In many ways this is a special case of the nature vs. nurture argument.
The books that I have read (The language instinct, The Symbolic Species et all), put forward the theory (which makes sense to me) that the ability to acquire language, or rather the raging hunger to inhale large quantities of verbiage, is innate in the human child's developmental process.
However, the language that is learned, be it English, Chinese, Tagalog or sign language, depends upon the environment in which the child grows up.
Language is not just another skill that is just learned in the same way that a general purpose computer can run any program. Our bodies, particularly throats and brains, are highly adapted and specialised to the tasks of language.
Steven Pinker points out that the adaptation goes the other way too: human language is by definition the words and rules that have survived being learned by countless generations of children, and therefore whatever is left must be suited to being learned by human children. Thus our languages do not vary without limit - Each language feature fits into one of relatively small number of categories because we have the mental equipment to be able to learn certain kinds of languages.
You may well ask: Do birds learn to fly, or is it innate?
taschenrechner answers the rhetorical question: Yes we have to learn to use the brains and throats that are designed for language, just as birds have to learn to use their wings thought they are designed only for flight. In both cases, the skill is both learned (baby birds can't fly, having neither the skills nor the strength) and innate (would not get far without those wings).
With regards Cognis' excellent write-up, I think it is fair to say that something that is "entirely learned" but learned over timescales over which evolution plays a significant role, so that special structures for the ability have become part of the body, has in some sense become part innate. Cognis is also correct that the simplest explanation is that we started speaking with simple calls, like monkeys or birds, and that complexity of speech co-evolved with better speaking apparatus to enable that speech.
A five-year-old's command of language, if it's just another complex, patterned input for babies and small children to take on board, is a pretty good trick. As unlikely as a child playing Handel's Messiah on the piano after hearing a couple of CDs of unrelated music. And this is almost every child, whatever their level of general intelligence. Put like that, the mysterious language instinct is the least unlikely explanation.
The rationale behind the associative approach to language acquisition is that the language environment an infant finds itself in is sufficiently structured to allow that infant to detect and learn the regularities that underpin competent use of the language. Learning is made through association, hence, language is seen as an associative process. This sort of approach is usually (but not always) a domain-general one, where the acquisition of language is seen as very much the same as any other learning process. As such, language may be seen as being closely related to other cognitive development. There is an emphasis in the associative account of individual differences having an important role to play. A key demonstration of this is that different environments can lead to the acquisition of different knowledge. Infants placed in any language environment acquire that language just as easily as if they were placed in another language environment. So, an infant raised by an English speaking family will acquire English just as easily if the infant would acquire Japanese if raised by a Japanese speaking family. Furthermore, the plasticity of development seems to suggest that there is a learning process taking place. It is possible for a child who has suffered brain damage affecting language to recover and develop full competence. This is due to language being a steady state, rather than the changing brain of the infant.
The opposing view to language being learned totally from the environment is the rationalist view that language is too complicated for this sort of learning. The approach rooted in nativism prefers to think of the brain as the steady state upon which language is acquired, rather than the environment. Key figures in this school of thought are Chomsky and Pinker, who see the infant as bringing significant innate knowledge to the language learning situation that allows them to acquire language. The key point of this idea is that there just is not enough information in the environment in order for the child to learn everything it needs to. This so-called "poverty of the stimulus" led to the idea that there are innate mechanisms, or innate knowledge, that provides the infant with the "head start" it needs to get the ball rolling. As such, the linguistic environment is used to fill in the gaps in the infant's innate knowledge - for example, telling the infant what the language of its community is. Ideas such as Chomsky's Language Acquisition Device (LAD) has been a major contribution to this school of thought. The LAD was postulated to be the innate mechanism that allows a child to latch onto the grammatical principles of the language that is to be acquired. Evidence for the LAD comes from that fact that all languages share a similar underlying "deep structure" of grammar - universal grammar as Chomsky called it. The LAD, along with other nativist principles, led to the idea that there are "modules" in the brain that are specified for language. The nativist view does not embrace the same domain-general view that empiricists do, favouring a domain-specific one, whereby language is seen as somewhat of a special process.
The nativist approach leads very much to rationalist approaches to language acquisition, leading to symbolic accounts and models of linguistic processes with systems of symbols and rules. On the other hand, associative approaches rely much more on things like Hebbian learning, and things like connectionist models with no explicit rules represented (see; There are no rules in language, only regularities).
The question is, perhaps, how do we decide between these two contrasting accounts? One linguistic developmental phenomenon that may provide an answer to this question is that of the categorical perception of phonemes. It has been found that infants as young as it is possible to test (maybe as young as a few days) demonstrate the ability to discriminate between different phonemes based on the same classifications that adults use. Using the High Amplitude Sucking Technique (HAS), Eimas et al. (1971) showed that infants could discriminate between /ba/ and /pa/, which differ only in voice onset time (VOT). It was found that infants only a few months old could detect the difference between the two phonemes in the difference in VOT crossed that categorisation boundary between the two, but did not differentiate between phonemes when the same size difference in VOT was tested within a phonemic category. Eimas also found that young infants displayed categorical perception in a liquid continuum (Eimas, 1975) and in a place of articulation continuum (Eimas, 1974). It has been confirmed that infants do this across all such continua. The interesting thing about this is that adults do not display the same behaviour. It seems that the adult language user has "tuned into" the phonetic contrasts that serve to differentiate words in their own language, at the neglect of being able to differentiate between contrasts that do not act to distinguish one word from another in their native language, but which may do so in other languages. Infants do not show such specificity. For example, Werker and Tees (1984) showed that English infants can differentiate the retroflex/dental contrast between /T/ and /t/ in Hindi, and the velar/uvelar contrast between /k'i/ and /q'i/ in the North American language of Nthlakapmx, but (first language) English adults cannot. It is also well known that Japanese adults have difficulty with the /l/-/r/ distinction present in English but not in Japanese, whereas Japanese babies do not have this difficulty. Kuhl (1983) has shown using the conditioned head turn procedure that even at six months of age, infants treat acoustically different tokens of the same phoneme, differing in intonation changes or speaker identity for instance, as equivalent.
Categorical perception of phonemes seems, at least initially, to be evidence certainly in favour of the rationalist (i.e. nativist) approach to language acquisition. The fact that this phenomenon occurs so very early in an infant's life might suggest that it is very unlikely that the infant could have learned to categorise phonemes. So, the ability must be innate. However, Kuhl and Miller (1978) may call into question those nativists that hold that language acquisition is a totally domain-specific skill by finding that chinchillas demonstrate categorical perception of phonemes with almost identical boundaries to humans. They argue that it is something about the nature of such auditory signals themselves that allow speech sounds to form "natural categories". This would seem to be evident in the fact that chinchillas have hearing very similar to humans in the respect of their auditory curve (Miller, 1970), and detection thresholds for complex sounds (Luz, 1969). This evidence, and evidence that categorical perception occurs for non-speech sounds too, seems to hint that language may not be so special after all. Furthermore, Kuhl and Miller's demonstration of categorical perception in animals (it has been demonstrated in other animals, such as cats) seriously undermines the idea that infants bring innate phonetic knowledge to language acquisition (hence their ability to categorically perceive phonemes) because animals surely have no innate linguistic knowledge, as they have no language.
One might be tempted to take a "tabula rasa" approach in light of this evidence. However, this position is untenable due to the fact that the infant would get nowhere as they need some sort of innate mechanism in order to actually learn the structure of language, as one does not "get structure for nothing". By the same token, a view of complete language innateness is also impossible, as infants are clearly not born knowing a language, and can learn any language as easily as the next. It is hardly surprising, then, that nobody holds either of these extreme views. Instead there is agreement that there is some innateness to language acquisition and some learning too. Nature and nurture are so interactive that one cannot be without the other. The question may not be whether nativism or empiricism is right or wrong, but rather how much innate ability is required for language acquisition, and how much innate ability evolution may have endowed humans with.
Recently, there seems to be a more satisfactory compromise made between nativism and empiricism through the genetic evolution of a connectionist network by Nakisa and Plunkett (1998) which was capable of detecting speech features after only two minutes of exposure to English (or any of the diverse range of languages tested). This model was capable of categorical perception similar to human infants very early in life. A series of randomly connected networks were trained on speech input and those that were most successful in representing different phonemes differently, and the same phonemes similarly, were deemed to have better "fitness" and were favoured when it came to "breeding" two parent networks together to produce a member of the next generation of networks. After 100 generations, networks were able to distinguish phonemes like a human infant. Not only is this a good demonstration of the value of genetic algorithms in discovering solutions to computational problems, but it shows how an innately given and genetically determined starting mechanism can interact with environmental input to produce very rapid learning and the patterns of data observed in newborns. This account of categorical perception shows how innate factors are needed to provide a foundation on which associative processes begin to actually acquire a language. However, the purer rationalist may well take issue with the Nakisa and Plunkett model due to its lack of explicit rules and symbols, since this is what the pure rationalist is drawn towards. Perhaps such a person may criticise the model due to the fact that its evolution was based on training input that was fully developed speech, but this is backwards to how language must surely have evolved. Humans must have developed categorical perception before language as categorical perception is a necessary prerequisite to being able to differentiate between words.
But this caveat in the Nakisa and Plunkett model, if it is a caveat at all, does not dismiss the view that language acquisition seems to be based on innate computational properties being used for a task the environment places on a person. The brain is basically made up a large number of different systems of neurons wired together differently - evident in the number of different Brodmann Areas (as defined by different types of cell organisation) found in the brain. It has been proposed that when a set of such heterogeneous computational systems are exposed to an environment, different computational systems are recruited to certain tasks on the basis of their inherent ability to perform them. Therefore, the brain becomes specialised due to different environmental challenges becoming internalised. As such, knowledge is not innate, rather neural networks that make up the brain begin ignorant, but their innate biases allow them to be recruited to a task they are suited for. This view is a compromise between the associative account and the rationalist account. As with most dichotomous views in psychology, the current thought on the issue seems to have found its way to a compromise.
References:
printable version chaos
Everything2 Help
