📊 Learning through conversation: what research says

Conversation as a learning tool is not a new idea. It is at the heart of the Greek philosophical tradition (Socratic dialectic), 19th-century pedagogies (Pestalozzi, Froebel), and major educational reforms of the 20th century (Dewey, Montessori). What is new is the ability to measure it rigorously. Cognitive neuroscience and education sciences over the past thirty years have produced a substantial body of work on the brain mechanisms that explain why learning through conversation works — and under what conditions.

Vygotsky and the zone of proximal development

The concept of the zone of proximal development (ZPD), introduced by Lev Vygotsky in 1934, is perhaps the most useful for theoretically understanding what a conversation with a historical character does. The ZPD refers to the gap between what a student can do alone and what they can do with the help of a more competent interlocutor. In the Vygotskian tradition, optimal learning occurs in this gap — when the learner is pushed slightly beyond their current abilities by a tutor who provides appropriate scaffolding.

A historical character on a platform like Parlova precisely plays this role of situated tutor. It does not give the answers — it situates them in their context. A student who asks Caesar why the Roman legions are so powerful does not receive a list of attributes (heavy infantry, discipline, military genius) — they receive an explanation from Caesar's own point of view, which presupposes certain knowledge and reveals others. This calibrated imbalance between what the student knows and what the character reveals corresponds exactly to the dynamics of the ZPD.

The Socratic method revisited by research

The Socratic method — asking questions rather than teaching answers — has been reevaluated by empirical research over the past twenty years. Studies conducted notably by Deanna Kuhn (Columbia University) on the development of argumentative thinking show that students exposed to structured Socratic exchanges develop significantly superior argumentation skills compared to those who receive the same content in lecture form.

The main reason identified by Kuhn is what she calls epistemic co-construction: in a Socratic dialogue, both interlocutors jointly construct the understanding of a problem. The learner is not the recipient of pre-constructed knowledge — they are an active co-constructor. This position generates deeper cognitive engagement and more flexible understanding, meaning it is more easily mobilized in new contexts.

Active retrieval vs passive memorization

One of the most robust findings in cognitive sciences over the past thirty years is the testing effect, also known as retrieval practice. The classic experiments by Roediger and Karpicke (2006) showed that students who test themselves on content — trying to recall what they have learned without looking at their notes — memorize this content two to three times better than those who simply reread the same content the same number of times.

Conversation with a historical character triggers a similar mechanism. When a student formulates a question to Caesar, they must first activate their knowledge about Rome to know what to ask — this is a form of retrieval. When Caesar responds with new or unexpected information, it is inscribed on a substrate of already activated knowledge, which promotes its integration into long-term memory. The conversation is an interactive retrieval practice.

Recent studies on educational AI

Studies on the use of conversational AI in education are still recent — most date from 2022-2025. But they converge towards a few clear conclusions. Studies on intelligent tutoring systems have shown since the 2000s that these systems produce learning effects comparable or superior to individual teaching by a human in specific contexts. More recent studies on LLMs (large language models) in educational contexts show variable results depending on how they are used.

The most positive results are obtained when AI is used in a Socratic rather than informative framework — that is, when it asks questions rather than gives answers, and when it is anchored in a specific rather than generalist context. A study published in the Journal of Educational Psychology in 2024 compared three conditions: lecture, reading a text, and guided dialogue with a historically situated AI character. The dialogue condition produced the best results on comprehension and long-term retention tests, particularly on analysis and transfer questions (applying knowledge to new contexts).

Practical implications for teachers

Research converges on four practical principles for teachers who wish to use conversation as a learning tool.

1. 1Structure before dialogue. Conversation works best when students arrive with activated prior knowledge. Preparation (short reading, sheet, discussion) is not optional — it is the condition for effectiveness.
2. 2Encourage questions that generate surprise. Questions to which students do not know the answer are more effective than verification questions. Genuine curiosity — not simulated — is the best state for learning.
3. 3Organize a systematic critical debrief. Conversation generates engagement. Debriefing generates understanding. Without collective synthesis, learning remains fragmentary.
4. 4Combine conversation and written trace. Writing after the conversation consolidates learning. Asking students to formulate what they have learned in one sentence or a short paragraph is one of the most effective interventions for long-term memorization.

These principles are not specific to Parlova — they apply to any form of conversational learning, whether with a human tutor, another student, or an AI character. What research confirms is that the structure of the interaction matters more than the technology that supports it. A well-designed dialogue with a historical character produces learning. A poorly designed dialogue with the same technology produces nothing — or produces the illusion of understanding, which is worse.