Thirty-Six-Month-Olds Use Story Context to Predict Who Gets Mentioned Next
By 36 months of age, children are not simply waiting for language to arrive word by word. In an eye-tracked comprehension task, they used prior story information to anticipate which character was about to be referred to, revealing a form of referential prediction that depends on a structured mental model of the situation rather than on the sentence alone.
The finding places a sophisticated cognitive operation squarely within early childhood language comprehension. Adults routinely build causally organized situation models as they listen, linking who is present, what has happened, and which entities remain available in the scene. Here, toddlers showed a version of that same process. When the narrative established that one goose had already left, children treated a later mention of “the goose” as if it pointed unambiguously to the remaining animal, even though the story still contained two geese in the broader context.
Eye Movements Revealed Prediction Before Disambiguation Arrived
The study used critical sentences such as “She asked the goose with the boots to play with her,” and the timing mattered. Before the sentence reached the point where a fully disambiguating phrase could do its work, children were already directing their gaze toward the target animal when story context made that referent plausible. In practical terms, their visual attention moved ahead of the unfolding syntax.
One sentence captures the central result plainly: the children were not only interpreting language, they were anticipating it.
That pattern is important because it separates simple word recognition from context-sensitive inference. If a child hears about two geese but one has been established as absent from the scene, the remaining goose becomes the only viable referent. The child’s brain must therefore integrate discourse history, scene continuity, and semantic plausibility in real time. The task did not measure passive recall; it probed whether young listeners could use a narrative model to predict a still-present character before the linguistic signal fully resolved the reference.
Language Skill Shaped Whether Children Generated the Inference
Individual differences in language ability predicted whether children generated these inferences. Stronger language skills were associated with better performance in using the unfolding story to infer the intended referent. Exploratory analyses also suggested that children with language delay did not generate the same inferences.
Those differences fit a familiar developmental logic. Referential prediction depends on more than vocabulary size or sentence processing alone; it draws on the capacity to maintain a coherent representation of the story, keep track of which entities remain relevant, and use that model to constrain interpretation. Children with more advanced language skills are more likely to have the processing resources needed to hold those elements together quickly enough to guide eye movements before the sentence is complete.
The contrast with language delay is clinically and developmentally meaningful. If inferential prediction is reduced or absent, the child may still hear the words correctly but fail to map them onto a rich enough situational structure to narrow the reference in advance. In day-to-day conversation, that would slow comprehension, especially in stories, instructions, or classroom talk where multiple characters or objects compete for attention.
Situation Models Appear Earlier Than Many Theories Would Suggest
The work extends the reach of situation-model theory into a much earlier developmental window than often assumed. Rather than waiting for older children to become skilled readers or more mature listeners, the capacity to infer who can plausibly be mentioned next was already visible at 36 months. That places predictive comprehension near the center of early language development, not at its endpoint.
The mechanism is cognitively plausible. A child who hears that one goose has left the scene no longer treats both geese as equally available. The referential field narrows because the narrative itself has altered the causal structure of the world being described. In that sense, the comprehension system is doing more than decoding grammar; it is simulating a tiny world and updating that world as events unfold. The result is a prediction about reference that emerges before the disambiguating words arrive.
What the Findings Mean for Early Language Assessment
The study’s design matters because it captures comprehension online, where prediction actually happens, rather than after the fact in a memory test. Eye tracking can detect whether attention shifts toward the target while the child is still processing the sentence, offering a window into the timing of inference generation. For developmental science, that timing is the story.
These data also raise a practical possibility for early assessment. If referential prediction depends on language growth and appears weaker in children with delay, then tasks like this may help distinguish children who merely process words slowly from those who struggle to build the situation model needed for real-time comprehension. The present findings do not establish diagnostic utility, but they do show that predictive reference is measurable in preschoolers and linked to language skill.
One implication is immediate: a child’s difficulty in conversation may reflect a failure of prediction, not only a failure of word knowledge.
A Narrow but Important Window Into Comprehension Development
The sample included 115 36-month-olds, and the report focused on a single kind of inferential task. Even so, the pattern is striking because it isolates a core feature of human communication: speakers rarely name every possible referent explicitly, and listeners must constantly infer what comes next from context. Children who can do that at age 3 are already operating with a more adult-like comprehension architecture than many assume.
The findings stop short of showing how early this ability begins, how stable it remains across different narrative structures, or how strongly it predicts later language outcomes. They also leave open whether the same predictive machinery operates in broader social exchange, not just in carefully designed story settings. What the data do show is narrower and more concrete: by 36 months, some children are using situation-level information to forecast reference, and language delay may blunt that capacity.
For developmental clinicians and researchers, that makes referential prediction a promising marker of how efficiently a child is building meaning from discourse in real time.