Theoretical Basis

The main ideas here come from the work of Krashen. He argues that the optimum way to develop a second language is by actual experience in meaningful contexts. Something he refers to as COMPREHENSIBLE INPUT. Contrary to received wisdom, he emphasises that we don't develop general language abilities by learning vocabulary and rules, or by talking in that language. I would however qualify this, by saying that talking is a skill which needs to be developed, and that this can be a form of comprehensible input, if the content can be understood by and is relevant to the learner.

A good review of Krashen’s ideas is given below, in a 2015 talk (split into 3 parts). He’s pretty entertaining, and makes a strong argument that is easy to follow.

You can also see a more technical review of his Comprehensible Input hypothesis here.

To summarise, the key processes for Krashen are:

► There should be meaningful involvement, which he believes happens best through reading, based on understanding and interest.

► That language learning always happens with the same sequence of knowledge and skills. This is different from the way it is usually taught, in the form of complete rules, with increasing logical difficulty. This results in problems, and it is therefore best to let the sequence arise by itself (with normal input exposure).

► It is possible to consciously monitor our language processes, but that again this interferes with natural and optimal development. Progress is best when we are not aware of what we are developing, but absorb language in the same (implicit and unconscious) way that we learned our first language - Krashen refers to this as 'acquisition'.

► If we don't enjoy involvement with the language, then Krashen argues that acquisition cannot happen; that we then have an affective filter which blocks development.

I definitely agree with his general ideas and approach and there are many parallels with the process of normal (formal) language development. For instance with Nelson's (1973) finding that mothers who impose structure (for correct word choice and pronunciation), slow development down.

Critically, there is direct support in terms of Krashen’s ideas, in terms of student progress. See this article (BLOG, referenced below) for a good review. When students read matched, interesting material, they make the same, or better progress than those who spend the same amount of time with formal vocabulary and grammar learning. Moreover, whatever gains are made with taught grammar and vocabulary are typically quite specific, short-term, and tend not to be applied or result in long term interest or involvement.

What this means for Learning

There are many resonances with various aspects of educational psychology, as well as a number of basic psychological principles. There are also a number of practical implications which come from this.

- Perhaps the most obvious one, is that you should carry out lots of interesting and enjoyable activities in the new language, and avoid formal vocabulary and grammar. As with our own early language development, learning should be largely incidental, with activities focussed on meaning and communication, with limited conscious awareness of our progress.

This seems to go against common sense, as much of formal education is highly structured, and carried out in a very conscious, monitored way. However, as Bruner (1997) reminds us, our current educational system is historically relatively recent, and the majority of past human learning happened as a form of apprenticeship. This sort of experience is simply impossible with normal class sizes, and to manage group delivery, teachers have to set up structured activities which they can deliver and monitor. Incidentally, 1:1 teaching (tutoring) has been shown to bring the average pupil to the level of the top 2% of children taught in normal classes (Bloom, 1984). This is due to the teacher being able to individually monitor the pupil's needs and to tailor learning closely to this.

A further key aspect is therefore the need to provide matched input. This is a theme which arises in many areas of psychology, and has a strong foundation in general cognitive development (see below on Piaget). It is also a key feature of motivation, with Csikszentmihalyi (2014) arguing that when we are closely matched with an activity, this results in a state of ‘flow’; this is essentially a state of non-conscious development, and can be a strongly reinforcing. This is of course quite different from normal grammar and vocabulary work, which require conscious effort and can be quite aversive.

In brief, research in this area has shown that the key aspect of match is to have a success level of greater than 95%, and less than 99%. This 'sweet spot' is much higher and narrower than most people imagine (teachers typically estimate around 60%); it ensures a sense of involvement, without getting bored (hence the small unknown %). With reading (either in one's first or a foreign language), this translates into a rule of thumb of about 1 in 20 words unknown. This level usually provides sufficient context for the overall meaning to be established, and for unknown words to be identified. Successive experiences of an unknown word progressively build up our knowledge, and it takes about 20 exposures, to reach the level where it can be defined.

If you apply this to reading, and vocabulary development, Nagy & Herman (1987) found that a moderate amount of daily reading (23 minutes a day) would result in learning about 1,250 new words a year. Nation (2014a, see video below in references) has used this approach to develop graded reading materials for ESL learners. His system uses progressive increases in the time spent reading to get coverage of sets of target words as they become less frequent. These books take students up to the level where they will be able to read normal texts independently. From Nation (2014b, p7), the reading time necessary to establish upper intermediate vocabulary is in the region of about 1 hr a day for 2 to 3 years (depending on your starting point). Although this may be an underestimate as the criteria used was simply recognition (with 8 exposures), it is actually not that far off my own experience, taking about 3 yrs to go from A2 to a good B2.

However, the problem for people learning French as a second language is that a complete series of graded texts don't exist. In addition to this, these readers can be somewhat restrictive. Nation’s work for instance uses a range of older classics such as A Christmas Carol. This loses the motivation which comes from a sense of independence and self-choice, as well as lacking contemporary relevance.

As an alternative to this, Cheetham (2015) argues that children's general literature in a foreign language forms a good basis for early reading in that language. Although there are some concerns over mismatch between the level of maturity of the subject matter and the reader, such reading can be self-guided and adjusted for interest and difficulty. Cheetham also argues that self-selected reading leads to focusing on specific authors and themes, with greater exposure to repetitions of words and stylistic features (an idea that is supported by Krashen’s work). This has therefore been the basis for my own reading, starting with children's readers, and I continue to focus on stuff which I enjoy.

Developing Krashen’s ideas

Although most of Krashen’s arguments and recommendations have received support, it is likely that his ideas need some expansion, as he seems to have been principally concerned with ‘language development’ as a single concept. In particular, he does not really consider how we develop our listening and spoken skills. Decoding a stream of spoken French is very different (and much harder) than reading the same content (where you can pace your involvement, and check things you aren’t sure of). Spoken French is typically fast, with words run together, and with limited intonation; if you lose track then you can become completely disrupted. Generating spoken French within a meaningful social interaction is at an even higher level. I can personally read most things in French (close to my normal rate of reading English), but still have difficulty listening to natural spoken French, as well as speaking French (although with this it is at least possible to control the rate and complexity).

The key skills involved here seem to be with the automaticity of speaking and listening, so that you can concentrate on the meaning. ‘Expertise’ seems to play a large part in this, and it is likely that this takes a considerable amount of time to develop fully – probably many years (as with first language acquisition). Maybe this equates to the 10,000 hrs to become a complete expert (as popularised by Gladwell, 2008). At one hour a day, this would take about 27 yrs, though I’d imagine if you were this much into the language, you’d be spending much more time each day involved with it.

Apart from this, I also think that some level of conscious learning can help to accelerate things, particularly to consolidate what you have already developed, and to clarify some higher level principles. This is particularly important where our first language skills tend to block the new understanding. A good example of this is in the use of 'que' and 'qui' to start a subordinate clause - for an English speaker it initially sounds wrong to say 'c'est le livre qui avait tombé' - we tend to think of it as 'which', or ‘that’, and want to use the word 'que'.

General Learning and Cognitive Development

I think that in order to get a better idea of what is going on with language development, it is useful to base this on general principles of learning and cognitive functioning. A connectionist paradigm is being increasingly used in psychology, and this is based on structures known as neural networks. It’s useful to do this as the underlying assumptions and principles are similar to the way in which the structures of the brain operate. These are also being used increasingly to develop effective AI and language processing abilities (as with Google’s speech systems, and translation ). Joanisse and McClelland’s (2015) review shows how this approach is becoming more popular.

A connectionist approach can be matched with existing theories of cognitive organisation and development, which operate more at the descriptive level. In particular, the cognitive developmental psychologist Piaget's ideas continue to be the basis for many developments.

So, starting with the fundamentals of connectionism, it’s useful to look at the underlying ideas which are the basis of Neural Networks:

These are systems of artificial neurons (brain cells). As in the brain, these are linked together by connections which can be modified in terms of how well they are able to transmit a signal. As with actual brain cells, each cell can briefly ‘fire’ (become active), when the combined input to it goes above a certain threshold. When this happens, the activation is transmitted to a large number of other cells (typically thousands).

The input to the network is a pattern of activated cells, and the output is in the form of activation of a different set of cells. The nature of the connections between these determines the form of processing which takes place. For instance, a visual pattern can act as the input (such as a picture of an underground station platform, as in Cho et al [1999]), and the output could be some key feature of this (in this case, how crowded the platform is).

One interesting aspect of the network is that the system is ‘trained’ to operate using a number of different inputs; values for the desired output are used to modify the weightings of the various connections – known as backward propagation. This means that the actual connections arise from the input information and the feedback received. Earlier AI developments were based on systems of rules which had to be meticulously programmed in – although these were often quite effective, this was typically due to a brute force approach, involving massive numbers of computations.

In most cases, for neural networks, the connection weightings and the overall system functioning are quite complex and cannot be predicted by humans. In the field of GO competitions (a Chinese board game), not only has a neural network based system beat the world champion Ke Jie (The Guardian 2017), but it did so using highly original and creative moves which humans have difficulty understanding.

Neural networks operate to derive associations in the input data – whatever they may be. For instance, Sejnowski and Rosenberg (1987 p.157) trained a neural network to ‘read’ (convert text to speech), and the system came to use part of its network to specifically analyse for vowels. Such a feature can be seen as an ‘emergent property’ – something which arises independently, as the complexity of a system increases.

Neural networks don’t use a form of conscious awareness or higher level guiding (executive) function, yet are able to come up with performances which are able to outperform humans. It is also increasingly questionable to what extent humans actually use what we call 'consciousness' to direct or manage our thinking. It may be that the majority of our cognitive processes are similar to neural networks, with limited awarness. Consciousness may only come in as a form of post hoc explanation (Oakley and Halligan, 2017), and may be something of a red herring in our understanding of cognitive processes.

A connectionist model of language development supports the use of a more experiential approach to learning, rather than conscious, direct instruction. The emergent properties of the network development may show rule based behaviours, but these are the consequences, not the determinants of learning. It is also interesting that the stages which networks go through are typically close to what we see with human learning, for instance with language networks showing early ‘babbling’, and errors which show over generalisation of rules, for instance saying ‘comed’ instead of ‘came’ (Rumelhart and McClelland, 1986, p.220).

This approach is quite different from the previously dominant paradigm in language development proposed by Noam Chomsky (see Wikipedia review). He has argued that we develop language according to a universal and inherent rule-based system – an underlying grammar, which transforms meaning to language and vice versa. His ideas came largely from the observation that all languages have an essentially common structure. The key inference from this approach is that learning should therefore focus on these rules, with a strong role for grammar teaching.

However, the commonality of language structures may just be the result of the similarity of human experiences and inherent causality – for instance, with the world generally being made up of ‘things’ ‘doing something’ to ‘other things’. A pattern-seeking AI (our brain) would automatically reflect this underlying structure, which would be the basis of any communication system. Again, from this perspective, language structures (grammar) are not causative, and teaching them would be a rather pointless or even damaging procedure.

In fact, with normal language development, there is considerable evidence for a progressive constructivist approach. Children appear to generate and successively modify rule-like structures, all without any apparent awareness or effort. The example of ‘comed’ above for instance, indicates that the child has developed the idea that –ed covers things in the past.

In the same way, children’s early language also shows many simplifications – for example by saying ‘doggy go’ meaning ‘the doggy’s gone now’. Trying to accelerate language by just correcting and getting the child to imitate the complete (correct) phrase is typically an exercise in frustration. In fact, the vast majority of (successful) parent-child interactions simply involve responding to the underlying meaning of what children say, NOT the correctness of their grammar (Lierberman, 2015). This has direct implications for second language learning, although it is obviously very difficult to duplicate childrens' support system (constantly attentive language experts) for foreign language learners..

Finally, we can relate all of this to a key theory of cognitive development. Piaget (see Wikipedia article) has described the general structure of cognitive processes, as well as the way in which these develop. Cognitive organisation is seen as the formation of schemas (Swiss French for ‘patterns’) – which are similar to the idea of a concept. The concept for ‘dog’ for example has the formal features of being a quadruped, carnivorous mammal, and is a popular household pet. A schema however is every key aspect related to a dog for a person. It includes features of a typical dog (the classic shape, barks, chases cats etc), but also our relationship and feelings (whether we like or are scared of them), as well as our typical actions with them – going for walks etc.

Hopefully, you appreciate that schemas can be seen as the result of the development of neural networks. The organisation and development of schemas therefore gives further insight as to how learning occurs.

Piaget described the developmental process as being the elaboration of schemas through a balanced process of assimilation (when new information is simply fitted in to existing schemas), and accommodation (when schemas are changed to adapt to new input). For instance, a child may have the schema of ‘fish’ which includes dolphins and whales (due to their shape, and the fact that they swim in the sea). Evidence that they come to the surface to breathe, may simply be assimilated – although it is likely to cause some tension with other information. Finally, more sophisticated information (such as bearing young alive and suckling them), can lead to a reorganisation (accommodation) and the new schema of ‘cetaceans’.

The balance between assimilation and accommodation can be related to the general idea of ‘match’, with optimal learning development and high levels of involvement being when there is mostly assimilation, but also with some accommodation.

Piaget sees learning and development as a form of construction by the individual, with progress being based on the individual’s abilities and interests, rather than a simple didactic sequence, based on the apparent logic of what is being taught/learned. Bruner (1961) developed Piaget’s ideas to see learning as a spiral, with areas being revisited as more general intellectual progress occurred.

These ideas reinforce the belief that second language learning should focus on the normal spontaneous sequence of native language development, with activities emphasising what is already known (but with continuing expansion), in the context of individual communication and understanding.

As mentioned earlier, you would conclude therefore that the optimal system for second language learning would be one of continuous 1:1 tutoring. This would need someone to match their language closely to the student, to provide an expanding language input, while emphasising the meaning and use of language in all its forms. For children this basically the role of a good parent/carer, and longitudinal studies have confirmed that there is a strong relationship between the amount and quality of language input a child receives, and their language development (Hart and Risley, 1995).

Unfortunately of course, this is impractical for second language learners. It is also impossible to achieve such matched language experiences just by being in the target language country. This is due to the the problem of immersion mismatch (the fact that normal life is well beyond the language capabilities of people who are not yet near to being fluent). Therefore, one has to look for situations which initially provide a high level of meaningful input, followed by opportunities to speak and listen, and then to communicate using writing.

High levels of enjoyable (matched) reading seem to be ideal as a form of direct input. Reading speed is typically much greater that normal listening or speaking, and readers can stop and check when they have any problems. This can be supported even further by aids such as rapid IT-based dictionary checks; these don’t interfere with the ability to construct plausible language and develop meaning. Skills developed from this can then act as the basis for interest-led and supported listening, then speaking with feedback, and finally writing.

Once a level of independence (B2 and above) is achieved, then meaningful independent experiences with the natural language become possible and enjoyable. People are then as it were ‘in the zone’ and their progress will then be based on what are essentially normal activities (in that language/culture), following individual needs and interests.

Theoretical Basis

Ideas and evidence for the approach