14.1 Earlier Thinking on Language Learning

Jean Piaget is famous for studying his own children, as a way to understand human cognitive development. A psychologist, Piaget came to understand development in children as a movement from egocentrism, a focus on the self, to sociocentrism, a sense of how individuals interact with the society around them. Piaget structured cognitive development in a series of stages: from birth to 2 years old, involvement with the experience of the senses and movement; from 2 to 7 years old, the beginnings of language use to ask questions about the world around them; from 7 to 11, children begin to use logic and reasoning; and from the age of 11 through the rest of their lives, children use abstract reasoning to solve problems in the world and in their interactions with others. Language acquisition, then, is part of a process of cognitive development that, for Piaget, leads to the reasoning abilities of adults. His stages answer the question of how we become adults, from the tabula rasa (blank slate) with which we begin our lives.

When researchers (mainly in psychology) began to think about how children learn language, they started with the simple observation that kids tend to repeat what they hear. The suggestion that children learn language through imitating their caregivers made practical sense, especially given the behaviorist ideas of positive and negative reinforcement. Noam Chomsky’s assertion about language acquisition – note the change in vocabulary from “learning” to “acquisition” – was that no, children don’t learn their native language by imitation or behaviorist-style reinforcement. Or by flashcards (see Figure 14.1). As Chomsky rightly pointed out, what comes out of kids’ mouths is not simply an imitation of adult language. Kids’ utterances are new and often unpredictable given the input they receive from their caregivers, and, as anyone who has spent time with a three-year-old knows, explicit instruction will only get you so far.

Figure 14.1 Spoiler alert: This is not how we learn language.

Piaget’s structure of ideas fit well with the development of notions of language acquisition among American generative linguists. Beginning in the 1960s and 1970s, a formal model for a physical symbol systems hypothesis (PSSH) in cognitive development arose from Chomsky’s Syntactic Structures (Reference Chomsky1957), the founding book in the generative school. PSSH holds that we learn to manipulate physical symbols in the brain according to particular, logical rules (Figure 14.2). These symbols come to exist as a physical structure that bears information. These symbols can be combined to form complex symbol structures, which can be manipulated and transformed to create our complex human thought processes. The physical symbol theory, as relevant to language, is known as the language of thought hypothesis, which suggests that we combine the symbols in our brains in the structure of sentences and that we respect that structure as we manipulate and combine sentences in order to create complex thoughts. In this view language is a logical system, with meanings in symbols organized into thoughts by means of a definite structure that we can call grammar (or syntax).

Figure 14.2 Physical symbol systems: the brain and the computer.

And as dated as the computer looks in this illustration, PSSH (physical symbol system hypothesis) still hangs on as part of the traditional core of research into Artificial Intelligence.

The way that we learn about physical symbols and their grammar in the brain fits Piaget’s development structure. In the first two stages, we acquire symbols and, as time goes on, the grammar to use them to ask questions about the world. In Piaget’s next two stages children learn to create more and more complex and abstract thoughts. One big question about the process is how children learn symbols and their grammar. In the 1960s Chomsky postulated that a Language Acquisition Device (LAD) must be present in the brain, to help children learn language. The main reason for thinking that there must be an inborn biological aid for language acquisition was the assumption of poverty of stimulus. That is, the language that children hear is not systematic enough in the way that they hear it. The world is too complicated for children to be able to learn about symbols and grammar from it. Thus, the LAD must be there in order for children to get to the adult cognitive state predicted by the language of thought hypothesis.

More recently, generativists talk less about the LAD, and more about Universal Grammar, grammatical principles thought to be shared by all human speakers, and about how parameters can be set in the brain that accompany the grammars of the different languages that people actually speak. Think of the principles as being a type of language menu and the parameters being the items that a language selects from the options provided by that menu. A series of different selections thus results in an individual language’s structural characteristics. Language acquisition, then, uses a more or less specific biological aid (the LAD or Universal Grammar and parameter setting) in order for children to learn the physical symbol system and its logical rules for making sentences and larger complex thoughts. After that, children can learn language as both a symbolic system and as a set of (more or less) logical rules for making sentences and larger complex thoughts. The LAD can only function for so long; generativists talk about the critical period, the time during which a child’s LAD is open, active, and ready to receive linguistic input, as being finite. The critical period ends in late childhood or early adolescence (perhaps because the onset of puberty means that the brain has other things to worry about!), and after this period the language learner must rely on the same cognitive faculties used to learn mathematics or geology.

You may ask, however, how much of this approach to language acquisition is a chicken-and-egg problem, where linguists’ expectations of language, as from Piaget and Chomsky, influence their thinking about the cognitive status of language. Are Piaget’s stages the best way to think about cognitive development? Are the physical symbol system and the language of thought hypothesis the best ways to think about how we put our ideas together? On the other hand, do Piaget’s assumptions about what constitutes adult cognition control how he thought about cognitive development? Does an assumption that there are physical symbols in the brain (nobody has ever found any, by the way) control how we think about the creation of grammar in the brain in the language of thought? As we’ll see below, usage-based linguistics looks at language learning as a process that involves chunks of information rather than small bits of meaning pieced together to make larger bits of meaning.

14.2 Language under Construction

The usage-based approach to this topic focuses not on “language acquisition” but instead on “language construction.” Citing the general capacities of human cognition and specifically the capabilities of intention-reading and pattern-finding (and not some sort of acquisition device or mental template), usage-based linguists such as Michael Tomasello have conducted empirical studies that demonstrate how children construct language for themselves as they interact with others in a social environment. Much like his explanation of species-wide (phylogenetic) development of language, Tomasello’s explanation of individual (ontogenetic) language learning begins with pre-linguistic communication, such as pointing, that signals the concomitant development of social cognition and language use. Children as young as 12 months old can understand that, when an adult points to something, there is intention behind it. Whether the intention of pointing toward the next room is to direct the child’s attention to the location of a missing toy, to indicate that it’s time to go to bed, or to inform the child that Mommy’s home, human children understand that there is information attached to the gesture. Intention-reading is a human-specific trait (i.e. a chimpanzee can direct her gaze to the object you are pointing at, but does not understand that you are doing so with the intention of communicating information relevant to her).

Linguists who study statistical learning in language acquisition have shown that even infants can use the frequency of different sounds in order to begin to find words in the continuous flow of speech sounds of their language. Constructions (as described in Chapter 11) have frequencies that allow learners to acquire them, so that they can find sequences of words that mean something in the midst of the continuous flow of speech. These frequencies from the complex system, whether the top-ranked variants on an A-curve or later the differential frequencies of variants that help us to distinguish language in different situations, depend on countable units like sounds, words, and constructions. Language in the brain, then, is not just a logical system of rules that children have to learn, but instead a set of sounds and words and constructions that children can learn in part from how often they hear them. Repetition is not enough to begin, but it asserts itself immediately when children come to the point where they are ready to use grammar in connected speech, and not just isolated words or short phrases. When Piaget said that language use began at the beginning of his second stage, he meant that children, his children, at that time could begin to use grammar to ask questions about their world. What we now know is that language learning begins much earlier, and that children have already succeeded in many language-learning tasks before they get to grammar.

The usage-based approach views the stages of language acquisition as a child’s ongoing construction of that language. The holophrastic utterance, defined by Tomasello (Reference Tomasello, Bavin and Naigles2016: 72) as “the smallest unit in which a person expresses a complete communicative intention” is already a composite structure in that it is both a name for a thing (or type of thing) as well as an attempt to prompt someone into action, such as Juice! as a request for juice. Two-word combinations such as More juice!, More cookie!, More milk!, have one element that remains fixed while the other changes. These combinations aren’t about syntax so much as they are about nailing down the meaning of a situated utterance. Longer utterances follow the same kind of pattern; they mirror the types of phrases that kids hear and, taken as whole, they involve a great deal of repetition. Take the phrase “Let’s go—,” which a child could hear in a number of situations: Let’s go to bed, Let’s go eat, Let’s go find your shoes. This kind of repetition is how children “extract words (with their functions) and, at the same time, how they find analogical patterns across utterances […] and thereby abstract meaningful grammatical constructions” (Tomasello Reference Tomasello, Bavin and Naigles2016: 75). Children are not learning language word by word, or according to settings in the LAD, but instead are continually trying to understand the functional role of an entire utterance, all of its pieces at the same time. Commonalities found across utterances and functions are only later realized to be separate “words” through the power of deduction.

Tomasello (Reference Tomasello2003: 110) writes that “language acquisition is not just triggered by the linguistic environment, as proposed by generative grammarians, but rather the linguistic environment provides the raw materials out of which young children construct their linguistic inventories.” The linguistic environment alone, however, is not enough. The locus of language learning for usage-based linguists like Michael Tomasello is interaction, which provides for the child both the types of constructions and the social contexts for their use.

14.3 Complexity and Acquisition

A complexity perspective provides yet another way to think about language acquisition. Cognitive development does not occur by learning physical symbols, or by learning constructions. Development is a dynamic process where there are no symbols or pre-existing constructions; neurons in the brain just keep firing so that the cognitive system remains in constant motion. We learn something when some patterns of firing are more common than others; we know something when some patterns of firing have become such a habit that they work more quickly than others. The dynamic process of cognitive development is thus a kind of juggling act where the information available mutually influences itself. Neuronal activity demonstrates the same kind of self-organized criticality that we find throughout nature, similar, for example, to the sand dune which is constantly changing due to elemental forces and yet is stable at the same time. Constant neuronal firing eventually yields stable patterns. Language learning thus comprises these emergent patterns in our cognitive activity as we interact with caregivers, and later others, over time.

One way to think about the difference between a complex systems perspective and earlier ideas of cognitive development is to compare how you might get water from one bucket (experience in the world) to another bucket (a child’s mind): You can try to use a teaspoon or a cup to move a little water at a time, or you can just pour from one bucket to another. The older ideas use small units, like a teaspoon or a cup (symbols, constructions) to get information into the developing cognitive system, while the dynamic complex systems idea simply pours, sometimes a little faster and sometimes a little slower, to get information into the system continuously. Contemporary evidence from neuroscience supports the dynamic, continuous process rather than the existence of any sort of unit of information in the brain.

Instead of viewing language development as a series of stages, complexity science looks at the interaction, repetition, and frequency effects as indicators of children’s developmental processes. Take a look at the following data on the frequency with which different grammatical constructions occur as a percentage of mothers’ and children’s speech from Michael Tomasello (Table 14.1).

The types of constructions that children use are not the same as those used by their caregivers (in this case, their mothers). The children have many more fragments and fewer questions, and almost no complex constructions (yet). Instead of matching, the constructions used by the mothers and their children are complementary. The mothers ask questions; the children answer with fragments (e.g. What do you need? More juice), which is very much in line with joint attention behavior. Neither the mothers nor the children use very many subject–predicate constructions, which is what generative grammar is focused on. Mothers use more imperatives, in line with their role as caregiver. What we see in Table 14.1 is that repetition and frequency alone cannot explain what the mothers and children actually do. Recurrence, frequency, and cognitive development may be hard to tease apart, but they all certainly appear in the data.

What is clear from empirical evidence is that the frequency profiles generated in the complex system of speech help children to learn their native language. Children might settle on went as the past tense form of the verb to go not because of a prescriptive rule or because of an “exception” rule attached to an entry in their mental dictionary, but instead because within the interactions that they have with caregivers, teachers, and books, went is the form most often found. Complex systems do not conflict with the newer findings from construction grammar. The fact that one verb is massively more common in any construction is just what we would expect from a complex system: an A-curve for a frequency profile. We know that the complex system of speech will produce just such a frequency profile for every construction, because of the property of scaling. This distributional pattern gives all children, learning whatever language is spoken in their homes, high quality input to learn from. The complex system ensures just the opposite of poverty of stimulus: Children have ready-made targets for things to learn in the highest frequency features on the A-curves for every construction and every other aspect of language. The fact that children first learn isolated nouns and names comes from joint attention, and the apparently sudden changeover at age 2 to children’s use of grammar just means that children at that age have progressed from isolated words and phrases to the full complex system of speech, including function words along with nouns and phrases. Soon thereafter, children undergo what appears to be an explosion in their knowledge of vocabulary: This too comes from the distributional pattern of the complex system, as children acquire many additional words that are just below the top rank on the A-curve. The process of learning your language then goes on for the rest of your life, as you continue to experience A-curves for linguistic features in all of the situations that you learn to command as you grow older.

14.4 Second Language Acquisition

Learners of a second language (L2) sometimes think that memorizing vocabulary and verb paradigms is the path to competency. It doesn’t take long to realize that learning an L2 isn’t that simple. Proponents of generativism argue that the same UG-based acquisition process present for first language (L1) acquisition is also present for L2 acquisition. In fact, the mental grammar of a learner’s L1 might very well constrain the learning of an L2 if there are lots of differences in the UG parameter settings. These types of constraints can create a kind of interlanguage, in which characteristics of an L1 are applied to an L2; interlanguage thus falls short of native-like use of the L2. A speaker who becomes stuck in interlanguage is said to have fossilized and at this point shows little promise of moving closer to proficiency in the L2. The generative concern with second language learning also has to do with the critical period hypothesis; older language learners may have limited or no access to the UG “instinct” for language. Generative advice for learning an L2? Do it younger and often. (Good advice no matter what side of the generative/structuralist debate you’re on.)

In terms of their approach to language learning (and language teaching) systemic functional (SFL), usage-based, and complexity perspectives have much in common. SFL, for instance, holds that language use is anchored to social contexts and that we process and store language in “chunks.” Because of this, the SFL approach to L2 focuses on language teaching that is both practice- and meaning-oriented in that “learning ‘the grammar’ of a language is not about learning to adhere to rules, but learning to turn experience and human existence into meaning by using the resources that the grammar of a particular language makes available” (Byrnes Reference Byrnes2009: 5). The way to do this is to increase language learners’ awareness of the relationships between form and meaning in different social contexts. Like the SFL approach, a usage-based approach to second language learning is less concerned with language structure than with the ways in which language use is anchored to the learner/speaker’s experience with the L2. From this viewpoint, grammar is seen as the “cognitive organization of one’s experience with language” in which language users actively categorize their linguistic environment, creating a “vast network of phonological, semantic and pragmatic associations,” which in turn is shaped and reshaped by repetition (i.e. the frequency of use) of specific linguistic elements within that environment (Bybee Reference Bybee, Robinson and Ellis2008: 216–7).

The SFL and usage-based approaches are very similar to the one taken by linguists such as Diane Larsen-Freeman, who apply complexity theory to second language acquisition. Larsen-Freeman (Reference Larsen-Freeman2003) sees the teaching of an L2 as “teaching grammering,” a phrase that emphasizes the action of language learning and underscores her belief that L2 teaching should be organic and holistic (as opposed to linear). What this means is that a particular grammatical construction is not the focus of the “Lesson on Day X,” but is instead present in a series of lessons, allowing the students to experience the grammar of the language they are learning, rather than memorize decontextualized grammar rules. Larsen-Freeman describes teaching her students to “learn to look” at their L2. For a particularly challenging structure, for example, she asks her students to think about “the company that the structure keeps,” giving them the same kind of meta-awareness that SFL advocates.

Another way to characterize the complexity approach is to talk about the ecology of learning – to see language learning as a whole ecosystem of interactants that may include a teacher, other students, a textbook, native speakers of the language, and classroom or non-classroom contexts. This kind of ecological perspective is espoused by linguists like Claire Kramsch who, much like Mufwene on language evolution, uses the idea of language ecology as a way to talk about humans’ relationship with language:

All of the non-generative approaches stress that learning (and teaching) a second language should be grounded in an embodied or situated experience of that language. Victor Kroskrity once explained that L2 learners are not mere “hosts for language,” and what we take this to mean is that students of a second language are actively involved as agents of meaning-making, drawing on their repeated experiences with an L2 in specific environments (or situations of use) to construct for themselves another language.