Measure Your Chinese Typing Improvement: A Personal Testing Protocol

To find out whether Chinese typing is becoming easier, keep the original task, your first-pass text, and the time spent checking it. A score alone cannot show whether you entered words more smoothly, skipped a line, or simply recognized a passage you had already practised.

This is a personal self-measurement protocol, with a local recording sheet and worked examples. It is not a report of a TypeChinese improvement study. The numbers below are fictional demonstrations; they are not participant results or evidence that a particular routine causes improvement.

Our Chinese typing speed benchmark answers how TypeChinese scores compare within its sample. This article answers a different question: how can you audit your own copying and correction process, including the time needed to produce checked text?

Choose one task you want to become better at

Write a concrete question before collecting scores:

Can I copy an unfamiliar short Traditional Chinese message, then check it accurately, using my usual desktop input method?

Copying keeps the intended text available for comparison. Composing a new message adds vocabulary recall, planning, and grammar decisions. Keep composition in a separate log if that is your real goal.

Choose an editor that lets you preserve ordinary text without automatically replacing it. A local text editor and a stopwatch are enough. You do not need an account, a spreadsheet service, or a public score.

For this protocol, count Han characters in the source, excluding punctuation, spaces, digits, and Latin letters. You must still reproduce and check those other symbols; they contribute to elapsed time even though they are outside this particular numerator.

For example, 我明天上午去圖書館。 contains 9 Han characters and one full stop. Its counted length here is 9. Do not use the editor's “word” count as a substitute without checking its definition.

Keep probes to ordinary Chinese characters and punctuation at first. Unicode distinguishes encoded characters from user-perceived characters, which can contain several code points; see Unicode's text segmentation specification. Rare variants, combining marks, and emoji need a separate counting rule before you add them to a test.

Freeze the conditions in a short setup note

Keep this note with your records:

Use the same short warm-up before each measurement. Keep it out of the recorded task. Choose a warm-up duration you can sustain, such as two minutes, and record it rather than extending it only on days when typing feels slow.

You can keep the input method's normal prediction enabled. Record that choice. Its dictionary can learn from your use, so the result describes you working with that configured system; it does not isolate finger skill from software assistance.

If you change computers, input methods, keyboard layouts, or help settings, start a new condition label. A more readable candidate window may make your setup more usable, but its before-and-after scores should not silently become one unchanged series.

Prepare unseen probes separately from practice

A probe is a short copying task reserved for measurement. Once you have read or practised it, mark it as seen. It can remain useful practice, but it cannot become an unseen probe again.

Ask a partner to prepare a small bank that follows the same brief. If you work alone, choose a source you have not read and reveal only the next section at test time. Record the source and familiarity honestly; you cannot fully blind yourself to text you wrote.

For a first bank, use this practical brief:

• One script throughout, with approximately 80–120 Han characters per probe
• Everyday messages with vocabulary you can read, rather than unfamiliar literary passages
• Similar sentence lengths and punctuation density
• No personal names, URLs, numerals, or rare characters in the initial bank
• Different wording and situations, with no sentence copied from the warm-up or practice material

These constraints make tasks more comparable; they do not establish that every probe has equal difficulty. Keep the full sources so you can investigate an unusually easy or hard result later.

Allocate probes in advance to a baseline block and later blocks. Use each only once for unseen measurement. Reserve another group of sentences for practising difficult words between blocks.

If you cannot prepare a bank, use new passages from one TypeChinese category and keep its scores in a separate app log. Save enough passage information to identify repeats. Categories contain varying text; a shared category name does not guarantee matched difficulty.

Run the same entry, snapshot, and review sequence

1. Open the source and a blank local editor. Confirm the intended input method. Start the entry timer immediately before your first keystroke.
2. Copy the complete source. Correct mistakes normally during entry; do not deliberately leave them to improve the timer result.
3. When you believe the copy is complete, stop the entry timer. Preserve this first-pass output before repairing anything else. Record the elapsed seconds.
4. Start a separate review timer. Compare the preserved output with the source, from the beginning, including punctuation and line breaks. Repair a working copy until it matches.
5. Stop the review timer when the final copy is verified. Record its elapsed seconds and the remaining mismatches found in the preserved first pass.
6. Keep the source, first pass, final copy, timings, and notes together. Record interruptions even if the result is disappointing.

Exclude the mechanical time spent duplicating or saving the snapshot from the review timer. Apply that rule every time. If you already started examining the text while saving it, note the inconsistent timing rather than pretending the separation was exact.

If you stop before completing a probe, retain the attempt as incomplete and do not calculate completed-task throughput. Decide in advance what counts as an interruption, such as a phone call, and use the same rule for every block. Keep excluded attempts in the raw log with a reason.

Rehearse the procedure with this original source

Use this Traditional Chinese message to practise the six steps before starting an unseen measurement block:

今天下午，我們先到圖書館整理資料，再把需要的文件放進同一個資料夾。請確認每份文件的標題和日期，並記下還要修改的地方。離開以前，把借來的書放回服務台，也提醒同伴帶好筆記。若有不清楚的內容，先提出問題，再決定下一步。

The source contains 95 Han characters and 11 punctuation marks, with no spaces, digits, or Latin letters. Its counted length for this protocol is 95; the complete text contains 106 code points. The paragraph above is also the correction key: the final copy should reproduce it exactly.

Use rehearsal-01 as the attempt ID. Record your actual entry and review seconds, preserve the first pass, and calculate 95 × 60 ÷ (entry seconds + review seconds) only after checking the final copy. If the first pass already matches, keep the review time and record zero remaining mismatches.

This is a seen rehearsal exercise, not a held-out probe. Once you have read this page, do not put this paragraph into your unseen baseline or later measurement bank. Its purpose is to check that your timer, snapshot, counting rule, and recording sheet work together.

Separate remaining errors from difficulties you remember

Your first-pass snapshot shows what remained when you thought the copy was finished. It cannot reveal every wrong key or candidate you corrected during entry.

This distinction matters in text-entry research: Soukoreff and MacKenzie's original error-metrics paper separates corrected errors from errors left in the transcribed text. The simple diary here does not capture an input stream and does not calculate that paper's total error rate.

For short, clear examples, classify remaining mismatches as follows:

With duplicated or rearranged text, more than one alignment may be plausible. Save the exact mismatch and describe it, such as “copied the second clause twice.” Do not force an ambiguous passage into a precise error total.

Add a separate diagnostic note immediately after the task:

• Reading or spelling: I could not recall the pronunciation or spelling of the intended word.
• Code recall: I recognized the character but had to look up or reconsider its shape code.
• Candidate choice: I entered a code but selected or searched for the wrong candidate.
• Copying position: I lost my place, skipped text, or copied it twice.
• Editing: I found a mismatch but struggled to select and replace just that span.

These are your observations, not automatically measured causes. Use “uncertain” when you cannot tell. Do not turn remembered pauses into a claimed keystroke count or accuracy percentage.

Calculate checked-text throughput

Use one deliberately named local measure:

Checked Han characters per minute = source Han count × 60 ÷ (entry seconds + review seconds).

Only calculate it after verifying the final copy. Keep the two times visible beside the result: a shorter entry time can still come with more review work.

Fictional calculation example

Suppose a completed source has 120 Han characters. The fictional entry time is 180 seconds and review plus repair takes 20 seconds.

• Combined time: 180 + 20 = 200 seconds
• Entry-only Han rate: 120 × 60 ÷ 180 = 40
• Checked-text Han rate: 120 × 60 ÷ 200 = 36

Neither rate is a TypeChinese score. TypeChinese's current CPM uses the passage's JavaScript text length, including punctuation, and its elapsed practice time. This local measure uses a different numerator and adds a separate review stage. Keep the columns and comparisons separate.

Here is another fictional pair, with the same counted source length:

B has a faster entry stage, but both take 200 seconds to reach checked text. The larger review burden is useful information that an entry-only score would hide. These invented values illustrate the arithmetic; they do not describe a real learner.

Keep a raw recording sheet

Copy these fields into a local note or spreadsheet. Use a row per attempt and store the three text files beside it.

For example, a fictional row might read: A / condition-1 / probe-04 / unseen / 120 Han / 180 s / 20 s / 1 substitution / 36.0 / candidate choice / complete.

Preserve the first-pass text even after fixing it. If a number looks unusual later, that evidence lets you check for skipped clauses, a counting mistake, or inconsistent timing.

Read the pattern before choosing the next exercise

Compare whole blocks of unseen probes, including their range of results and error notes. A block with fewer copying mistakes but similar throughput may still address your original problem. One unusually fast task is a reason to inspect its source, not enough to declare a trend.

If entry is the main burden and notes repeatedly mention candidate choice, practise a small set of those words using your normal keyboard. If review dominates, work on line-by-line checking and precise cursor placement. If reading itself is difficult, choose more familiar language before treating the issue as keyboard speed.

TypeChinese can supply separate practice between measurement blocks: start with Simplified HSK 1 vocabulary or Traditional conversational passages. Use keyboard entry for keyboard practice; keep dictation, pasted text, and other assistance out of that series.

After changing one part of your routine, record what you changed and continue with unused probes. Personal before-and-after observations cannot establish that the change caused a gain: text difficulty, IME learning, fatigue, and other practice may also change. Report what your records actually show, and keep collecting comparable evidence before drawing a stronger conclusion.