Language Emergence
Brian MacWhinney
1 Stipulationism
In the 1950s, researchers thought that children learn language through imitation, guided by principles of shaping and reinforcement (Skinner, 1957). By the end of the decade, the new information-processing psychology (Newell & Simon, 1972) and the rapidly growing theory of generative grammar (Chomsky & Halle, 1968) were challenging this view. These new paradigms expressed the complexities of human behavior not as links between habits, but as complex systems of interlocking rules. The power achieved by these systems relied on the ability of the modeler to stipulate the right set of rules in terms of their elements, combinations, and patterns of rule orderings. The successes of these stipulative systems can be attributed to the precision of their formulation and the expressiveness of the formal production system language on which they relied (Hopcroft & Ullman, 1979). Stipulative rule systems promoted the articulation of enormous cognitive architectures of seemingly impossible complexity (MacWhinney, 1994). As these systems grew in complexity, testing the empirical grounding of their individual components became increasingly impossible. Through its descriptive successes, stipulationism ended up sewing the seeds of its own conceptual destruction.
In the 1980s, the rise of connectionism (Rumelhart & McClelland, 1987) provided an alternative to stipulationism. Neural networks viewed children as learning cues, rather than rules (MacWhinney, Leinbach, Taraban, & McDonald, 1989). In the 1990s, researchers began toexplore still other alternatives to rule systems, including ynamics systems theory (Thelen &Smith, 1994), optimality theory (Tesar & Smolensky, 2000), and biological models of neural plasticity (Elman, 1999). Formal linguistic theory also began to move away from stipulationism,attempting to extract a minimal set of principles from which broader syntactic patterns could emerge (Chomsky, 1995).
2
2 Emergentism
Historically, emergentism began as a reaction against stipulationism. However, it is wrong to think that any model of language development that does not specifically stipulate particular rules or hard-wired modules is emergentist. If we simply used the absence of stipulated rules as our criterion, we would allow even the most half-baked, inarticulate idea to count as an emergentist solution. Clearly, something more is needed for an emergentist account. These additional ingredients are mechanism and generativity. Just like the “big, mean” rule systems of the 1970s,an emergentist ccount must provide a specific mechanism that works to generate the observed behavioral patterns. In an emergentist account, generativity emerges not fromstipulated rules, but
from the interaction of general mechanisms. Let us consider some examples.If you spend time watching the checkout lines at a supermarket, you will find that the number of people queued up in each line stays roughly the same. There are rarely six people in one line and two in the next. There is no socially articulated rule governing this pattern. Instead, the uniformity of this simple social “structure” emerges from other basic facts about the goals and behavior of shoppers and supermarket managers.
Honeybees are certainly no smarter than shoppers. However, working together, bees are able to construct an even more complex structure. When a bee returns to the hive after collecting pollen, she deposits a drop of wax-coated honey. Each of these honey balls is round with approximately the same size. As these balls get packed together, they take on the familiar hexagonal shape that we see in the honeycomb. There is no gene in the bee that codes for hexagonality in the honeycomb, nor any overt communication regarding the shaping of the cells of the honeycomb.
Rather, this hexagonal form is an emergent consequence of the application of packing rules to acollection of honey balls of roughly the same size.
Nature abounds with examples of emergence. The outlines of beaches emerge from interactions atoms can pack into sheets. Weather patterns like the Jet Stream or El Ni emerge from interactions between the rotation of the earth, solar radiation, and the shapes of the ocean bodies. Biological patterns emerge in very similar ways. For example, the shapes of the spots on aleopard or the stripes on a tiger emerge from the timing of the expression of a pair of competing genes expressing color as they operate across the developing leopard or tiger embryo (Murray,1988). No single gene directly controls these patterns. Rather, the stripes emerge from the interactions of the genes on the physical surface of the embryo. The shape of the brain is very much the same. For example, Miller, Keller, and Stryker (1989) have shown that the ocular dominance columns described by Hubel and Weisel (1963) in their Nobel-prize-winning work may emerge as a solution to a competition between projections from the different optic areas during synaptogenesis in striate cortex.
Emergentist accounts of brain development provide useful ways of understanding the forces that lead to neuronal plasticity, as well as neuronal commitment. For example, Elman (1999) shows how the learning of linguistic categories emerges from the interaction of constraints on the activation of sheets of neuronal tissue. Similarly, Quartz and Sejnowski (1997) have shown that plasticity may also involve the growth of new patterns of connectivity. On the macro level, recent fMRI work (Booth et al., 1999) has shown how children with early brain lesions use a variety of alternative developmental pathways to preserve language functioning.
These emergentist formulations of the neural grounding of cognition allow us to consider new ways of dealing with age-old confrontation between nativism and empiricism. We teach students that the opposition between nativism and empiricism is the fundamental issue in developmental psychology. However, what students end up learning is that everything in human development involves some unspecified interaction between nature and nurture. Because we often fail to explain how this interaction occurs, students end up being confused about the underpinnings of
the science of human development. Emergentism addresses this problem directly. It replaces the traditional opposition between nativism and empiricism with a new conceptual framework, explicitly designed to account in mechanistic terms for interactions between biological and environmental processes. The goal of
emergentism is to replace accounts based on stipulations with accounts in which structures emerge from the interaction of known processes. However, it must do this without sacrificing mechanism and generativity.
We must temper this strong formulation of the emergentist program with practical reality. The primitive state of our understanding of basic neurological and developmental processes means that models often still have to rely on stipulation to characterize structures that we do not yet fully understand. For example, a model of the effects of auditory processing deficits may need to include a hand-wired representation of information passed on to language processing from the auditory cortex. This type of stipulation regarding structures that are not at the core of a given
model is not a theoretical commitment. Rather, it reflects the primitive nature of our current modeling techniques.
3 Five Time Frames
Emergentism does not imply a radical rejection of either nativism or empiricism. On the contrary, emergentism views nativist and empiricist formulations as the partial and preliminary components of a more complete account. The traditional contrast between nativism and empiricism revolves around the fact that they describe developmental processes that operate across different timeframes. When we discuss some ability in the infant, we often ask ourselves, “Is this ability innate or learned?” There is nothing wrong with this question, as long as we realize that it really a question about the timeframe involved in the delineation of the relevant emergent processes.
We can distinguish five separate timeframes for emergent processes and structures (Lorenz,1958).
