “not A” and picked out shapes B and C instead. Similarly, the baboons learned that shape B meant “only B,” but adding the “not” symbol changed it to mean shapes A and C. The baboons could even learn a similar rule for shape C, but they struggled much more to learn the reverse pattern: that shape C picked out shapes A and B, and “not C” picked out only C. Even though baboons have trouble using the “not” symbol in new contexts without more training, their ability to learn such abstract combinations suggests a basic capacity for compositionality.
Studies with young children and animals suggest that the ability to combine small pieces of meaning into larger ideas might have ancient roots. However, humans seem unique because we created complex symbolic systems, like language, to discuss these ideas. Language allows us to talk about things that we can’t see or touch, like abstract thoughts, imaginary situations, or our feelings. Over time, language may have evolved to help us organize and share our thoughts more effectively. On the one hand, combining symbols in meaningful ways is so hard no other animals have done it like we have, but on the other, it’s so easy for us that even a baby can do it!
References
Dautriche, I., Buccola, B., Berthet, M., Fagot, J., & Chemla, E. (2022). Evidence for compositionality in baboons (Papio papio) through the test case of negation. Scientific Reports, 12(1), 19181.
Fodor, J. A., & Pylyshyn, Z. W. (1988). Connectionism and cognitive architecture: A critical analysis. Cognition, 28(1-2), 3-71.
Hirsh-Pasek, K., & Golinkoff, R. (1996). The origins of grammar. Cambridge, MA: MIT Press.
Partee, B. H. (2004). Compositionality in formal semantics: Selected papers of Barbara H. Partee. Malden, MA: Blackwell Pub.
Pomiechowska, B., Bródy, G., Téglás, E., & Kovács, Á. M. (2024). Early-emerging combinatorial thought: Human infants flexibly combine kind and quantity concepts. Proceedings of the National Academy of Sciences, 121(29), e2315149121.
Lily Zhu
Author
Lily is a first year graduate student who studied Cognitive Science and Data Science at Johns Hopkins University. She is interested in how children integrate multiple sources of information for language learning and develop their compositional knowledge. In her free time, Lily enjoys running, gaming, and experimenting with desserts.
Elika Bergelson
Principal Investigator