Someone claims dictation tests are “scientifically worse” than free tracing tests, and it is worth taking seriously, because the answer turns entirely on what the words mean. Under the usual definitions, the claim is backwards. Here is what dictation and tracing actually test, and what the science really favors.
Define the terms first
The confusion is definitional, so pin it down:
- Dictation, in the usual sense, means producing a character from a prompt that is not the character itself, a spoken word, a meaning, a gap to fill, with nothing to copy. That is production from memory.
- Tracing usually means following a model that is shown to you, drawing over or beside a visible character.
- Free tracing is ambiguous: if it means tracing a shown model, it is recognition; if it means writing freely from memory with no model, it is actually the same as dictation.
So before judging which is “worse,” you have to know whether tracing means copying a model or writing from nothing.
Under the usual definitions, the claim is backwards
If dictation means producing a character from memory and tracing means following a shown model, then dictation is the better test, not the worse one. Producing a character from memory is recall, which the testing effect shows builds and demonstrates durable learning, and generating it yourself rather than copying engages the generation effect. Tracing a visible model is recognition-adjacent: the answer is in front of you, so it neither builds nor proves the ability to produce the character. For Chinese specifically, handwriting beats typing for learning words precisely because production beats selection. So the science favors dictation over model-tracing, the opposite of the claim.
When the two are actually the same
If “free tracing” means writing freely from memory without a model, then it is not really tracing at all; it is from-memory production, the same thing good dictation tests. In that case there is no meaningful difference, because both require recall. The apparent disagreement dissolves once you notice both describe producing the character with nothing to copy, which is the strong test either way.
What actually makes a writing test good
| Test | What it requires | Strength |
|---|---|---|
| Dictation (from a prompt, no model) | Recall, production | Strong |
| Free writing from memory | Recall, production | Strong |
| Tracing a shown model | Following, recognition | Weak |
| Multiple choice | Recognition | Weakest |
The pattern is clear: any test that makes you produce the character from nothing is strong, and any test where the character is shown is weak, which is the deeper point behind why OCR and translation worsen character amnesia and why Hack Chinese does not test handwriting.
Why this matters for practice
If you believed dictation was worse and switched to tracing a model, you would be trading a strong test for a weak one and wondering why your handwriting was not improving. The correct takeaway is the reverse: lean into dictation-style, from-memory production, whether you call it dictation or free writing, and avoid mistaking model-tracing for a real test, the same trap as more gamified stroke modes that still only trace, or apps people leave like in quitting Skritter after an update and seeking a modern replacement.
A plan for strong writing tests
- Define your test: are you producing from memory or copying a model?
- Favor dictation-style prompts: a word or meaning, character hidden.
- Write the character from memory, with nothing to trace.
- Check stroke order and structure afterward.
- Avoid mistaking tracing a shown model for a real test.
How Hanzi Write Practice fits
Hanzi Write Practice is built on the strong side of this. It hides the character and prompts you to produce it from memory, which is exactly dictation-style, then checks stroke order and structure with spaced repetition. It does not let you trace a shown model and call it a test, because that would only build recognition. So whatever you call it, the practice is from-memory production, the test the science favors, on the foundation of the case for a writing app.
Bottom line
The claim that dictation is worse than free tracing is usually backwards: dictation, producing a character from a prompt with no model, is recall, which the science favors, while tracing a shown model is recognition, which is weak; free from-memory writing and good dictation are the same strong test. Hanzi Write Practice is built on that production, and it is in early access, so join the list.
Frequently asked questions
Are dictation tests really scientifically worse than free tracing tests?
Usually it is the reverse. If dictation means producing a character from a prompt with nothing to copy, it is recall, which the testing and generation effects show builds durable learning, while tracing a shown model is recognition, which is weak. If “free tracing” means writing freely from memory with no model, then it is the same as dictation. Either way, the strong test is producing the character from nothing, which is what Hanzi Write Practice is built on.
What is the difference between dictation and tracing?
Dictation has you produce a character from a prompt that is not the character itself, such as a spoken word or a meaning, with nothing to copy, so it tests recall. Tracing usually means following a model that is shown to you, so it tests recognition. The presence or absence of a model to copy is the key difference.
Why is producing from memory better than tracing a model?
Because producing the character yourself engages recall and the generation effect, building a durable trace and proving you can write it, while tracing keeps the answer in front of you and only trains recognition, which fades. Any test where the character is shown is weaker than one where you must produce it.
Is free writing from memory the same as dictation?
Essentially yes. Both require producing the character with nothing to copy, which is recall, so they are the strong form of testing. The disagreement in the original claim usually comes from an ambiguous use of “tracing,” which can mean either copying a model or writing freely from memory.
Want a test that builds writing? Join early access and produce characters from memory.