Skip to content
University of GroningenNiels Taatgen

Research

I use cognitive modeling, computer simulations of human thinking, to study how people learn, control their attention, perceive time and combine simple building blocks into complex skills, and I increasingly apply what we learn to education.

In cognitive modeling we construct computer simulations of human thought. The goal is to understand human thinking better and to build more rigorously testable theories of cognition. I work primarily with the ACT-R architecture, developed by John Anderson, and with my own extension PRIMs.

My research is driven by a sense of wonder that humans can do almost any new task they are confronted with. Two themes are central:

A complementary thread of my work is methodological: I want to make cognitive modeling more constrained, by letting models learn their own task knowledge and by making real, prior predictions of experimental outcomes rather than post-hoc fits.

Research topics

AI in Education · current focus

Building knowledge graphs from student data, and how personalized practice leads to better learning results.

Open page

PRIMs and transfer of cognitive skills

An ACT-R extension that breaks rules into primitive building blocks to explain when training on one task improves another.

Open page

Multitasking

Threaded cognition and the problem-state bottleneck: when does multitasking lead to interference, and why do people do it?

Open page

Time perception

A model of interval timing in which an internal clock, attention and learning together produce our sense of time.

Open page

Inaugural lecture (2010)

Draden door de Geest: how to keep hold of the strings in a world of multitasking. In Dutch, with the original figures.

Read

Main theoretical contributions

Skill acquisition

When I started using ACT-R it lacked a plausible mechanism for learning new production rules. During a sabbatical in 1999 I developed production compilation: a single mechanism that specialises general problem-solving strategies (e.g. analogy) into task-specific rules, and which in some models leads to qualitatively different behaviour after enough practice.

First applied to learning the English past tense (Taatgen & Anderson, 2002), production compilation has since been used to model German plural learning, phonetic representations, the balance-scale task, and the Kanfer–Ackerman Air Traffic Controller task, accounting for both aggregate data and individual differences.

Control in dual tasking and complex dynamic tasks

A later thread of this work asked how skill acquisition makes people not only faster but also more robust and flexible. The solution was to abandon strict hierarchical task representations in favour of a weak hierarchy: instructions specify the conditions under which they apply and are organised in instruction sets. This produces a better balance between top-down and bottom-up control, and led to the minimal control principle: people organise their task knowledge to minimise the amount of top-down control required.

Key references

Taatgen, N. A., Huss, D., Dickison, D., & Anderson, J. R. (2008). The acquisition of robust and flexible cognitive skills. Journal of Experimental Psychology: General, 137(3), 548–565.

Taatgen, N. A. (2005). Modeling parallelization and flexibility improvements in skill acquisition: From dual tasks to complex dynamic skills. Cognitive Science, 29(3), 421–455.

Taatgen, N. A., & Lee, F. J. (2003). Production compilation: A simple mechanism to model complex skill acquisition. Human Factors, 45(1), 61–76.

Taatgen, N. A., & Anderson, J. R. (2002). Why do children learn to say "broke"? A model of learning the past tense without feedback. Cognition, 86(2), 123–155.

Learning from instructions

Can a good theory of learning improve real instructional situations? Anderson's algebra-tutor work suggests yes. My own work, funded by NASA and the US Air Force, looked at how people learn from instructions and eventually develop knowledge robust enough to handle interruptions and novel situations. The domain we used was programming the flight management system (FMS) of a Boeing 777.

Standard pilot training has crews memorise procedures as ordered lists of steps. This works poorly: pilots forget infrequent procedures and struggle to adapt them to new circumstances; they are neither flexible nor robust. Our cognitive models suggested a better representation: don't just list actions, also specify the conditions under which each action applies and the effect it should have. Take the first step of a typical procedure:

1. Press the LEGS key.

Augmented with context, the same step becomes:

If you want to go to the LEGS page, press the LEGS key, which will bring you to the LEGS page.

In experiments comparing list and context instructions, people who received the context version learned faster, generalised better to new problems, and pressed the LEGS key significantly less often when it would have been unnecessary. Eye-tracking showed an additional benefit: people who got context instructions performed safety-critical verification checks substantially more often (71% vs. 46%).

Funding

Air Force Office of Scientific Research.

Key references

Taatgen, N. A., Huss, D., Dickison, D., & Anderson, J. R. (2008). The acquisition of robust and flexible cognitive skills. Journal of Experimental Psychology: General, 137(3), 548–565.

Taatgen, N. A., Huss, D., & Anderson, J. R. (2006). How cognitive models can inform the design of instructions. Proceedings of ICCM 2006, 304–309.

Team

Current PhD students

Frequent collaborators

Former PhD students

Former postdocs