Symbolic learning & the grounding problem


Harnad, S. “The Symbol Grounding Problem.” Physica D 42 (1990). 335-346.


In a short paper, the authors attempt to define symbolism as a cognitive science but find that the theory fails due to the symbol grounding problem: that symbols are composed only of other symbols and thus self-referential.

They define 6 basic learning behaviors:

  1. discriminate
  2. manipulate
  3. identify
  4. describe
  5. produce descriptions
  6. respond to descriptions

which cognitive theory must explain. Examining the first and third behaviors, the authors propose a dual representation: iconic (symbol) and categorical (internal analog transforms). However, they admit one “prominent gap:” no mechanism to explain categorical  representations. The authors thus dismiss symbolism as a sole solution and turn to connectionism as a hybrid solution: “an intrisically dedicated symbol system…connected to nonsymbolic representations…via connectionist networks that extract the nonvariant features.”


The authors likely succeed for theorists but this was a little dense given my lack of background. I think I got the idea that a symbol (for example, a swastika) exists by itself and combined with “rules” (our prior learning and knowledge that the symbol has a recent association with the Nazi Party) to produce a composite symbol (loathing). I also took away that humans, especially in groups, are too complex to be semantically interpretable, and that connectionism (based not on symbols but on pattern activity in a multilayered network) may offer some answers.

The dual representation–iconic (symbol) and categorical (internal analog transforms)–seem to suggest a symbol paired with a real-world (our experience with/background on/knowledge of) event; however, the authors later define that as an interpretation. In addition, I’m not certain why the iconic representation is not symbolic as the authors state.

The conclusion makes sense (although this is classic Vygotsky–and connectionism seems like just another word for community): if a category is defined as a symbol (image) plus our experience with that symbol, I started to believe that all our knowledge is interconnected (within a single human) with past experiences–and agree that it may not be possible to model learning in a purely symbolic (i.e., no connection to the real world) fashion.


Instance-based and rule-based learning


Taatgen, Niels A. and Wallach, Dieter. “Whether Skill Acquisition is Rule or Instance Based is determined by the Structure of the Task.” Cognitive Science Quarterly: 2.2 (2002). 163 -204.


The traditional view of skill acquisition as “a gradual transition from behavior based on declarative rules in the form of examples and instruction towards general knowledge represented by procedural rules” is challenged by instance theory. While the latter seems to explain the inability of experts to verbalize rules, research on the directional asymmetry of rules seems to support the traditional importance of rules.

ACT-R’s instance-based architecture is based on 2 key arguments:

  • the strategy to use or the memory to retrieve is based on which has the highest expected gain (optimization)
  • declarative memory is activated (and filtered) by environmental demands and past experience (which has been encoded as production rules by procedural memory)

At the symbolic level of ACT-R, procedural rules are applied to declarative chunks which store information in a proposition; chunks are either new (perceptions) or created internally by prior knowledge/experience. Each rule contains a condition- and an action-part, and declarative items are pattern-matched to the condition and applied in the action. The subsymbolic level of ACT-R deals with the choice of which rule to apply according to Bayes’ Theorem (increases base-level activation each time it is retrieved and decays over time).

The article seems to imply that if instance-based learning fails (because the problem is too time-consuming–no discernible pattern), learners will attempt to derive some sort of rule. The authors argue that instance-based learning works best when the relationships between variables is very difficult; rule-based learning (simplified cases) is more successful with a large number of cases with obvious relationships.  The authors then test this hypothesis with two detailed experiments.

The results:

  • Previous research showed no learning through observation without direct rules (explicit relationships)
  • Subsequent research, indicated that exploratory participants did better than observers, but that observers could better verbalize and construct a causal model.
  • This research shows evidence of learning by observing even without rules; and participants seem better able to answer questions about old systems than new. Both results support instance-based learning.


ACT-R seems to offer tremendous explanatory power. However, instance theory seems related to the concept of expert (tacit?) knowledge:

  • know when to apply
  • gets better over time (more instances)
  • gets better with use
  • additive for community

Because production rules that propose new declarative rules are not accounted for in the ACT-R architecture, this “missing link” may be the elusive transfer element; the authors propose partial-matching in ACT-R retrieval as a solution.

In instance theory, encoding and retrieval seem more closely linked to temporal/spatial reality than to attention as the authors claim; however, the view of memory as evolving from algorithmic processing to memory-based processing succinctly describes expert knowledge.

The recommendations for design strategies seem profound:

  • Instance-based learning takes over from rule-based over time
  • Creating declarative rules is the most important (first) step
  • Analogy works best at the start but declines quickly
  • Declarative rule works well at start and persists
  • Instance continues to improve over time to become best

ID Theory

Reigeluth’s own contribution to the volume he edited (Instructional Design Theories and Models: A New Paradigm of Instructional Theory, Vol. 2. 1983. NJ: Erlbaum. pp. 5-29) is immense. He draws a clear distinction between design (prescriptive, decision-oriented) theory and descriptive (predictive, conclusion-oriented) theory. Learning theory is descriptive; design theory shows us how to accomplish our goals and includes three primary characteristics:

  • goals
  • methods
  • situations

The idea that design theory is probabilistic is equally true of descriptive theory. The final distinction made the most sense:

descriptive theory concerns validity, while design theory concerns preferability

A learning situation has both conditions and outcomes. The  conditions (the 4 he listed actually seem like 3) generally map to outcomes.


  1. What is to be learned
  2. The learner
  3. The environment and constraints


  1. Effectiveness (of the learning)
  2. Appeal (to the learner)
  3. Efficiency (of the delivery environment)

In addition, the methods of design theories are componential (with different parts, kinds, and criteria) although the individual components cannot be simply “added” to increase the probability of learning success.

Reigeluth then argues for a new model of learning based on the transition from an industrial to an information society. His vision of the current educational paradigm as based on mass production and standardization as befits an industrial (factory) approach was clearly articulated; his argument that we must move to mass customization was equally compelling, and I especially appreciated the analogy that the factory model was designed for sorting not for learning. However, his vision for a path to change was less helpful. While I buy the argument that people learn at different rates and that if time is constant, achievement must vary, the sole alternative (allowing time to vary) does not necessarily follow; achievement may also vary with the quality of the instruction (and by quality, I mean the broadest sense of the word: instruction matched perfectly to the learner’s needs at the moment the learning is delivered).

The distinction between basic and variable methods seemed somewhat artificial; variable methods that are proved become basic methods. At the same time, the alternative methods chart offers practical advice even though several of the methods are not discrete but compositions of other methods. The most challenging topic was reserved for the end: the argument that, “the only viable way to make decisions about instructional strategies that meshes with cognitive theory” is to do so during instruction, a proposition that implies an adaptive system.

The Design Profession

The career recommendations was extremely practical in listing designer, project coordinator, and artist as the primary personnel paths. The industry categories were similarly useful; accrediting bodies (such as associations) also offer job opportunities. The professional development sections were somewhat cursory (writing a blog should be added to the promotional suggestions). However, the 2 pages that summarized design models with visuals was superb:

  • Dick and Carey provides all the elements (ADDIE) but only a linear process.
  • R2D2 provides an iterative process solution. Assessment is the junction and bridge between Dissemnination and Definition; the model can also embrace Nokia’s spiral between explicit and tacit knowledge.
  • The layers model inserts the real-world constraints of time (budget); higher layers can produce meta-designs such as simulation archetypes.
  • The rapid model marries ADDIE to concurrence.

Development – A Factory model

The Development phase is treated as a factory model which may be more appropriate for large-scale projects than for individual courses. The team management section could be a separate text, and while the graphics section should add a paragraph on 3D modelers, it’s a practical and useful overview. The scheduling section should include usage of Gantt charts.

The most thorough aspect of the chapter was on post-project activities which are often ignored:

  • post mortem debriefing
  • deployment
  • recommendations
  • training
  • documentation
  • summative evaluation
  • client evaluation
  • project cost analysis

Demonstration phase

A transition between Design and Develop/Deliver, this phase is often incorporated directly into a single iterative unit release methodology. The optional deliveries were very helpful:

  • treatment
  • scenarios
  • templates (content as well as layout/technical)
  • requirements spec (especially useful for software)
  • storyboards
  • scripts
  • prototypes

Practical suggestions are prevalent:

  • media asset lists
  • reading levels
  • translations
  • reviewer training

The storyboard form could have been demonstrated as a template–with the navigation (top) annotated as a common element.

A Millenial Learning Style?


Reeves, Thomas. Do generational differences matter in instructional design. Retrieved 2/6/2009 from


Most of this paper reviews conflicting research on differences among the three most recent generations (boomers, gen X, and millenials) but tends to embrace the conclusions of Twenge: NSD. The author takes most research to task for lack of rigor, especially for failing to address socioeconomic status. By admitting to the existence of, but adopting the most conservative view of,  generational differences, the author concludes that these differences do not constitute a sufficiently important variable to justify modifying instruction; similarly, he dismisses learning style differences as having little validity or utility. At the same time, the author lists games and simulations as intriguing areas for further research, and notes that distance education is equally effective as classroom instruction


While the paper correctly criticizes the lack of rigorous scholarship by “optimists” such as Prensky, the willingness to embrace the almost equally suspect “pessimists” seems somewhat arbitrary, particularly for a paper that professes a pure research-based approach. The argument that the research has concentrated on higher-income Caucasian learners seems wholly justified and points to a major gap in the literature.

The brief discussion of the potential of games for education was puzzling. While the author spent the bulk of the paper dismissing cursory research in generational learning styles, he was eager to embrace equally suspicious research into the efficacy of games in education. His dismissal of learning style differences was based on a single paper (Coffield) which only considered the validity of Kolb’s LSI instrument, not the concept itself.