Before bilingual children can say anything, they must learn to distinguish between the two languages that are spoken to them, and they must learn to make useful perceptual distinctions in each of them in order to understand what is said to them. Conversational interaction and non-verbal communication, such as pointing and gaze, aid children in attending to and processing aspects of their “language input environment” (De Houwer, Reference De Houwer2009, Reference De Houwer2011), “exposure” in Carroll's terms. As Carroll points out, children must build up their linguistic categories based on their intrinsically category-free “exposure”: they must process speech to acquire language. This reminds me of Wijnen's (Reference Wijnen, Beers, van de Bogaerde, Bol, de Jong and Rooijmans2000) notion of language intake, which is the “data base children use to derive hypotheses on the structure of the target grammar” (p. 174), and which constitutes children's selection from what I will continue to call input, pace Carroll. Wijnen sees data selection from the input as determined by the processing of physical and distributional characteristics of spoken utterances addressed to children as well as the linguistic knowledge and skill children have already acquired. In my view, Wijnen's dynamic model can be applied to any kind of linguistic domain other than grammar, including semantics, pragmatics, the lexicon and phonology (for evidence on the latter two, see McGillion, Herbert, Pine, Vihman, DePaolis, Keren-Portnoy & Matthews, in press).

Children's knowledge base and the opportunities to practice their linguistic skills are continuously expanding, as are children's nervous systems. The limits and possibilities of very young minds (say, at age 1) and their concomitant states of knowledge and histories of language practice are not the same as those of older child minds (say, at age 10). Such differences in maturity and opportunities for language practice will lead to different kinds of intake options from the dual language input environment, so that one particular kind of input may have different effects on children at different levels of maturity and practice. Furthermore, bilingual children's opportunities and willingness to practice (use) a particular language may depend quite directly on social-psychological factors, such as their language attitudes (see also Carroll). Quite fundamentally, parental attitudes and beliefs are important in shaping bilingual children's language input environments (De Houwer, Reference De Houwer, Extra and Verhoeven1999).

Parallel to children's constantly changing linguistic maturity and practice levels, qualitative and quantitative aspects of their language input environments are changing, too, and do so in response to children's levels of language skill. For instance, mothers talk more to older than younger bilinguals (De Houwer, Reference De Houwer, Grüter and Paradis2014); mothers may stop using a particular language in response to a child's perceived difficulties with it (De Houwer & Bornstein, Reference De Houwer and Bornstein2016); features of infant directed speech are typically absent in child directed speech (CDS) to six-year-olds; and parents don't normally tell complicated stories to babies. Also, young bilingual children's linguistic performance in each language rapidly and dynamically changes in response to changes in the overall amount of input in a language that they receive (De Houwer, Reference De Houwer2009).

The challenge is to connect the ever-changing features of input in each language with bilingual children's maturing development, and to map their language intake along the way. This assumes that we at least know what bilingual language input environments look like. I join Carroll in calling for more, and more reliable measures of CDS. Carroll is spot-on in criticizing the common group approach: grouped data are inappropriate for examining input effects on bilingual development, since these are fundamentally individual-based processes (Lanza, Reference Lanza, Cenoz and Genesee2001). Also, in studies of input in bilingual acquisition once-off questionnaires are often used to assess the “division of labor” amongst languages. These can only give a very approximate picture of children's “data base”, as Carroll also notes. Repeatedly using a language diary is likely to be more accurate (De Houwer, Reference De Houwer2011). Actual recordings of CDS in bilingual families (De Houwer, Reference De Houwer and Clark1997, Reference De Houwer, Grüter and Paradis2014; De Houwer & Bornstein, Reference De Houwer and Bornstein2016) are preferable, but require huge resources. It is in and through interaction, after all, that children hear language.

Finally, I concur with Carroll that studies should distinguish between children with bilingual input from birth and those who start hearing a second language later. It is a basic error not to do so (De Houwer, Reference De Houwer2009).

Carroll has done the field of bilingual acquisition a great favor in asking that we fundamentally reflect on the notion of input. Let's rise to the occasion.