InkChinese

The story in the video is 门上的春联, recorded with a test profile. The app is private to my family, behind a Google sign-in.

The problem

My 6-year-old hears and speaks Chinese at home but is only starting to read it. That is the typical profile of a heritage learner in the US: a large spoken vocabulary and weak literacy. Research on US Chinese–English bilingual children finds that speaking tracks conversation at home, while reading tracks exposure to print and practice with a phonetic script (Ke, Xia & Zhang, 2023).

So the bottleneck is not vocabulary. It is connecting characters to the sounds and meanings the child already knows. Flashcards teach characters without a story, and audiobooks tell the story without asking the child to do anything. I wanted short lessons a child can finish alone, in words they already say, that build reading one character at a time.

The lesson: four steps

Every story is a lesson in four steps, shown as four buttons at the top of the screen. A step opens only after the one before it is done.

1. 听 Listen. A natural Mandarin voice reads each page while the characters light up one by one, then the page turns itself.
2. 字 Characters. The lesson's new characters come up on a practice grid, on the page where they appear, and the child writes each one stroke by stroke with a finger or an Apple Pencil.
3. 读 Read. The child reads the page aloud. Speech recognition follows along character by character; if the child is stuck, the character pulses, and after a pause the app says it and moves on.
4. 复 Review. Five mixed questions: pick the character you heard, complete a word from the lesson, fill the blank in a sentence from the lesson, and write a character from memory. Four out of five passes.

Why these steps: what the research says

Before designing anything I read the research on how children learn to read Chinese, and rated how strong each finding was. Each step answers a specific finding.

• Listening first, with the text in view. Studies of shared and repeated reading with picture books in mainland China and Hong Kong found that young children learn many characters incidentally from rereading, and can read them later in new contexts (Beijing grade 1; a one-year study of 300 lower-primary pupils). So narration with highlighting is a learning step, not filler.
• Writing, but only a little. Reading Chinese depends strongly on writing: orthographic awareness and motor memory matter more than in English (Tan et al., PNAS 2005). Current mainland textbooks follow 多认少写, recognizing many characters and writing few. So writing starts on day one, but each lesson writes at most 12 new characters while the child reads many more.
• Focused practice, then straight back into the story. I found no rigorous head-to-head study of teaching characters on their own (集中识字) versus inside texts (随文识字); the widely cited comparisons are single classrooms or teacher surveys. Smaller experiments suggest isolation helps children learn a character's shape and context helps its meaning. So each new character gets a short focused moment on the pad, on the same page where it appears in the story.
• Characters before pinyin, and pinyin that fades. The 2016 mainland textbook reform moved character learning ahead of pinyin, so pinyin no longer gates reading (report). Taiwan's grade-1 zhuyin teaching transfers to character recognition, but remedial zhuyin alone did not improve comprehension. So pinyin, or zhuyin for traditional script, is a bridge: optional, and it fades on characters the child already knows.
• Reading aloud. Programs for heritage learners put reading and writing first for children who already speak at home (CAL program profile). Speaking is practiced as reading aloud, so the child connects the print to the voice they already have.
• Short, mixed review. The general research on spacing and retrieval practice is strong, but trials specific to Chinese characters in children are small. So review is a short, mixed quiz at the end of each lesson, not a flashcard engine. The write-from-memory question first shows the character for three seconds and says it, because a sound alone matches many characters and would not be a fair question.

How the curriculum is built

The catalog is the hard part of a reading app, and most of the work went into rules for writing and placing stories.

Follow a real textbook. The spine is 暨南大学《中文》, a series used in many overseas weekend Chinese schools. Its five volumes are the app's five levels, with required textbook lessons in the book's own order. Companion stories sit in optional units after them. Finishing every required lesson at a level unlocks the next level's lessons and stories. Optional stories never hold a child back.

Two lists per level. Each volume has a list of characters it explicitly teaches (the 生字表: 133 in volume 1, 275 cumulative by volume 2) and a larger set of characters that actually appear in its reading texts (236 in volume 1, 433 cumulative by volume 2). Lessons teach and test from the first list. New stories are written within the second, so the language stays natural without going beyond what the child has met.

A coverage rule for every new story. Comfortable independent reading needs roughly 95–98% of the words to be known (Hu & Nation, 2000; widely used, though debated). Companion stories target 98% of characters from the level's list, with a hard floor of 95% of all character occurrences and 95% of distinct characters. That second floor stops a story from hiding new characters behind lots of easy repetition. A level 1 story may use at most one character from outside the list and a level 2 story at most two, each one declared with its word and meaning and supported by audio or the picture.

Length and sentence limits by level.

	Level 1 story	Level 2 story
Body length	10–50 characters	51–160 characters
Distinct characters	at most 30	at most 60
Longest sentence	12 characters	24 characters
Story shape	one familiar scene, repetition with a small change	one goal or problem, 2–3 actions, a visible outcome
Comprehension check	who or what; match an action	order the events; explain one cause stated in the text

Review is built into the writing. Each story picks 3–5 characters already taught, uses them naturally 2–4 times, and they come back across 3–5 stories. A page holds at most 16 characters.

Page turns follow the story, not the punctuation. Each page is one meaningful moment: an action, a discovery, a change of feeling. A setup and its reveal go on different pages, and a reaction stays with the action it belongs to. Stories are storyboarded and read aloud before any art or narration is made.

What this is not. These limits are my editorial choices, informed by the research above and checked against how my own child reads. They are not published proficiency thresholds, and a review by teachers and native speakers is still to come.

Features

• 434 books across five levels: the textbook articles, short companion stories and classical poems.
• Pinyin that fades on characters the child already knows, or traditional script with zhuyin.
• Tap any character to hear it and see its stroke order.
• Icon-first, Mandarin-only interface. Long-press any button to hear what it does, so it works before the child can read the menus.
• Progress per child, synced across devices, so each lesson picks up where it left off.

How it's built

• Reader: Vite, React and TypeScript, laid out for the iPad.
• Content pipeline: Python and one SQLite catalog. It adds pinyin and zhuyin, splits stories into pages, and checks every story against its level's character list and limits before any audio is generated.
• Narration: the Mac's own Siri Mandarin voice, with per-character timings so the highlight follows the voice.
• Writing: stroke data from hanzi-writer.
• Built with Claude Code, starting October 2026, by watching real lessons and fixing whatever got in the way.

The hardest part

Getting every page to agree with itself. Many Chinese characters have more than one reading, and the pinyin, the narration, the tap-to-hear clip and the quiz all have to use the reading that fits this sentence. A full audit of the catalog corrected the readings in nearly 200 books, and the rule now is to annotate the whole story before it is split into pages, so a page break never loses the context.