From commands to invitations: the pragmatic softening of child-directed speech, 1960s–2000s
- 2.47M
- parent tokens, four corpora, 1960s–2000s
- −41%
- imperatives per 10K words
- +76%
- softeners per 10K words
- 8.5 → 2.9
- commands per softener

Child-directed speech is the simplified register adults use with young children. Its structure is well described. Its pragmatic texture across historical time is not. American parenting is documented to have moved from authoritarian to authoritative norms over the same decades; if that shift reshaped what parents say to children, the corpus should show it.
The corpus
CHILDES was never built as a historical record, but its holdings span five decades. Four corpora, restricted to mother and father speech to children aged 18 to 42 months, give 2.47 million parent tokens: Brown (1962–66), Kuczaj (1973–75), HSLLD (1988–96), and Providence (2001–05). Two lexicons are tracked, imperatives (stop, don’t, no, sit, come, look, put…) and softeners (please, maybe, could, would, honey, careful…), each in a narrow and an expanded form so that no single word list can carry a result.
What moved
Imperative tokens fall from 482 to 285 per 10,000 parent words, a 41% drop that holds under both lexicons. Softeners rise from 57 to 100 per 10,000, up 76%. The ratio collapses from 8.5 commands per softener in the 1960s to 2.9 in the 2000s, nearly a threefold change in register. Lexical diversity also grew, with social and emotional vocabulary up 30%.
What did not
The point of the controls is to make the finding falsifiable. Function words, number words, and mean length of utterance are register-defining features that should not drift with parenting culture, and they do not: all three are flat within a few percent across the decades. Technology vocabulary is likewise flat, which is its own small finding.
Reading it
Blum-Kulka and Ervin-Tripp mapped directive mitigation synchronically; no prior study tracked it diachronically in acquisition corpora. The limitation is that corpus and decade are confounded, since each decade is a different lab with different methods. The robustness gates mitigate that without removing it, and the next step is expansion to every English corpus in CHILDES with more than 500 utterances.