This special issue celebrates the fortieth anniversary of the Journal of Child Language by presenting reflections on the past and future of research on child language acquisition. In the spirit of this celebration, I review informally the major principles we have identified, the ideas we have rejected, and some of the crucial issues remaining to be explored. Beginning with the positive findings, we can sort the principles we have confirmed into four categories: learning principles, processing principles, neurolinguistic principles, and social interaction principles.

Let us begin with consideration of some of the overarching principles governing child language learning.

1. 1. Structure and productivity. Wundt, Leopold, Brown, Berko, and many others have demonstrated the pervasive impact of structure building in language acquisition. Children build structures and generalizations on all linguistic levels, including phonology, morphology, semantics, syntax, and conversation.

2. 2. Compositionality. In general, children learn simple structures before more complex structures. For example, they first learn words and then learn to put these words together into phrases. In phonology, children learn to control their vocal cords and tongue before producing consistent target sounds (Oller, Reference Oller2000). At the same time, children can also pick up complex patterns as single rote items from which they then pull out smaller components over time.

3. 3. Generalization. Structures arise from generalization over instances. Within each of the linguistic levels, generalization takes on slightly different forms. In phonology, phonemic categories are abstracted as central tendencies across exemplars. In word learning, concepts are extracted from specific examples.

4. 4. Lexical basis of syntax. In syntax, lexical combinations give rise to item-based patterns and then feature-based patterns through a process of increasing levels of generalization (MacWhinney, in press). The pervasive nature of this process is reflected in the tight correlation between lexicon at 1;8 and grammatical abilities eight months later (Bates & Goodman, Reference Bates, Goodman and MacWhinney1999).

5. 5. Statistical learning. Statistical learning operates sequentially to extract candidate words and distributionally to extract sound categories and phonotactic patterns (Thiessen & Erickson, Reference Thiessen, Erickson, MacWhinney and O'Grady2014).

6. 6. Bootstrapping. The categories and patterns that arise from generalization and statistical learning can then facilitate acquisition of further categories and patterns. For example, learning of Hungarian vowel contrasts then allows the child to acquire Hungarian vowel harmony. There is an interplay between generalization and episodic encoding of specific instances. Some children move quickly to form generalizations; others are more conservative. Some encode episodes in great detail; others use less precise encodings.

7. 7. Availability. Initially, children extract patterns from forms that are easily detectable and highly frequent. Detectability is facilitated when cues appear at the end of words and phrases or when they are constructed as full syllables or separable properties (Slobin, Reference Slobin and Slobin1985).

8. 8. Prototypicality. Because patterns are extracted from specific instances, early patterns function as prototypes. For example, Cantonese-speaking children work initially with a sentence pattern in which animacy, topicality, and word order are all expressed through the subject (Chan, Lieven & Tomasello, Reference Chan, Lieven and Tomasello2009). Prototype organization is also evident in phonology (Kuhl, Reference Kuhl1991) and semantics (Mervis & Rosch, Reference Mervis and Rosch1981).

9. 9. Reliability. Although availability, detectability, and prototypicality play major roles in early learning, later learning is influenced more by cue or pattern reliability (McDonald & MacWhinney, Reference McDonald and MacWhinney1991).

10. 10. Ambiguity. Because ambiguity represents a specific form of unreliability, it can delay acquisition. For example, the use of the German article der to mark either masculine nominative or feminine/plural genitive can lead to occasional confusions (Pelham, Reference Pelham2011).

11. 11. Distinctions are functionally contrastive. Children learn to set up categories and distinctions on the basis of contrasts in meaning or function (Clark, Reference Clark and MacWhinney1987). Some distinctions map directly to contrasts in the real world, others to obligatory grammatical distinctions, such as gender or verb class. The extraction of distinctions arises primarily through self-organization.

Many of the core principles derived from the study of adult language processing apply with equal force to children's processing and learning.

1. 1. Competition. Competition arises between patterns that have been formulated on different levels of generality or based on conflicting datasets.

2. 2. Multilingual competition. We have increasing evidence for concurrent activation of multiple languages during processing in adults. Although adults develop attentional processes that can limit crosstalk, there is still interlanguage transfer, particularly in the context of code-switching environments. These same transfer effects can be identified in children (Yip & Matthews, Reference Yip and Matthews2007).

3. 3. Combination vs. rote. The most common competition is between analytic or combinatorial forms and synthetic or rote forms, although there are also competitions between alternative analytic forms and analogistic patterns (MacWhinney, Reference MacWhinney1978).

4. 4. Minimization. We have learned that the processor operates in ways that minimize attachment, distance, and memory costs (O'Grady, Reference O'Grady2005). These forces operate in children as well as adults, but they can be seen most clearly in older children.

5. 5. Formulation. Like adults, children correct their errors, both for form and substance (MacWhinney & Osser, Reference MacWhinney and Osser1977).

Ongoing advances in neuroscience continue to teach us about the neuronal bases for language development.

1. 1. Early differentiation. Genetic and epigenetic variation can lead to different patterns of brain structure and connectivity, some of which lead to language disorders (Bishop, Reference Bishop2013).

2. 2. Language sparing. Children who experience brain lesions often develop normal language functioning (MacWhinney, Feldman, Sacco & Valdes-Perez, Reference MacWhinney, Feldman, Sacco and Valdes-Perez2000). Language seems to be uniquely favored in this regard, because of the fact that it utilizes alternative methods for constructing messages.

3. 3. Vulnerability of timing. Problems with the timing of neuronal integration seem to be central to various language disorders, as well as delays in normal acquisition (Howell, Reference Howell2011; Roelofs, Reference Roelofs, Guendozi, Loncke and Williams2011).

4. 4. Articulatory challenges. The control of phonological output is a major development challenge. It involves a high level of precise control and coordination of fine-grained motions that must be extended to a wide variety of articulatory targets (Browman & Goldstein, Reference Browman and Goldstein1992).

5. 5. Entrenchment. The development of prototypes is one reflection of the general tendency for neuronal patterns to become entrenched through learning. This effect is the major determinant underlying age-related declines in second language learning ability (MacWhinney, Reference MacWhinney, Gass and Mackey2012).

6. 6. Plasticity. Perhaps the most dramatic evidence for plasticity is the fact that ASL speakers make use of basic language processing areas for controlling the production of sign language (Corina, Vaid & Bellugi, Reference Corina, Vaid and Bellugi1992).

7. 7. Specialization. Although neurolinguistic development reflects flexibility, sparing, and plasticity, there are also limits to the plasticity of certain cognitive functions. For example, although children with focal lesions recover language functions, they cannot recover from right hemiplegia. These limitations are particularly evident in the entrenchment of motor and auditory processes (Guion, Flege, Akahane-Yamada & Pruitt, Reference Guion, Flege, Akahane-Yamada and Pruitt2000).

8. 8. Individual variation. Children differ markedly during the earliest stages of learning. Some emphasize clarity or faithfulness in pronunciation; others use filler words (Peters & Strömqvist, Reference Peters, Strömqvist, Morgan and Demuth1996), grunts, and mushy articulations. Much of this variation is based on underlying genetic variation, as further filtered through epigenesis. Despite this wide initial variation, all normal children end up learning their native languages.

We have also learned a great deal about the impact of social structures on children's language learning.

1. 1. Trust and identification. Children operate within a zone of proximal development grounded on trust in and identification with their caretakers (Tomasello, Reference Tomasello2003).

2. 2. Common ground. Like adults, children rely on the construction of common ground to advance conversation (E. Clark, Reference Clark, MacWhinnney and O'Grady2014).

3. 3. Cooperativeness. Caretakers try to facilitate children's learning of word meanings by giving them labels that correspond to what they think the children have in mind (Brown, Reference Brown1958).

4. 4. Simple input. Caretakers simplify the child's learning task by providing simple input, often in the shape of single-word utterances or labels (MacWhinney, in press).

5. 5. Scaffolding. Caretakers provide further support for language learning by scaffolding utterances and exchanges through completion, recasting, and repetition (Bohannon & Stanowicz, Reference Bohannon and Stanowicz1988).

6. 6. Fine-tuning. Caretakers provide input that is finely tuned to the child's advancing developmental level (Sokolov, Reference Sokolov1993).

7. 7. Input variability. Some children receive extensive rich input; others little input at all (Hart & Risley, Reference Hart and Risley1995). Children who receive less adult input will have smaller vocabularies and subsequently weaker academic performance (Snow, Reference Snow and MacWhinney1999).

8. 8. Code selection. Multilingual children show early sensitivity to the many social conditions determining code selection (Genesee, Boivin & Nicolaidis, Reference Genesee, Boivin and Nicolaidis1996).

It would be easy to expand each of these lists, by including additional topics and more specific instantiations of each of the principles. My hope is that a simple listing of this type will make it clear how much progress we have made in the last half-century of child language research. Of course, not all of these findings are universally accepted. It is important for academics to challenge ideas through debate and discussion, and each of these principles will surely be further refined through this process. However, we can see from this listing that we now have an increasingly solid grasp on the basic nature of first language acquisition.

Much of this progress has occurred through studies challenging earlier simple formulations. Let us consider some of the ideas that have been rejected in this way.

1. 1. Degenerate input. According to Chomsky (Reference Chomsky1980), the input to the child is so full of retracings, errors, and incompletions that it provides no clear indication of the possible sentences of the language. However, Newport, Gleitman, and Gleitman (Reference Newport, Gleitman, Gleitman and Snow1977) showed that “the speech of mothers to children is unswervingly well-formed”.

2. 2. The logical problem. Gold (Reference Gold1967) demonstrated that, under certain assumptions, language identification in the limit would be unattainable. However, more recent work (A. Clark, Reference Clark, MacWhinnney and O'Grady2014; MacWhinney, Reference MacWhinney2004) has questioned the relevance of Gold's result.

3. 3. Universal phonology. Early accounts of the development of phonology claimed that, during the babbling period, children produced all the possible sounds of human language. However, closer studies have shown that the range of sounds produced during babbling is limited to certain basic patterns (Cruttenden, Reference Cruttenden1970; Davis & MacNeilage, Reference Davis and MacNeilage1995).

4. 4. Discontinuities. Views of language development that emphasize discontinuities, such as Jakobson's (Reference Jakobson1968) claims about the discontinuity between babbling and early phonology, have been rejected.

5. 5. Shaping. Behaviorist accounts of language learning were based on notions of shaping and differential reinforcement that have now been discarded (Chomsky, Reference Chomsky1959) .

6. 6. Cognitive determination. During the 1970s, researchers pursued the idea that advances in object permanence during stage 6 of the sensorimotor period would trigger advances in vocabulary and syntax. However, studies failed to confirm these predictions (Corrigan, Reference Corrigan1979).

7. 7. Mutual exclusivity. The idea that children have a built-in constraint against referring to things by different names has been invalidated by evidence that bilingual children are happy to accept multiple names, as well as children's willingness to learn forms that describe things in alternative contexts.

8. 8. Parameters. The idea that children learn language by setting values on a small set of syntactic parameters was difficult to formulate and never received clear empirical support.

9. 9. Back propagation. Attempts to model language acquisition using back propagation networks without additional computational structures failed to account for important aspects of the learning of morphology and syntax.

Some of these concepts may be resurrected in modified form in future accounts, but for now they play a marginal role.

Child language studies are being infused by many new cross-disciplinary insights and methodological developments. Laboratory methods such as event-related potential (ERP), near-infrared spectroscopy (NIRS), preferential looking, and eye-tracking provide us with increasing detail regarding the flow of processing. The development of standard stimulus sets for objects, words, sounds, and interactions has added further structure to these experimental efforts, tracking infants' abilities on a month-by-month basis as they progress through language learning and socialization, although often through a cross-sectional methodology. Child language studies are also beginning to link up with work in developmental biology that characterizes the genetic and epigenetic basis of language development. Perhaps the most dramatic methodological advances have come from the application of computational technology to child language. Using corpora from CHILDES, researchers are formulating increasingly detailed computational models of the learning of phonology, lexicon, syntax, and conversation.

Despite all of this progress and promise, there are large areas of language learning that are not receiving adequate attention. Although we have formulated many possible accounts and mechanisms, our understanding of the causes of disorders such as stuttering and specific language impairment remains vague. Another major gap involves our understanding of how children refine word meanings over time across episodes in real life, using feedback from multiple encounters. By restricting our attention to laboratory experiments, we have failed to develop a picture of how children sharpen word shapes, and how they come to deal with polysemy, homonymy, and ambiguity. This problem is particularly acute when we come to the learning of school-based concepts such as symmetry, dependability, majority, species, and politics (MacWhinney, Reference MacWhinney, Gershkoff-Stowe and Rakison2005). Understanding the refinement of such terms will require us to dig more deeply into children's learning of social and scientific structures. To do this, we will need to link up work on the structuring of mental models of complex systems and processes in adults with their development in children.

There is a similar gap in our understanding of the development of persuasion. When I began recording my sons' language development, it was my intention to study this, but other things have intervened. Fortunately, Karniol (Reference Karniol2010) has shown how one can use transcripts from CHILDES and other sources to make progress on this issue.

These last two gaps illustrate an overarching problem we have in terms of tracking acquisitional patterns that operate across longer timeframes. Relying only on cross-sectional data, we cannot understand the dynamic aspects of these processes. For example, we need to understand exactly what interactions and social configurations can lead a bilingual child to reject use of the home language, when entering school. To study such processes, we must commit ourselves to increasingly ambitious attempts to record the development of individual children or perhaps several children in similar sociolinguistic contexts. The growth of methods for collecting, transcribing, and analyzing large video corpora may make this increasingly possible in the future (Roy, Reference Roy2013).