Citation: Salomon, G., & Perkins, D. N. (1989). Rocky roads to transfer: Rethinking mechanisms of a neglected phenomenon. Educational Psychologist, 24(2), 113–142.
URL / source: Educational Psychologist via Lawrence Erlbaum Associates. (Direct DOI: 10.1207/s15326985ep2402_1)
PDF: salomon-perkins-1989-rocky-roads-to-transfer.pdf in this folder.
Why this matters for our project
This is the paper that defines the conceptual map we need. Two contributions:
- Two-mechanism theory of transfer — low-road (extensive practice → automatic triggering of skills in similar contexts) and high-road (mindful abstraction → explicit principle that gets applied across very different contexts). Different conditions produce each. Our pedagogy must consciously target which one we want.
- A direct empirical anchor for "CT transfer is a notorious problem" — they cite Pea & Kurland (1984) directly: "Although Papert (1980) spoke of LOGO as a source for the acquisition of generalizable thinking patterns and 'powerful ideas,' Pea and Kurland (1984) found no evidence of such as measured by transfer to planning tasks" (p. 114). This single sentence saves us from needing to track down Pea & Kurland separately.
The two mechanisms in one sentence each
"Low-road transfer primarily reflects extended practice; distance of transfer depends on amount of practice and the variability of contexts in which the practice has occurred. High-road transfer, on the other hand, depends on the mindful abstracting of knowledge from a context." (p. 115)
The three illustrative stories (p. 117–118)
Salomon & Perkins ground the abstract distinction in three vignettes — each maps cleanly to a kind of transfer we want for our adult digital-literacy users:
Car to truck (low-road). You've driven a car so much that the skill is automatic. Sliding into a truck cab triggers the same well-practiced behaviors with minor adjustments. Stimulus-controlled, automatic.
Sixth-grade study habits → adult work scheduling (high-road, forward-reaching). A teacher drilled "set aside time" so deeply in childhood that the principle now spontaneously suggests itself in new contexts as an adult. Abstraction was formed during original learning, "pops up" later.
Childhood "count to 10" → adult impulse-buying (high-road, backward-reaching). As an adult facing a new problem (impulse purchases), the person deliberately searches for relevant strategies and retrieves "count to 10" from a totally different domain (controlling tempers). Abstraction is formed at the moment of need by reaching back to past learning.
These three stories give us a vocabulary: when we say "we want users to transfer skills to novel digital contexts," we should be precise about which of these three we mean — because they require different design.
Low-road mechanism — what produces it
Two ingredients:
- Varied practice. "Practice that occurs in a variety of somewhat related and expanding contexts will force the cognitive element in question to adapt in subtle ways to each of these contexts, yielding an incrementally broadening ability that gradually becomes more and more detached from its original context and more and more evocable in others." (p. 120)
- Automaticity. "Processing and, by extension, behavior, becomes fast, effortless, and unlimited by processing capacity. It is carried out in well-integrated 'bundles' under stimulus control." (p. 121)
"The key aspects of this process of low-road transfer are varied practice and practice to automaticity." (p. 120)
Limitation: "The automaticity of mastery, attained as a result of much repeated practice, short circuits the link between situation and behavior." (p. 121) — meaning low-road transfer can misfire when the situation looks similar but isn't (their example: Americans driving on the wrong side in England — automatic right-side driving transfers negatively until conscious effort overrides it). For our domain this is the smartphone-to-desktop problem from a different angle.
High-road mechanism — what produces it
Two ingredients:
- Mindful abstraction. "The deliberate, usually metacognitively guided and effortful, decontextualization of a principle, main idea, strategy, or procedure, which then becomes a candidate for transfer; or, alternatively, the rarer case of learning of such a principle, idea, and so on, in abstract form in the first place." (p. 126)
- Mindfulness as metacognitive control. "By mindfulness we mean the volitional, metacognitively guided employment of nonautomatic ('controlled') processes." (p. 125)
"Some researchers see in the cultivation of metacognitive guidance the necessary, if not sufficient, condition for what we call here high-road transfer." (p. 126)
"The abstract formulation provides the bridge from one context to the other. But why mindfulness? The reason is that: (a) the abstraction must be understood, and (b) the understanding requires mindfulness; automatic processes just do not yield novel abstractions that are well understood. Having an abstraction in the superficial sense of possessing a rote abstract representation without understanding it will not suffice. For transfer, the abstraction must be genuinely comprehended, not just learned as a formula." (p. 126)
"Active learning wherein people achieve abstractions by themselves, although not necessarily producing better learning outcomes, facilitates farther transfer than so-called passive reception." (p. 126)
The Gick & Holyoak (1983) finding they cite (p. 127) is striking: subjects who wrote their own summaries comparing two analogous stories transferred to a new problem at 91%; subjects who wrote poor summaries transferred at 30%. The act of producing the abstraction in your own words is the discriminating variable.
Two sub-types of high-road transfer
- Forward-reaching: abstraction formed during original learning, then applies spontaneously later (study-habits story). This is what classroom instruction can engineer with explicit pattern naming.
- Backward-reaching: abstraction formed at the moment of need, by reaching back to prior learning (count-to-10 story). This is what real-world novel-task performance requires — and it is much rarer without explicit metacognitive habit.
Design implications for our project
Mapping Salomon & Perkins' two mechanisms onto our five design moves:
| Their mechanism | Our design move(s) | What it requires |
|---|---|---|
| Low-road | Cross-domain task families (varied practice) | Same skill in ≥3 surface forms; practice to fluency before moving on |
| High-road, forward-reaching | Explicit pattern naming + metacognitive debrief during original learning | Co-pilot names the principle, prompts learner to articulate when else it applies |
| High-road, backward-reaching | Far-transfer assessment in novel contexts | Test whether learner retrieves a prior pattern when facing a genuinely new problem |
| Both | Contrasting cases (Schwartz & Bransford) | Generates the differentiated structures both mechanisms can act on |
The mindful-abstraction finding is the strongest empirical anchor for our co-pilot's "name the pattern after success" move. Salomon & Perkins are explicit: passively receiving a ready-made abstraction transfers worse than producing your own. Implication for the AI co-pilot: don't just announce the pattern. Prompt the user to articulate it, then confirm and refine.
The automaticity-as-double-edged-sword finding (p. 121) is the empirical anchor for the Urban Institute smartphone-to-desktop failure. Smartphone users have low-road automaticity in a phone context. When the new context (desktop, file system, hierarchical organization) looks superficially similar but operates on different procedural rules, the automatic transfer misfires. Implication for the spec: explicit mental-model instruction must precede or accompany varied practice in early levels — otherwise users build automaticity on the wrong abstractions.
Direct quotes worth citing
"We argue that transfer is not at all a unitary phenomenon. Rather, transfer can occur by different routes dependent on different mechanisms and combinations of mechanisms." (p. 115)
"Some researchers see in the cultivation of metacognitive guidance the necessary, if not sufficient, condition for ... high-road transfer." (p. 126)
"Active learning wherein people achieve abstractions by themselves, although not necessarily producing better learning outcomes, facilitates farther transfer than so-called passive reception." (p. 126)
"The automaticity of mastery, attained as a result of much repeated practice, short circuits the link between situation and behavior." (p. 121) — i.e., automatic skills can transfer negatively when surface cues match but underlying rules differ.
Caveat
Salomon & Perkins is a 1989 theoretical paper — a synthesis and proposed framework, not a primary empirical study. Its predictions are qualitative ("low-road transfer occurs when..."). For our pedagogy doc this is exactly the right kind of source: a clean theoretical map we can use to organize our five design moves. For specific empirical effect sizes we lean on Schwartz & Bransford 1998, Bastani et al. 2026, and HPL II's syntheses.