A complex network-based analysis visualizes how language is acquired in childhood
Sept. 7, 2026
- A study co-led by the IBE (CSIC-UPF) and IdISBa, with the participation of IFISC (CSIC-UIB), transforms spontaneous conversations of children into mathematical networks to study how language is acquired in early childhood.
- Based on networks constructed from 142 conversation sessions in Dutch, the team has identified a common trajectory of linguistic development and pinpointed a tipping point around the age of two.
- The work, published in Physical Review Research, opens a promising avenue to analyze the acquisition of syntax, with potential applications in the clinical field and in the study of language evolution.
Learning to speak is a highly complex process. As the first words appear and sentences lengthen, language also reorganizes and matures. This is the premise of a mathematical approach first proposed 20 years ago at the Instituto de Biología Evolutiva (IBE), a joint center of the Consejo Superior de Investigaciones Científicas (CSIC) and Universidad Pompeu Fabra (UPF).
Now, a research team led by the IBE (CSIC-UPF) and the Health Research Institute of the Balearic Islands (IdISBa), with the participation of IFISC (CSIC-UIB), has developed a quantitative method that allows researchers to study how syntax evolves during language acquisition. Using mathematical networks built from infant speech, the team has identified an abrupt transition around the age of two and analyzes how children with typical and atypical development are positioned along this trajectory.
From infant conversations to syntactic networks
The work analyzes 142 sessions of child speech in Dutch from the CHILDES database: 82 from children with typical development and 60 from children with atypical development, including Down syndrome, hearing impairment, and specific language impairment.
To study these conversations, the team analyzed the syntax of each sentence using dependency trees, which show how words connect to each other. Afterwards, they merged the trees from each session into a "syntactic network": a mathematical representation in which each word functions as a node and each syntactic relation as a link.
"Each session was transformed into a syntactic network and characterized by 18 properties, such as vocabulary size, number of connections, presence of cycles, or similarity between words" comments Víctor M. Eguíluz, a researcher at the Institute for Cross-Disciplinary Physics and Complex Systems (IFISC, CSIC-UIB) who participated in the study.
A marked transition around the age of two
With these metrics, the research team constructed a mathematical space in which each point represents the syntactic structure produced by a child in a session. In this space, the children's networks do not appear randomly dispersed, but rather follow an orderly maturation trajectory.
The analysis reveals that syntactic organization becomes more complex over time and undergoes a sharp transition around two years and two months of age. During this period, networks shift from simpler structures, resembling trees or chains, to denser and more interconnected configurations, with syntactic cores and cycles reflecting a greater combinatorial capacity. In linguistic terms, words begin to participate in more (and more diverse) syntactic relations.
"This jump is reminiscent of a phase transition, like when water suddenly freezes upon crossing zero degrees: the syntactic network remains relatively stable until it surpasses a threshold and abruptly reorganizes its structure", comments Luís F. Seoane, principal investigator of the BIT-Lab at the IBE (CSIC-UPF) and co-leader of the study.
Infant syntax follows a common trajectory
The study also compares this trajectory with that of boys and girls with atypical development, particularly with Down syndrome, hearing impairment, or specific language impairment. The results show that their syntactic networks tend to align with the same path observed in typical development, although they often fail to reach the final phases of the process.
"The data show that syntax progression can halt before completing this transition in children with atypical development, even if they are older", explains Lluís Barceló-Coblijn, IdISBa researcher and first author of the study. "This 'stagnation' along a shared route suggests that language maturation follows a shared pattern, but that certain clinical conditions can hinder the final key step", he adds.
The study could contribute in the future to the development of new tools to characterize linguistic trajectories in clinical contexts. The team warns, however, that it will be necessary to expand the analysis to more languages, social contexts, and pediatric populations, as well as to validate the method in specific clinical studies.
A new avenue to study language evolution
This mathematical map of early linguistic development allows the team to participate in a historical debate: how the acquisition of syntax in early childhood can shed light on the origin and evolution of human language.
"We observe a defined structure, with phases and abrupt jumps, that could offer clues about the evolution of language. Just as embryonic development preserves traces of our evolutionary history, we speculate that some early stages of linguistic development could help us understand possible ancestral forms of language use", points out Seoane.
Although language leaves no fossils, the team suggests that some phases of its acquisition could act as a window into earlier stages of its evolution. This approach opens the possibility of studying child linguistic development not only as an individual learning process, but also as a way to explore how human linguistic capacity could have emerged and organized.
Photo: Pexels
Transitions between discrete developmental stages during language acquisition through syntax network morphospace; Physical Review Research 8, 023174 (2026); Lluís Barceló-Coblijn, Víctor M. Eguíluz and Luís F. Seoane.
