Deep learning is learning in which students look for meaning, connect new knowledge to what they already know, and become able to use it in situations they have not met before.
Ref: Winje & Løndal (2020); Pellegrino & Hilton (2012)
One term, two meanings
The term is not used the same way everywhere. Winje and Løndal (2020) analysed how 71 research publications on primary and secondary education (1994 to early 2018) define deep learning. They found two main conceptions:
Deep learning as meaningful learning.
The largest group (63 of the 71 publications). Core elements: searching for meaning in the material, relating new knowledge to prior knowledge and everyday experience, and intrinsic motivation. This conception goes back to the research on deep and surface approaches to learning by Marton and Säljö (1976).
Deep learning as transfer.
A smaller group (8 publications): the ability to take what was learned in one situation and apply it in a new situation. This is in line with the report of the US National Research Council on deeper learning (Pellegrino & Hilton, 2012).
Bloom and RTTI
Learning taxonomies make it possible to sharpen the definition, and to measure it.
The Revised Bloom’s Taxonomy for Learning (Krathwohl, 2002) uses “Apply” when a student has to find a procedure to solve a problem. That covers both choosing a known procedure and adapting a procedure to solve the problem.
The RTTI classification table explicitly mentions “new situations”. RTTI splits Bloom’s “Apply” in two: Training (known procedures) and Transfer (a concept applied in new, unknown situations).
This lets you read the two conceptions of deep learning in RTTI terms. Meaningful learning (making connections, linking to prior knowledge) is close to Insight. Deep learning as transfer is exactly RTTI’s Transfer category. So anyone who wants to measure deep learning looks at Insight and Transfer.
Questions
Questions are not only used to check whether a learning goal has been reached, but also to engage students. One of the best answers a child can get to the question “why?” is a counter-question: “what do you think yourself?”.
Relevant questions for deep learning include: explain …, state a causality, choose the best alternative and justify it, demonstrate that …, discuss … etc.
A sheet with the signal words (verbs) for the four cognitive categories can be downloaded for free from tli.cloud (“RTTI signal words”). Both teachers and students can use it to recognise the question type.
Another way to look at the quality of an answer is the SOLO taxonomy (Biggs & Collis, 1982). It ranges from a prestructural response up to an extended abstract one, where the integrated understanding is generalised to a new topic or area.
A balanced test contains questions in all four cognitive categories: reproduction, training, transfer and insight. Deep learning is mostly found in Insight and Transfer, so those are the questions that matter when you want to measure it. Of course there is a difference between students in grade 8 and students in their final year of secondary school. A test for grade 8 contains more questions on Reproduction and Training, while final-year students get more Transfer and Insight questions.
But teaching and tests in grade 8 must also treat material from the categories transfer and insight. Then students can (begin to) develop and practise their learning strategies for those cognitive categories too. Besides, some students remember facts most easily through understanding: without understanding, no remembering.
A test for grade 8 can contain as much as 25% training and 5% insight questions. For the final year the situation is almost reversed, with most focus on Transfer and Insight; still, those students must keep practising their strategies for Reproduction and Training, so keep up to 30% of the questions for those categories.
The unexpected focus on Transfer and Insight that students often meet at the start of upper secondary school, without a real transition, puts many students in a difficult position: learning strategies that worked well in lower secondary school are suddenly useless.
Learning strategies for deep learning
Several learning strategies that are useful for deep learning are described in publications by Drost & Verra, such as Formative Evaluation with RTTI. You will also find some on this site under Transfer.
Some of the other strategies focus on linking the new topic to existing knowledge and experience. Asking yourself W-questions such as what, who, how and why is a simple and effective strategy. The techniques reviewed by Dunlosky et al. (2013) include elaborative interrogation and self-explanation, both rated as moderately useful.
Hattie and Donoghue (2016) conclude that strategies work best when they are matched to the phase of learning and embedded in the content of the subject, rather than taught in separate sessions. The Education Endowment Foundation (2025) guidance on metacognition and self-regulated learning includes explicit teaching of how to plan, monitor and evaluate one’s own learning.
Teaching for deep learning
Teachers can keep in mind:
- How can I get students to think about which procedure or approach they consider best suited in a new situation?
- How do students get the opportunity to pick from known procedures and put together a procedure of their own?
- Are there several possible answers?
Example: after practising a couple of procedures, you give students a new challenge in which they have to choose or combine to solve it.
In the limited time you have, you can show the way a couple of times, but students must eventually walk it themselves. A depth task in the test must concern a new, unknown situation. If the student has seen the task and the solution before, it is Training, not Transfer. How “new” a situation is can vary along several dimensions (Barnett & Ceci, 2002).
Deep learning (RTTI categories Insight and Transfer) must be seen in connection with Reproduction and Training; none of the categories exist in isolation. By using the right verbs you get good depth questions on the test. With a good distribution of questions over the cognitive categories, you can measure how effective the deep learning for the topic has been.
References
- Barnett, S. M., & Ceci, S. J. (2002). When and where do we apply what we learn? A taxonomy for far transfer. Psychological Bulletin, 128 (4), 612–637. https://doi.org/10.1037/0033-2909.128.4.612
- Biggs, J. B., & Collis, K. F. (1982). *Evaluating the quality of learning: The SOLO taxonomy (Structure of the Observed Learning Outcome)*. Academic Press.
- Education Endowment Foundation (2025). *Metacognition and self-regulated learning: Guidance report*. https://educationendowmentfoundation.org.uk/education-evidence/guidance-reports/metacognition
- Pellegrino, J. W., & Hilton, M. L. (Eds.). (2012). *Education for Life and Work: Developing Transferable Knowledge and Skills in the 21st Century*. National Academies Press (National Research Council).
- Winje, Ø., & Løndal, K. (2020). Bringing deep learning to the surface: A systematic mapping review of 48 years of research in primary and secondary education. *Nordic Journal of Comparative and International Education, 4*(2), 25–41. https://doi.org/10.7577/njcie.3798
For more information, check https://tli.cloud.