教学错误样本分析
Chinese Grammar Errors: Analysis of 988 Teaching Examples
A reproducible breakdown of 988 curated Chinese error examples by explanation topic and source-word level, with CSV, methods and limitations.
Scope: these are AI-assisted, checker-reviewed teaching examples, not real learner submissions. Percentages describe this corpus only. The level is the source word’s HSK 2.0 level, falling back to its 2021 list level; it is neither a learner’s tested proficiency nor the 2025 exam level.
Download all 988 classified examples (CSV) · Read examples by level · Try the quiz
样本分布Which topics appear most in this teaching corpus?
Of 988 unique wrong sentences, 320 (32.4%) receive the primary topic “Word order & placement”. These figures measure editorial coverage. They do not estimate a real learner error rate.
| Topic | Count | Share | Source level 1 | 2 | 3 | 4 |
|---|---|---|---|---|---|---|
| Word order & placement | 320 | 32.4% | 55 | 57 | 93 | 115 |
| Other / uncategorised | 168 | 17.0% | 17 | 25 | 39 | 87 |
| Measure words & quantities | 146 | 14.8% | 42 | 28 | 38 | 38 |
| 的 / 地 / 得 & modifiers | 115 | 11.6% | 18 | 16 | 29 | 52 |
| Word choice & collocations | 109 | 11.0% | 11 | 14 | 20 | 64 |
| 了 / 过 / 着 & aspect | 47 | 4.8% | 3 | 8 | 11 | 25 |
| 是 & adjective predicates | 43 | 4.4% | 7 | 2 | 7 | 27 |
| Questions | 15 | 1.5% | 4 | 4 | 3 | 4 |
| Negation | 9 | 0.9% | 2 | 0 | 1 | 6 |
| Complements & verb constructions | 8 | 0.8% | 1 | 0 | 1 | 6 |
| Connectors & clause patterns | 6 | 0.6% | 0 | 1 | 3 | 2 |
| Comparisons | 2 | 0.2% | 0 | 0 | 2 | 0 |
| Total | 988 | 100.0% | 160 | 155 | 247 | 426 |
Each example is counted once. Percentages are rounded to one decimal place and may not sum to exactly 100%.
例句Examples behind the numbers
Word order & placement: 320 examples
✗ 不我认识他。
✓ 我不认识他。
不 must come directly before the verb, not before the subject. Source word: 不
Other / uncategorised: 168 examples
✗ 我们们是朋友。
✓ 我们是朋友。
我们 is already plural, so 们 cannot be added again. Source word: 我们
Measure words & quantities: 146 examples
✗ 我买了苹果一个。
✓ 我买了一个苹果。
The quantity phrase 一个 must come before the noun, not after it. Source word: 了
的 / 地 / 得 & modifiers: 115 examples
✗ 我想很累。
✓ 我觉得很累。
想 does not take an adjective predicate like 很累 to describe your own state; use 觉得 for "feel". Source word: 想
Word choice & collocations: 109 examples
✗ 我今年二十年。
✓ 我今年二十岁。
年 is used for years and dates, but age is expressed with 岁. Source word: 年
了 / 过 / 着 & aspect: 47 examples
✗ 我着看书。
✓ 我看着书。
着 must follow the verb; it cannot come before it. Source word: 着
是 & adjective predicates: 43 examples
✗ 我在家是。
✓ 我在家。
在 is itself the main verb 'to be at', so no extra 是 is needed: 我在家 means 'I am at home'. Source word: 在
Questions: 15 examples
✗ 你叫什么吗？
✓ 你叫什么？
吗 is for yes-no questions, not for questions that already have a question word like 什么. Source word: 吗
Complements & verb constructions: 8 examples
✗ 我说话中文。
✓ 我说中文。
说话 is an intransitive verb-object phrase and cannot take an object like 中文; use 说 + language instead. Source word: 说话
Connectors & clause patterns: 6 examples
✗ 所以下雨，我没去。
✓ 因为下雨，我没去。
所以 marks the result, not the reason; use 因为 to introduce the cause. Source word: 所以
Comparisons: 2 examples
✗ 我一样你。
✓ 我跟你一样。
一样 is not a verb that takes an object; use 跟/和 + person + 一样 to compare people. Source word: 一样
方法与限制Method, provenance and limitations
- Use the reviewed “Common mistakes” entries attached to the existing HSK word pages. Generation and automated verification are described in the public enrichment script.
- Deduplicate by the exact wrong sentence. Where a sentence occurs under more than one word, the first source word in the vocabulary data supplies the level and explanation.
- Retain the first correction when an editorial field lists slash-separated alternatives. Source word levels use HSK 2.0 first, then the 2021 list.
- Match the English explanation against the ordered rules below. Assign the first matching topic; preserve unmatched examples as “Other / uncategorised”. This is a transparent topic heuristic: overlapping rules, Chinese strings inside examples, and wording choices can affect assignments.
The corpus is selected for teaching utility, not randomly sampled. AI verification can be wrong; some examples span more than one error type. No learner identities, live checker inputs or browser mistake histories were collected. No confidence intervals or population-level claims are warranted.
Classification rules (in priority order)
- Measure words & quantities:
measure word|classifier|quantity phrase|numeral|count(?:ing|able)|number[- ]measure - 的 / 地 / 得 & modifiers:
的|地|得|possessive|attributive|adverbial marker|degree complement - 了 / 过 / 着 & aspect:
了|过|着|aspect|perfective|experiential|past[- ]tense|completed action - Word order & placement:
word order|before|after|precede|follow|position|placement|between|at the (?:end|start|beginning) - Negation:
negat|negative|\bnot\b.*不|不.*没|没.*不 - Questions:
question|interrogative|吗|呢|几|多少 - 是 & adjective predicates:
是|copula|adjective|stative - Complements & verb constructions:
complement|resultative|directional|separable|verb[- ]object|把|被 - Comparisons:
compar(?:e|ison|ative)|比|更|最 - Connectors & clause patterns:
conjunction|connector|clause|因为|所以|虽然|但是|如果|不但|而且 - Word choice & collocations:
means?|meaning|use[ds]? (?:for|with|to)|collocation|instead|refer|cannot (?:take|be used)|can't (?:take|be used)
Report and CSV: CC BY-SA 4.0, attribution SentenceCoach / Cheng Xu, snapshot 2026-11-04, classification method v1.0. This is a separate teaching-error report; it does not change the version or DOI of our existing sentence-ordering dataset.
Common questions
What do learners of Chinese get wrong most?
This teaching corpus cannot answer that population question. Its largest categories show which topics our curated examples cover most, not which errors learners make most often. A representative study would need independently sampled real learner writing.
Are these real learner submissions?
No. The examples were drafted with AI assistance for SentenceCoach word usage notes and checked by the sentence checker. They are instructional examples, not observed student mistakes, human-labelled research data, or anonymous counts from live checks.
How are the examples classified?
A deterministic first-match rule assigns one primary topic by matching keywords in each English explanation. Unmatched explanations are retained as Other / uncategorised. The CSV exposes every assignment for inspection; the method is heuristic, not a manually adjudicated linguistic annotation.