Stephen van Vlack

Sookmyung Women`s University

Graduate School of TESOL

Human Learning and Cognition

Spring 2006

Week 4 Answers - Terry, Chapter 3 & Lamb, Chapters 4 and 5

1. What is classical conditioning and why is this important?

Classical conditioning, otherwise known as Pavlovian conditioning, is when an organism forms a simple association between two previously unrelated stimuli. One of the stimulus is important to the organism and is called an unconditioned stimulus. The other stimulus is a conditioned stimulus. The association made between tho two must be learned as there is a demonstrable change in behavior as a result of the association having been made. As the result of the association the organism will have transferred the natural responses to the unconditioned stimulus to the conditioned stimulus even though prior to the their dual and temporally controlled exposure they were not at all related. In a nutshell this is what classical conditioning is.

Although for the purposes of trying to control the process and/or concepts of morality classical conditioning experiments are usually carried out in a laboratory and with organisms other than humans, it is possible to see how this can be applied to the outside world and to the human world in general. Conditioning is simple type of learning which takes habituation just one baby step further. In conditioning we are taking what was learned through habituation and extending on to another stimulus which just so happens to co-occur in the world. Such type of learning is obviously largely responsible for our uncanny ability to survive for it means that we can extend our behaviors beyond a simple flight response generated by habituation. It may also be the very underpinning of our ability for language.

The very defining feature of language is its arbitrariness as proposed by de Saussure; that is the arbitrariness of the actual structures (but never the meaning). We would then wonder, if there is no logical reason why some structural units should be expected to be found adjacent to others (to think so would be ludicrous) then how did they come to co-occur like they do? This is a question of linguistic evolution and unfortunately can not concern us here. What does concern us here, however, is the question of how a child can learn these patterns of association. There must be a type of learning which supports the rather quick learning of language. Conditioning might just be the missing link.

Think about context. Context is the clue to all learning and those that can see it and differentiate its parts are the good learners.

2. How might classical conditioning-type studies be carried out on language?

There are several possible studies or experiments that we ourselves could try to carry out on language acquisition in the classroom.

Feature-hopping

The basic idea in the feature hopping sequence is that all words are composed of sets of features. Some of these features are direct attributes of the word while others are less direct attributes and take the form of links to other different words and concepts. Now, when two stimulus (and in this case what we are really addressing here is the language and functions) are linked and a new stimulus comes in the two stimulus are associated. It is in this association that features hop or are mapped from one stimulus onto the other. It is therefore possible for us as language teachers who are very interested in language learning to try to devise a test to see how features of one linguistic stimulus are mapped onto another linguistic stimulus. For us as second-language teachers this of course would be done in relation to feature mapping across the two languages. We would assume that the mapping would occur from the dominant language to the less dominant language at least in the initial stages. This can become simply by having lexical items co-occur and then test to see if the features of one map onto the other. So, if by explaining a new or unfamiliar English word by translating it into Korean, we want to see if the features of the Korean word are immediately carried over or hop onto the English form.

Co-occurrence

The idea studying co-occurrence comes from basic observations that in order for two stimulus to be associated they need to co-occur and not just once but probably several times depending on the type of stimulus, the specific context in which they occur, as well as the intervals between the co-occurrences. Well, one thing that we can do as a way of testing is to have some lexical items co--occur and then to test and see if the students or subjects after a certain period of time also use these new combinations in their own language use. This is simple learning of patterns through association as a result of co-occurrence.

Feature generalization

The underlying idea of behind feature generalization is that a stimulus (and that includes words/phrases/sentences) are composed of features. We also know that associations can actually be extended from the original associated link to other things or concepts which have similar features. So, for example if somebody associates the actual animal dog with the lexical item dog than the attributes of the animal dog (at first the specific features of a particular dog, let`s say a chihuahua) will he mapped onto the lexical item. When the learner encounters another dog, such as a dachshund or a St. Bernard and learns than they are all dogs then then we would expect them to map the same types of features of dog onto these new dogs based on the fact that there seem to be common features. Over time though the features that make up dog will change and be generalized to accommodate all the animals that are in fact dogs while at the same time excluding animals that aren`t dogs like coyotes or wolves. This of course could be turned into an experiment in the classroom.

3. What is extinction and how does it work?

Extinction is the eradication of the conditioned response due to breaking the co-occurrence of the unconditioned stimulus and the conditioned stimulus. An interesting thing about extinction (and this gives us valuable insight on learning and forgetting in general) is that the effects of extinction are only temporary so long as the pattern of co-occurrence is reinstated at a later date. This confirms what we said last week about the theory that once learned things are never actually forgotten though they might go dormant for a period of time. In any case, the potential for recovery is always there - maybe. What probably actually happens in extinction situation is that the costs of recovering the information simply becomes comes too high. This is based on the idea of firing. We know that firing is a process in the brain by which electricity is entered into a neuron and is disseminated to its synapses. Each synapse requires a different amount of electricity to fire it, or make it work. The amount of electricity required is called the threshold level. As a synapse is used more and more often its threshold level will go down. So, to a large extent the threshold levels of a neuron`s synapses are determined by their frequency. If synapses are not used frequently their threshold levels are going to go up and the longer they're not used to more they'll go up. Once they reach a certain level it becomes hardly unlikely that these synapses will be fired because they simply cost too much energy. The brain due to its plasticity will simply find another way around and threshold levels will therefore continue to go up in what seems to be a rather vicious cycle. This doesn't mean we can't fire a synapse that is required or coded for a certain association, it just means that we avoid doing it because it's not efficient. If we need at some point to actually do that we can but it is going to cost too much energy.

4. How does contiguity affect conditioning?

Contiguity refers to the timing of the conditioning. This relates to the intervals between the presentation of the two stimuli as well as the order of the stimuli. In looking at contiguity the biggest lesson we need to take from this is that a forward sequence is the most effective. This probably because a forward sequence would allow the organism to effectively prepare for the coming unconditioned stimulus. It is also very important to remember that contiguity as a variable in conditioning is very much reliant on the task itself, the species involved in the experiment, and the type of response expected. All this ties into brain differences among species and among the different functions as related to the type of response expected. If the stimulus is visual in nature then the neurological structure of the optic nerve and how this is connected to other parts of the brain becomes an important factor. Also, some experiences are more intense than others, such as food poisoning, and this will have a large effect on contiguity.

5. What other things affect conditioning?

There are several other factors that have a strong effect on conditioning. They are:

Prior exposure - familiarity,

Compound stimuli,

Surprise,

Relevance,

and Inhibition.

Prior exposure is a sword that cuts both ways. Prior exposure in which the two stimuli did co-occur wills serve to reenforce or heighten the conditioned response. Prior exposure, however, in which the two stimuli did not co-occur will increase the chances that a conditioned response will not occur.

When compound conditioned stimuli are presented several different effects occur depending on the timing and salience of the conditioned stimulus in relation the unconditioned stimulus. Most of this is intuitive and, therefore, does not require further explanation here as are the effects of surprise and blocking when two or more conditioned stimuli are used..

The amount of relevancy of the two stimulus to each other seems to have a fairly large effect on not only what will be associated but also on how quickly the association will take place.

An inhibitory response is one in which the absence of the unconditioned response will cause a conditioned response. What is interesting for us is that an inhibitory response can only be generated based on an understanding (knowledge) of the components that go into making the unconditioned stimulus. In order to know that something will not happen you need to be able to predict under what conditions it would happen and to know about what elements are present and what elements are missing.

6. How is conditioning related to learning?

Classical conditioning as mentioned in the answer to question 1 is a simple example of what is called associative learning. This is learning where one learns to associate two previously unrelated concepts, ideas, stimulus. It should be obvious that we are following the stimulus-stimulus (s-s) model here.

7. How can classical conditioning be used to explain some language learning phenomenon?

Form to function mapping

According to some theories of language, the basis of language itself is the mapping of forms to functions. A good indicator of this is the fact that all language has to be uttered or exist in a functional framework. This means that we always say things for a particular reason which not only has to be clear to the speaker but also to the listener as well. Therefore, we can see that functions are the basis and the groundwork of all language. As a person goes about using language they encounter functions as the basic stimulus and then they hear a piece of language which co--occurs within the context of the function. These two are associated and their features are mapped onto each other. This is form to function mapping and it is the basis of language. It would seem true then that the development of such mappings is based on the type of learning which we see in classical conditioning. There's an association between a specific function which we notice in the world and this seems to be like, at least initially, an unconditioned stimulus, and the conditioned stimulus which is a piece of language itself. Associations are formed initially between functions and chunks of language. Because we know that stimulus that share features will associate with other similar things, such as the initial unconditioned stimulus (in this case the function), we can easily see how similar forms will be associated with single functions. And it works the other way around as well. Similar functions will mapped onto singular forms simply because the functions have similar features.

8. How is it that `language` is a series of associated connections and not units?

Language has to be a series (and quite a complicated one at that) of associated connections between synapses because that is all that the brain does. The brain builds connections from neurons which connect at synapses. Information is not stored in neatly identifiable units like we see on paper but in the synapses of neurons. This is a complicated model when we consider that any bit of information is going to require a huge amount of synopses firing. To activate any type of linguistic unit this involves the simultaneous firing of millions of neurons. This is where the power of a parallel distributed processing (PDP) model comes in. PDP model basically says that a large number of disparate neurons in different neural networks can and do fire simultaneously. In a serial model, on the other hand, activation would need to spread in a specific order from one neural network to another neural network. While spreading activation is part of how the brain works it is only part and we require simultaneous firing of synapses.

What we are really describing at this point is the very basic underlying structural units of language. We have not looked at how an entire utterance has been put together yet we are simply looking at how the brain might store things like sounds and words/morphs/morphemes. In looking at this the system might appear highly behaviorist in nature in that the information stored in the brain is information which was brought in from the outside. As I said in class learning is essentially perceptual. We do also have conceptual learning but conceptual learning must occur after perceptual learning has reached threshold level. Systems in the brain need a certain amount of information to organize themselves into a system before they can begin to function. Looking just at sound it should be obvious that people learn sounds that are around them, and yes they do change them, but any change they make is going to be based on what they acquired from the environment. The same can be said for word units (we`ll call them that for simplicity`s sake). We learn words which we hear from others and more or less in the form that we hear them, of course interpreted through our own internal cognitive processes. This would we can see that the system is extremely complex and requires a large number of intersecting neural networks firing either simultaneously or in rapid succession.

A specific sound, for example, has no specific representation in the brain. The sound is made up of a neural network composed of different articulatory jesters. We note that these articulatory gestures need to be there because the brain needs to tell our muscles how to actually make the sounds. So, already we have a fairly complex neural network which involves taking specific articulatory gestures and firing them both simultaneously and in highly constrained orders. How this happens so relatively neatly is based on the design of neurons. The connections between synapses vary based on the strength of the connection and other variables. These simple physiological variables, based on experience (either productive or receptive), create neuronal firing patterns in the brain. These patterns have a tendency to reoccur. When a set of particular articulatory gestures are sequenced through links then what we might think of as a word unit is possible to be either uttered or understood. From here we have links to other aspects of language such as concepts and other word units. Everything is interconnected in thousands and thousands of ways. For example the neural network required to put together a sound like /m/is going to be connected to all the other sounds that /m/ is connected to. The number of connections here just for this small aspect of language is mind-boggling, but this is what the brain does. The system is extremely simple in its design and extremely complex in its output.

From a teaching point of view one of the main things we want to glean from all this is that there is a behaviorist (I use this word only for comprehensive ability`s sake. It is actually much more than a simple behaviorist system) foundation to language. This foundation, however, is only the beginning. Every user of language is constantly changing inside. As teachers this tells us first and foremost that the language we present our students and how we present that language to the students has an effect on their brain structure. This effect is what were trying to do as teachers. We are trying to change the structure of our students` brains, and particularly as regards language. How we do this is by providing linguistic input. People need to experience language. Use creates use. This very basic idea and of course we will be fine tuning this a lot as the course goes on but it is a very important place to start.

9. What are some of the basic types of nodes?

Lamb lists several different types of nodes.

-Nodes are bidirectional - yikes!

-There are different types of nodes according to lines - internal and external

-Nodes also branch in different ways - to single or multiple branches.

-Some seem to be ordered while others are not.

We need to take some of these claims with a very large grain of salt. First of all, nodes are not bidirectional and that this point we need to simply say that nodes are indeed synopses. The fact that Lamb is using these terms reflects his background as a linguist working with or well versed in computational models for natural language simulation, rather than a neurologist. Back when this book was written it was largely believed that synapses were bi-directional. We now know they aren`t, but this doesn`t necessarily challenge the model. It simply makes it a little more complicated, but we are the new was complicated in the first place so no big problem there. Lamb also mentions that there are different types of nodes or synapses which either are somehow internal to a neural network or form a link between different neural networks and are therefore external. This is certainly true but there is no strict delineation or clear delineation between internal and external synapses. Certainly neurons have many different branches, but synapses don`t. A synapse is just a link between one neuron and another. One synapse can branch off into millions of other directions but only through the help of neurons.

What we need to remember is that this model is just that, a model. It is not necessarily a very faithful rendition of the anatomical structure or functioning of the brain, although it does match very closely. In order to get the brain down on paper or into a computer things are going to need to be different, and for the most part much simpler. We can`t create models or computers with anything remotely near the number of connections a human brain would have. So, don`t worry too much about the specifics of his model. It is the general spirit which we need to embrace.

10. What are nections and how do they work?

Nections are neurons and they connect and manage the dispersal of electricity to nodes. To say that neurons manage and disperse electricity to nodes or synapses does not mean to indicate that there is some sort of cognitive processing going on inside of a neuron. How neurons manage to work as based on their connections to other neurons. As we mentioned previously a single neurons may have up to 50,000 dendrites. These different dendrites are going to be part of and connected to different neural networks. A neuron needs to be able to regulate which connections/synapses are going to be activated and which ones are going to be inhibited. And neurons does this by releasing different types of chemicals, some of which act as enhancers and some of which act as inhibitors to electrical conductivity. How this happens is not based on some sort of internal processing within the near on but simply on the strength of the neuron and its connections to other neurons as well as the state of the synapse. A synapse which has been primed is easier to fire and harder to inhibit. And neurons on is getting electricity sent through it from other neurons which are firing, thus causing connections and disbursing electricity into certain places, or certain synapses of the neural. It is therefore not surprising that these are the synapses which will be fired while others are simply inhibited from firing. This is really how it works and we can really see how the brain requires a PDP type of system to make this actually happen.