The Science Behind duiduidui!

Beyond Spaced Repetition

Traditional flashcard apps use spaced repetition: show a card, wait a while, show it again. It works, but it's blunt. It doesn't know if you're overwhelmed. It doesn't understand that knowing two characters means you're probably ready for the word they form together. It treats every card as an island.

duiduidui! goes further. It uses a Bayesian proficiency model that tracks not just whether you know a card, but how confident the system is in that assessment. A card you've never seen is genuinely unknown. A card you've reviewed fifty times with mixed results is a known challenge. These are fundamentally different situations that call for different strategies, and duiduidui! handles them accordingly.

Two Layers of Intelligence

The learning system operates on two levels simultaneously:

Your overall skill level is tracked using Item Response Theory — the same mathematical framework used in standardized testing. As you review cards, the system builds a precise estimate of your Chinese proficiency, complete with confidence intervals. It knows not just where you are, but how sure it is about that estimate.

Your knowledge of each individual card is modeled separately using Bayesian statistics. Every character, word, and phrase has its own proficiency estimate that updates with each review. The system distinguishes between cards you've mastered, cards you're actively learning, and cards you haven't encountered yet.

These two layers work together. Your skill level guides which new content to introduce. Your per-card proficiency determines what to review and when. The result is a learning experience that feels effortless because the difficulty is always calibrated to your sweet spot.

Compositional Knowledge Modeling

Chinese is uniquely suited to compositional learning. Characters are made of radicals. Words are made of characters. Phrases are made of words. At every level, meaning builds from smaller parts.

duiduidui! models this structure explicitly. When you successfully review a phrase like 好吃 (hǎochī, delicious), the system doesn't just update that one card. It propagates evidence down to the components — 好 (hǎo, good) and 吃 (chī, eat) — adjusting their proficiency estimates too. And when you learn a new character, the system looks ahead to identify which words and phrases are now within reach.

This compositional approach means every review does double duty: you're learning the specific item in front of you while simultaneously building the foundation for everything that contains it.

Intelligent Card Selection

Choosing what to show you next is the most important decision the app makes, and duiduidui! takes it seriously. Rather than simple scheduling, the system evaluates every candidate card across multiple dimensions:

• Information value: How much will this card teach the system about your abilities? Cards near the boundary of your knowledge are the most informative.
• Learning value: How much will reviewing this card actually improve your Chinese? Cards in your optimal challenge zone — not too easy, not too hard — maximize learning.
• Cognitive fit: Do you have the mental bandwidth for this card right now? If you're near capacity, the system favors reinforcement over new material.
• Compositional readiness: Do you know enough of the components to have a fighting chance? The system won't show you a four-character phrase if you haven't seen any of its characters.

The weighting between these factors shifts dynamically. Early on, the system prioritizes assessment — figuring out what you already know. As uncertainty decreases, it shifts toward optimizing learning. You never notice the transition because it happens gradually, card by card.

Cognitive Load Management

Information overload is the silent killer of language learning. Most apps let you add cards until you're drowning in reviews, then blame you for not keeping up.

duiduidui! measures your cognitive load in real time, using a model grounded in information theory. Each active card contributes to your mental burden based on three factors: how uncertain you are about it (high uncertainty = high attention demand), how close it is to your skill level (harder cards relative to your level take more effort), and whether you're starting to master it (partially learned cards gradually release their hold on your attention).

Your total capacity grows as you advance. A beginner might handle 10 active cards comfortably; an intermediate learner, 100. The system manages this automatically, throttling new introductions when you're loaded and opening up when you have room. The result is a learning pace that always feels sustainable.

Difficulty That Makes Sense

Every item in the duiduidui! dictionary has a difficulty score that reflects how challenging it is to learn. For the foundational characters, these scores were hand-curated based on character complexity, usefulness, and real-world learning experience. For multi-character words and phrases, difficulty is calculated mathematically from the components — harder parts make for harder wholes, but the relationship isn't simply additive.

This creates an emergent curriculum: single characters build from simplest to most complex. Radicals appear after you've seen enough characters to recognize the patterns. Words and phrases naturally follow their component difficulties. You never have to wonder what to study next — the difficulty ladder is built into the content itself.