Children quickly gain enormous linguistic knowledge duringearly development, in part due to low-level features of theirparents’ speech. Some posit that parents contribute to theirchild’s language development by tuning their own languageaccording to their child’s developmental abilities and needs(Bruner, 1985; Snow, 1972). Here, we investigate this hypoth-esis by examining ‘alignment’ at the level of syntax and func-tion words in a large-scale corpus of parent-child conversationsand measuring its association with language development out-comes. To do so, we employ a statistical model of alignment toestimate its presence in our dataset and its predictive impact ona measure of vocabulary development. Our results corroborateprevious findings, showing strong alignment for both parentsand children; in addition, we demonstrate that parental align-ment is a significant predictor of language maturity indepen-dent of demographic features, suggesting that parental tuninghas strong ties to a child’s language development.

Children do not learn language from passive observation ofthe world, but from interaction with caregivers who want tocommunicate with them. These communicative exchanges arestructured at multiple levels in ways that support support lan-guage learning. We argue this pedagogically supportive struc-ture can result from pressure to communicate successfully witha linguistically immature partner. We first characterize onekind of pedagogically supportive structure in a corpus analy-sis: caregivers provide more information-rich referential com-munication, using both gesture and speech to refer to a singleobject, when that object is rare and when their child is young.Then, in an iterated reference game experiment on MechanicalTurk (n = 480), we show how this behavior can arise from pres-sure to communicate successfully with a less knowledgeablepartner. Lastly, we show that speaker behavior in our experi-ment can be explained by a rational planning model, withoutany explicit teaching goal. We suggest that caregivers’ desireto communicate successfully may play a powerful role in struc-turing children’s input in order to support language learning.

Young children learn language at an incredible rate. Whilechildren come prepared with powerful statistical learningmechanisms, the statistics they encounter are also prepared forthem: Children learn from caregivers motivated to communi-cate with them. Do caregivers modify their speech in orderto support children’s comprehension? We asked children andtheir parents to play a simple reference game in which the par-ent’s goal was to guide their child to select a target animal froma set of three. We show that parents calibrate their referringexpressions to their children’s language knowledge, produc-ing more informative references for animals that they thoughttheir children did not know. Further, parents learn about theirchildren’s knowledge over the course of the game, and cali-brate their referring expressions accordingly. These results un-derscore the importance of understanding the communicativecontext in which language learning happens.

How do children learn the typical features of objects in theworld? For many objects, this information must come from thelanguage they hear. However, language does not veridicallyreflect the world: People are more likely to talk about atypicalfeatures (e.g., “purple carrot”) than typical features (e.g., “or-ange carrot”). Does the speech children hear from their parentsalso overrepresent atypical features? We examined the typical-ity of adjectives produced by parents in a large, longitudinalcorpus of parent-child interaction. Across nearly 2000 uniqueadjective–noun pairs, we found parents’ adjectives predomi-nantly mark atypical features of objects, although parents ofvery young children are relatively more likely to comment ontypical features as well. We then used vector space models toshow that learning the typical features of common categoriesfrom linguistic input alone is challenging even with sophisti-cated statistical inference techniques.