Ofsted Confusion: Neural Networks, pseudo-science and learning

 I have to start with this slide from HMI Michael Wardle's Webinar on Curriclum Design, because it looks so much like a map of Norfolk with a vastly upgraded transport system.

I think he is using it entirely metaphorically. As a picture of the brain's outline. And a generic representation of a network framed by that brain. To illustrate the point that learning is a complex web of knowledge where connections are made over time.

I don't think this is meant as a representation of an actual physical network in the brain, any more that it represents the actual road network in Norfolk.

But we have to be careful. I HAVE seen representations of "neural networks" and computer generated graphics of "a synapse firing as a memory is created" purporting to show learning physically happening. This was in a different presentation which was urging me to use this understanding of how learning happens in the brain, to help inform my approach to planning learning and teaching. 

As science, it was clearly utterly bogus. A discrete memory is not physically formed by the firing of a synapse in a little package of just the right amount of knowledge to cross the barrier from short term working memory to long term storage memory.

This isn't to say that "cognitive science" isn't useful. But we need to be careful to understand that talking about "storage" or "retrieval" are all metaphorical models. In the same way as we now talk about the brain being "hard-wired" for something, these are using the metaphor of computer storage to help express our understanding of what is going on in the brain.

In previous eras, memory was thought of as a hydraulic system, a series of cogs, a telephone system - whatever the technology of the time suggested. So it's not surprising that today we use the metaphor of computer memory. Except of course, that this is a metaphor within a metaphor. Storage and memory are already metaphors when applied to computers.

Then we also talk about "networks" in the way language works and the brain deals with language. Research on the brain can show what regions of the brain are activated when processing language. The main conclusion would seem to be one of immense flexibility and complexity, that is very far from offering an easy granular process to be followed to create learning. When Michael Wardle says learning depends on a web of knowledge and schemata formed over time, I don't think he means that we can look into the brain and see networks forming or firing. I think he means that it helps if you already know stuff, and new stuff you learn interacts with what you already know. In a complex, dynamic and changing relationship which is very far from a physical network of roads and roundabouts. (Yes, we have roundabouts in Norfolk.)

This storage network metaphor turns up in research in language-learning. For example this study talks about the way words "link" and "compete for space" in the memory. For example it suggests that if you learn a list of words such as lion, tiger, wolf, jaguar... then these will compete for storage and retrieval, and make recall marginally slower. But if you learn words like lion, yawn, yellow, mane... then these words will form a network and reinforce each other in learning and recall.

The metaphor of computer memory with networks and storage might be helpful here and these could be interesting (if marginal) findings. But as a teacher I would have questions as to whether this is less to do with the mechanics of "storage" and more to do with the interest and affective aspects. A list of animals to learn in a deliberate act of memorisation. Versus a set of words that set off the imagination with interesting descriptive detail. The research throws up interesting ideas that can be explored and interpreted in different ways. But reading the article didn't cast light on how the brain worked. It cast light on how the authors thought about the working of the brain.

As linguists, we know other ways that words form a system. We want our learners to see the underlying patterns. Not just so they can apply grammar to be accurate, express themselves, or show complexity. But also because we think it is a shortcut for memorisation. It's easier to remember things that make sense. Knowing a rule can save you from memorising a million examples. This is all true. But it's looking at it from the point of view of someone in possession of the knowledge of the whole system. For a learner, it is entirely different.

In his webinar, Michael Wardle focuses on how we could take this formal grammatical network of French and try to build it in the pupils' brain. He gives the example of introducing pets and adjectives. We have all gone through the complexities of teaching word order and adjectival agreement. Michael Wardle's approach is to start by "hiding" all adjectives which would change for feminine. So he would teach pupils to say they have a red dog, because the word rouge can transfer to saying they have a red tortoise without having to worry about changing the ending.

I can see the logic of this. And as a cunning ploy I am not above using it myself. It's similar to the way when I first went to France, I asked for deux of everything. To avoid making any errors of un / une. Expensive and waist-expanding, but at least I wasn't committing a grammatical mistake. And every year there is a pupil in Year 7 who just says they are going to describe everything as orange because it's invariable. And then I share compound adjectives with them and say as long as you say it's dark blue you don't have to do the agreement either. As I say cunning tricks. That I use half-jokingly.

But this isn't a joke. This is Ofsted setting out how to construct a curriculum. It depends on totally controlling what pupils say, so that it becomes necessary to prohibit self expression and communication at this stage of language-learning. Pupils can talk about red dogs and red tortoises. But as soon as someone has a green spider, then that is inadmissible because it doesn't fit the grammar pattern we want to see exemplified. We saw the same idea in The Great Pet Debate where perro and gato were fine, because they exemplified masculine and feminine endings. But pez, ratón and serpiente were not allowed.

I do think it is important to build pupils' core of language. And that once that core of language is established, then more and more will stick to it.

But it I think there is a sleight of mind going on, thinking that one "network" can be slipped into the mind instead of another. The idea that there are neural networks in the brain, and that knowledge can be described as a web, and that language has an underlying structure of links, does not mean that you can construct a grammatical framework and plug it into the pupil's brain like an extra graphics card into a computer. In the same way that I couldn't use the map at the top of this post to get from here to Hunstanton.

It is a nice theory that collapses on contact with reality of real human learners. It is an attempt to create a grammatical network, built up out of carefully selected knowledge, to make a tightly dictated schema that is predicated on the understanding of the grammatical system. And then it hopes to replicate that grammatical system as a network in the brain. Because of the power of the diagram on the slide to convince us that this is how the brain looks. And the mixed metaphors of neural networks, a web of knowledge, language as a grammatical system.

There are many, many more things in the human brain for learning and language to stick to. There's the sound of words, the way words look, the pupils' own language(s). And their lives and lived experience and ability to interact, and stories, and visualisations and silly associations and mnemonics and emotion and relationships and family and likes and dislikes. And yes, pets. Language engages and interacts with all of these. Not just with itself in an abstract network.

It is an inconvenient but undeniable truth that language attempts to refer to the world. It is not a closed self-referential system. Language is meaning as well as form.

So I might teach pupils cat and dog because they exemplify masculine and feminine endings. But the pupil with three snakes, a mouse and a fish will have very strong "networks" of experience that the words for cat and dog might not stick to. Let alone the owner of the spider and their "web" of knowledge. Meaning is the primary role of language and its strongest way of making links and memories. To put form above meaning in the hope of constructing a network in the brain is misguided and misconceived.

This is the rabbit hole we are being dragged into. By NCELP, Ofsted, and the new GCSE panel. It is an ideology based on creating a synthetic, abstract language, where the emphasis is on exposure to language chosen for its form so that linguistic patterns can be identified. It teaches a limited, reduced and deliberately simplified set of language, so that the learner might easily discern the patterns and supposedly turn them into a mental "network". This can only be done by removing the "obstacle" of the learner wanting to say or learn things that don't fit this carefully selected web of words. So communication and self expression have to be delayed.

And it's not just pets. As we have seen with the proposals for the new GCSE which seek to limit pupils to "high frequency" words, things like step mum, foster family, wheel chair are not to be taught. So if the pupils wanting to use these words can't learn them, then no pupil should be allowed to talk about their actual family. But that's OK because we are going to be talking about red dogs and red tortoises. To practise the language, not the meaning. But it's meaning which creates the strongest experiences, memories, networks.

I could stop there on that point that the goal of memorisation is best achieved when it links to a far richer network of meaning and real experience than just a self-referential linguistic system.

But I will go further (sorry). But I'll make it brief with some points to think about:

The complexity of language is real. But the human brain is up to the job.

The learner's interlanguage is by definition incomplete, partial, and includes misconceptions.

Expressing themselves has an important role in forming the unconscious schemata of the learner's interlanguage. Making meaning from the words and structures they have, exploring how it can be used, how it fits together, and what their current limitations are.

So I will finish with this:

If Ofsted want us to keep a close eye on what pupils can manage and what overwhelms them...

If Ofsted want us to think about how knowledge sticks...

If Ofsted want us to think about how knowledge accumulates...

If Ofsted want us to keep a close eye on how knowledge can be recalled fluently...

...then the best way to achieve all this is the opposite of what they are proposing. The best way is to acknowledge that language has meaning as well as form. The best way is to have a strong focus on curating a growing repertoire of language that pupils can use in order to express themselves with increasing independence, fluency and complexity.