Skip to main content
Digital Fluency

Schwartz Bransford 1998 — Time For Telling

Research · foundational

Citation: Schwartz, D. L., & Bransford, J. D. (1998). A time for telling. Cognition and Instruction, 16(4), 475–522. URL / source: JSTOR scan of Cognition and Instruction (Lawrence Erlbaum). https://www.jstor.org/stable/3233709 PDF: schwartz-bransford-1998-time-for-telling.pdf in this folder.

Why this matters for our project

Direct citation for two of our five design moves:

  1. Contrasting cases — Schwartz & Bransford operationalize this exactly: have learners analyze a set of cases that differ on the dimensions you want them to attend to, before you tell them the underlying principle.
  2. The Bastani-style "refuse-to-tell-until-effort" pattern — the paper is essentially an empirical argument that direct instruction (telling) is most effective when it follows a period of preparation that surfaces what the learner doesn't know. This is the cognitive-science foundation for the AI co-pilot rule "encourage user attempt first."

It also frames a third design move implicitly: pattern naming after struggle. The lecture/text in their experiments works best when it gives names and a unifying frame to distinctions the learner has already noticed.

Central thesis

Constructivist instruction is often misread as "discovery learning" (figure it out yourself, no telling). Schwartz & Bransford reject that read and offer a sharper claim: telling, lectures, and texts can be extremely effective — but only when learners have first generated differentiated knowledge structures that prepare them to receive the telling.

"When students have had an opportunity to generate well-differentiated knowledge about a domain, then teaching through a lecture or text can be an extremely effective form of instruction." (p. 476)

"There are points of knowledge development that are indicative of a 'time for telling' or a 'readiness' for being told something." (p. 476)

The paper is asking: what produces that readiness?

The mechanism: differentiated knowledge structures from contrasting cases

Their core hypothesis: analyzing contrasting cases — comparing exemplars that vary on the dimensions you want learners to attend to — develops the differentiated structures that make later telling productive.

The scissors illustration (p. 477–479) makes this vivid: a novice and an expert both "understand" the sentence "the dressmaker used the scissors to cut the cloth." But the expert's representation is differentiated — they know dressmaker's shears differ from barber's shears differ from nail scissors in specific structural ways tied to function. The novice has a generic "scissors" schema. Direct instruction about scissors design lands very differently on those two readers.

"Novices can easily think they understand when, in reality, they have missed important distinctions... A scissors expert will have a more finely differentiated concept of scissors than most casual comprehenders." (p. 477)

"Contrasting cases help people notice specific features and dimensions that make the cases distinctive." (p. 479)

Critically, the contrasting-cases analysis alone is usually not enough. It produces noticing, but not theory. Telling (the lecture / text / co-pilot explanation) provides the unifying framework that lets the learner make sense of what they've noticed.

"These activities by themselves, however, are usually insufficient for students to induce the principles necessary to understand a domain at a satisfactorily deep level. This is because novices often lack an overriding framework that helps them develop a theory or model to explain the significance of the distinctions they have discovered. This is one place in which telling can have powerful effects on people's abilities to learn; it can help them make sense of the distinctions that they have noticed." (p. 480)

Experimental design (Experiment 1)

Result (foreshadowed in abstract; see pp. 488 onward in the paper)

The contrasting-cases-then-lecture treatment outperformed:

The crossover design controls for individual differences, motivation, and content — students performed better on the half of the lecture that complemented cases they had analyzed than on the half that complemented cases they had only read about.

The verification (true/false) test did not discriminate between conditions. The transfer-to-prediction task did. Standard recognition assessments, in their words, "may not capture deeper levels of understanding" (p. 486).

Direct quotes worth citing in pedagogy.md

"When telling occurs without readiness, the primary recourse for students is to treat the new information as ends to be memorized rather than as tools to help them perceive and think." (p. 477)

"Contrasting cases help people notice specific features and dimensions that make the cases distinctive." (p. 479)

"Telling can have powerful effects on people's abilities to learn; it can help them make sense of the distinctions that they have noticed." (p. 480)

"The cases ... involve opportunities to actively differentiate among exemplars of psychological phenomena and concepts ... students tended to develop an overly superficial understanding of the experiments and concepts, limiting their ability to transfer to new situations. These experiments use the method of contrasting cases to help students develop knowledge that is more differentiated than we have been able to achieve in the past." (p. 480)

Implications for our design

  1. Co-pilot intervention timing: "tell" only after the user has tried — empirically grounded. Telling without readiness produces memorization, not transfer (p. 477).
  2. Contrasting cases as a pedagogical move: present near-neighbor tasks that vary on the dimension we want users to attend to. Don't describe the difference; let the user notice it through analysis, then name it.
  3. Pattern naming after the fact: the co-pilot's job after a contrasting-cases sequence is to name the unifying principle the user has noticed but not yet articulated. This is the "time for telling."
  4. Assessment design: standard recognition assessments (true/false, multiple choice, completion checks) may pass without indicating real learning. Far-transfer assessment (predicting a novel scenario) is the discriminating signal. This validates our planned distinction between near-transfer and far-transfer metrics in the spec's Assessment Engine.
  5. For our adult-novice population: the differentiated-knowledge-structures argument suggests our task design should explicitly construct readiness for AI-assisted explanation. Throwing a help-text at a struggling user is "telling without readiness" — exactly the failure mode this paper warns against.

Caveat on transfer

Schwartz & Bransford explicitly note (p. 480) that "students tended to develop an overly superficial understanding ... limiting their ability to transfer to new situations." Contrasting cases address this — but the paper does not solve far transfer in general. It demonstrates that one specific instructional move (contrasting cases before telling) produces better transfer than alternatives, in undergraduates, in cognitive psychology, on a one-week-delayed prediction task. The mechanism plausibly generalizes to our domain, but the population and topic don't.