Saturday, 24 October 2009

Second Language Acquisition and Memory

Evolutionary approaches have recently been adopted to try to explain the differences in adult and child second language acquisition (SLA). Hagen (2008) uses Paradis's (2004) theory of implicit and explicit learning to argue that adult SLA was selected out because it was not useful to our early ancestors. However, this order of evolution seems unlikely and a more parsimonious and longer-term explanation can be achieved by looking at the differing social pressures of adults and children.

Children acquire languages rapidly, effortlessly, implicitly and fluently. Adults, on the other hand, find second language acquisition laborious, difficult, and often do not achieve native fluency. Hagen (2008) addresses this difference from an evolutionary perspective. Firstly, Hagen shows that the difference between adult and child SLA is physical. Evidence comes from studies of SLA, and anthropology, which shows that the language faculty co-evolved with the physical evolution of our species. This part is well argued, and contains a diverse sources which, rightly, place the problem of the critical period as an evolutionary adaptation to the environmental conditions of the ancestors of humans.

Hagen cites Paradis’ (2004) theory of implicit and explicit knowledge being linked to procedural and declarative memory respectively. Procedural memory stores knowledge of unconscious, automatic skills such as walking. Declarative memory stores knowledge that can be consciously accessed, but with greater difficulty. The ability to incorporate knowledge into procedural memory atrophies in adults, which explains differences in loci of adult second languages in the brain. Hagen’s explanation for Bilingualism, however, focuses on relatively recent, certainly post-linguistic, environmental pressures. The main section of his argument considers the pros and cons of the ability for adult SLA. For the negative effects, Hagen cites Dyson’s (1979) argument that language diversity evolved to establish and maintain cultural differences between competing groups. This resonates with Accommodation theory. The benefits are mainly restricted to the ability for trade.

Hagen argues, however, that adult SLA would generally not be advantageous in pre-agricultural society (low populations dispersed widely). He links the critical period with studies of predatory animals brought up in captivity who don’t develop predatory behavior. As Hagen puts it, “If you don’t acquire it early, you might as well not acquire it at all” (p. 58). Hagen concludes that, since violent conquest of another culture yields greater rewards than cooperation during lean times, there was no pressure to develop adult SLA. On the other hand, the ability to socialize is vital to survival in a human society, so child acquisition is selected for.

Hagen’s explanation comes down to a discussion of the nature of man. Hagen argues that evolution, and evidence from history, favours an aggressive nature rather than an egalitarian one. Learning an enemy’s language is more costly and less beneficial than conquering them.

There are problems with Hagen’s approach, however. Although Hagen stresses that the differences between adult and child acquisition are physical, the explanation relies on sociological factors, originating less than 80,000 years ago. That is, although human language acquisition behaviour seems to “fit rather nicely into the mosaic of evolutionary theory”, Hagen has not explained the pressures that would cause the structural changes in the brain that would lead to differentiation in the first place. In the first half of the paper, Hagen suggests that adults lose their ability to acquire languages for two reasons: Firstly due to general cognitive aging. Secondly, because the language faculty is built upon the same instincts that lead to any young animal learning basic survival skills such as walking, and has therefore never been useful at later stages.

Part of the problem is that Hagen seems to be suggesting that SLA was selected against in adults (noted by Hirchfield, 2008, with a rebuttal by Hagen). That is, there was a point when language learning was equal for adults and children, and the adult ability eroded by drift. If this original adult ability was identical to the child ability (that is, there was no change over a lifetime), this seems to run against the idea that child and adult acquisition are physically different. An alternative situation is where adult and child learning was originally different, but achieved the same goal. This seems an inefficient solution.

Rather than a loss of an ability, perhaps the structural differences between adult and child acquisition reflect adaptations to different tasks. MacWhinney (2007, p. 2) argues the following: Children need to learn language for socialising in order to obtain protection and to inherit knowledge about how to survive. Adults face a different task – they must socialise in order to secure mates and provide for their children. Whereas there would be a pressure for children to acquire basic knowledge about the world (how to walk, red berries are poisonous) and how to communicate with their group, there would be a pressure on adults to keep track of social relationships (e.g., who is sleeping with who) and communicate about resources (e.g., where the food is, organisation of hunting).

I argue that this difference in social pressures is supported by two different kinds of memory: semantic memory, which encodes facts separated from context (e.g. the sky is blue), and episodic memory, which also encodes temporal information (e.g. it rained yesterday). Semantic memory has been argued to be related to procedural memory rather than declarative memory (Hurford, Flaherty & Argyropoulos, 2007). At any rate, it is certainly evolutionarily older than episodic memory, if not unique to humans, and develops later in childhood.

Adults can use the contextual informtion of episodic memory to keep track of complex social interactions. For example, it may be a bad idea to have a general sense that ‘X is a nice guy’, if X is nice only two thirds of the time. However, rather than an erosion of procedural/semantic memory in adults, it is the development of declarative/episodic memory that may be overriding procedural/semantic memory during adult second language acquisition.

I propose that the balance between using semantic and episodic memory was selected for so that semantic memory was favoured in childhood and episodic memory was favoured in adulthood. To test this assumption, a model was run to examine the ratios of semantic and episodic memory in a population of agents who age and compete for mating opportunities by assessing the social interactions of others.

The Model
In this model, agents can interact and store knowledge about events in either a procedural or declarative mode. Adults must keep track of how others behave in order to maximise their chances of offspring. It is hypothesised that this pressure will cause adults to use their declarative memory more than children.

Each agent is either male or female and has an age and an energy level. Each also has a set of genes that define a linear model for making decisions. That is, each variable involved in a decision can be modified by a slope and intercept with which to modify it. Agents receive energy at each time step (i.e. they eat) as long as they are above a certain age (they are initially dependent on others). Agents become capable of reproduction after a certain age with a small random variation. Agents can die either because they run out of energy or from old age.

Each agent has a declarative and procedural memory store which can store pieces of knowledge. A piece of knowledge can be either reflexive – another agent did something to me – or transitive – two other agents interacted with each other. The piece of knowledge also contains an attitude – whether the interaction was good or bad. For instance, a piece of knowledge can represent “Agent X gave me food”, or “Agent X and Agent Y mated”.

The declarative/episodic store is a LILO stack with a limited capacity. New pieces of knowledge are added to the start of the stack. This models episodic memory (e.g. “I gave Agent X some food earlier, but then they refused to mate with me”). The procedural/semantic memory stack works differently. Only one attitude can be stored with each interacting agent or pair of agents. New pieces of knowledge can gradually alter the attitude to another agent or interaction. When the capacity is reached, the least strongly held opinion is deleted. This models semantic memory (e.g. “Agent X is friendly”).

The Model
At each time step, the agent decides either to:
• Mate, based on age, energy and puberty
• Give food, based on age and energy
• Teach another agent, based on age and energy
• Observe another event, based on age and energy
There are a number of time steps in a year.

Agents can choose to give food to another agent. In doing so, their own energy depletes and the receiver’s energy increases. The receiver learns that the giver gave them food (a positive attitude).

Agents can choose to transfer a random piece of knowledge to another agent. The receiver also learns that the teacher chose to teach them something.

Agents can choose to learn any event that occurred in the current time step.

If an agent decides to mate, they choose another agent of the opposite sex to mate with, based on their knowledge. For each agent they know about, an overall attitude score is calculated based on the sum of the attitudes attached to each piece of knowledge and a weighting gene. Once the potential mate is chosen, they attempt to mate. The reciever may choose to accept or decline, based on their own attitudes towards the initiator. If they accept, then both agents expend energy and, if the receiver has not expired, they become pregnant and a new agent is added at the end of the year. An agent cannot become pregnant with more than one child at a time.

Sexual reproduction is modeled with genes chosen randomly from two parents of different sexes. Each gene represents either a slope or intercept for a decision variable. There is a small chance of drift.

Model initiation
The model was initiated with 50 agents with random ages, energies, sexes and gene settings.

Preliminary Results
The graph below shows some preliminary results from one run of the model. The Yellow line shows the population size, which slowly increases (on right scale). The solid lines represent the percentage of the population choosing various activities (Red = Mate, Blue = Observe, Green = Teach, Brown = Give). The dotted lines show the average proportion of choices to use episodic memory to store new knowledge compared to semantic memory, for adults (black) and children (brown).

The graph shows that children's average use of episodic memory is lower than adults'. This suggests that, over time, agents evolve to use mainly procedural memory as children, but then start using more episodic knowledge when they become adults. The robustness of these results need to be examined further.


The preliminary results agree with the hypothesis that the ability to store temporal social information using episodic memory during adulthood was selected. If the use of episodic memory conflicts or draws resources from semantic memory, this may explain the difficulty adults have in learning a second language.

This theory suggests a change over a much longer period than Hagen's and is also based on an adaptive advantage for declarative/episodic memory which is more likely to be selected than a neutral effect of procedural/semantic memory in adults.

The theory may be important in explaining linguistic diversity also. As Nettle (1999, p26) points out (under his 'Neutral model' of linguistic variation), in order for languages to change so as to become mutually unintelligible, two mechanisms are required. First is a source of variation. This is easy to find in both speech production by adults and learning patterns in children. The second type of mechanisms are amplifiers of variation - factors that maintain differences between two populations. This includes geographical isolation, but as Nettle points out, most societies have considerable levels of exposure to the languages of other cultures. This last point is very strong, but perhaps overlooks who is participating in cross-linguistic activities. If adults are the main interactors (trade, marriage, politics, war), then linguistic diversity may be maintained because they are poor at acquiring novel linguistic variants. That is, variation may be maintained if only the members who are acquiring the language (children) are isolated from other linguistic variants.

Since there is a huge amount of linguistic variation in the world, the selection process described in the model above should have occurred very early in language evolution, if not pre-linguistically.

L. Kirk Hagen (2008). The Bilingual Brain: Human Evolution and Second Language Acquisition Evolutionary Psychology, 6, 43-63


  1. Thanks for the review of Hagen's book. I was curious as to how integrated it was (being an SLA title) to the rest of what's been going in Language Evolution over the last decade. I'll pick it up in any case since I'm interested in pursuing any links between SLA / Bilingualism and the origins of language.

  2. I think second language will always five you a priority! tells about pros and cons of bilingualism!

  3. The more languages one knows, the more opportunities one have, Egnlish is international and therefore compulsory to know properly for everyone but every other language gives even more advantages than English alone