Information Processing and Connectionism in Language Acquisition
Recent Psychological Theories
Information Processing
Cognitive psychologists working on an information processing model or theory of human learning and performance tend to see Second Language Acquisition (SLA) as “the building up of knowledge systems that can eventually be called on automatically for speaking and understanding.”
Information Processing
This psychological theory compares the human brain to a computer. It includes the idea that the brain has a very large capacity to store information in the long term but a more limited capacity for information requiring our full attention. It is also based on the following idea: After a certain amount of practice, things, i.e., stimuli, that at first required attention eventually become automatic. Thus, they leave more attention available for focus on other stimuli.
According to the ‘information processing’ model, at first, learners have to focus their attention on whatever aspect of the L2 they are trying to understand or produce. This is based on the assumption that there is a limit to the amount of information that a person can pay attention to at one time.
For example, at the earliest stages (i.e., basic levels) of SLA, a learner pays attention to only the main words in a message. She is not able to notice the grammatical morphemes attached to some of those words. Gradually, through experience and practice, learners become able to use certain parts of their knowledge so quickly and automatically that they are not even aware that they are doing it. This allows the learner to focus on other aspects of the L2. Gradually, these aspects also become automatic.
The performance that becomes automatic originates from ‘intentional‘ learning, for example in formal study, but this is not always the case. Anything that uses up our mental ‘processing space’, even if we are not aware of it or attending to it ‘on purpose’, is a possible source of information or skills that can eventually be available automatically, if there has been enough ‘practice’. In this context, ‘practice’ is not a mechanical activity, but one that involves effort – physical and/or intellectual – on the part of the learner.
Schmidt (1990) emphasizes the role of ‘noticing‘ in SLA. He argues that everything we come to know about L2 is first ‘noticed‘ consciously. This contrasts sharply with Krashen’s views: like the cognitive psychologists’ view, Schmidt does not assume that there is a difference between acquisition and learning.
Besides the development of automaticity through practice, some psychologists suggest that there are changes in skill and knowledge that are due to ‘restructuring‘.
The notion of ‘restructuring’ explains that sometimes things that we know and use automatically may not be explainable in terms of a gradual build-up of ‘automaticity through practice‘.
‘Automatic activities’ seem rather to be based on the interaction of knowledge that we already have, or on the acquisition of new knowledge that – without extensive practice – in some way fits into an existing system and transforms it, that is, it restructures it. This restructuring may lead to sudden bursts of progress for the learner, but it can also sometimes lead to apparent backsliding when a systematic aspect of language learning incorporates too much or incorporates the wrong things.
For example, when a learner finally masters the use of the regular –ed inflection marking past tense, irregular verbs that had been ‘practiced’ correctly may be affected: For example, after months of producing: ‘I saw a film yesterday‘ the learner may now utter ‘I sawed a film yesterday‘ or even ‘I seed a film yesterday‘. Clearly, this is ‘overapplying’ the general rule for the past tense form of regular verbs.
Connectionism
Connectionism, or ‘neuron-like’ computing, developed out of attempts to understand how the human brain works at the neural level and, in particular, how people learn and remember.
In 1943, the neurophysiologist Warren McCulloch of the University of Illinois and the mathematician Walter Pitts of the University of Chicago published an influential treatise on neural nets and automatons. According to their theory:
- Each neuron in the brain is a simple digital processor
- The brain as a whole is a form of computing machine
As a learning theory, connectionism views language as a ‘complex system of units that become interconnected in the mind as they are encountered together’. The more often units are heard, the more likely it is that the presence of one will lead to the activation of the other.
Unlike innatists, connectionists do not find it necessary to hypothesize the existence of a neurological model that is designed for language acquisition alone. Like most cognitive psychologists, connectionists attribute greater importance to the role of the (social) environment than to any innate knowledge in the learner. They argue that what is innate in the learner is simply the ability to learn, not any specifically linguistic faculty, like Universal Grammar (UG) or Language Acquisition Device (LAD).
Connectionists argue that learners gradually build up their knowledge of language through exposure to thousands of instances of the linguistic features that they eventually learn. Thus, while innatists see the language input in the environment as a ‘trigger‘ to activate innate knowledge, connectionists see the input as the principal source of linguistic knowledge.
After hearing language features in specific situational or linguistic contexts over and over again, learners develop stronger and stronger mental or neurological ‘connections‘ between these elements. Eventually, the presence of one situational or linguistic element will activate the other(s) in the learner’s mind. These connections may be very strong because the elements have occurred together very frequently, or they may be relatively weaker because there have been fewer opportunities to experience them together.
For example, learners might get the subject-verb agreement correct, not because they know a rule but because they have heard utterances as: ‘I say‘ and ‘he says‘ so often that each subject pronoun activates the correct verb form.
Connectionists have shown that a learning mechanism, simulated by a computer program, can not only ‘learn‘ what it hears but can also generalize, even to the point of making overgeneralization errors. Connectionist studies have so far dealt almost exclusively with the acquisition of vocabulary and grammatical morphemes, that is, aspects of the language that even innatists will grant may be acquired largely through memorization and simple generalization.
How this model of cumulative learning can lead to knowledge of complex syntactic structures is a question that is currently under investigation.