How to Use Chinese Corpora Without Misreading Frequency

Why this article matters

A corpus can make learners smarter, but only if they stop treating frequency as truth. “This word appears 10,000 times” means little until you know where the corpus comes from, how it was segmented, what genres it contains, what period it covers, and what exact form you searched. Corpus literacy is a serious skill.

Core concepts

The article

Chinese corpora are powerful because they let learners inspect real examples. Instead of asking “Can I say this?” you can ask “Where does this expression appear, with what words around it, in what genre?” That is a better question. But corpus results can also mislead.

The first danger is corpus composition. A news-heavy corpus will overrepresent policy, institutions, formal reporting verbs, and official place names. A social-media corpus will overrepresent slang, fragments, emotional stance, and platform events. A literary corpus will overrepresent narrative description and older words. If you compare frequencies without knowing the source mix, you may think a word is common in everyday speech when it is actually common in news headlines.

The second danger is segmentation. Written Chinese has no spaces, so corpus tools must decide where word boundaries are. Is 研究生物 one word sequence meaning “graduate student biology,” or 研究 / 生物 meaning “study biology”? In real search, segmentation decisions affect counts, collocations, and examples. A string search can catch forms that a word search misses; a word search can avoid false hits that a string search includes.

The third danger is form variation. Simplified/traditional variants, regional vocabulary, old spellings, abbreviations, and named entities can split results. Searching 信息 may not tell you about 資訊 in Taiwan materials. Searching 普通话 may not tell you about 國語 or 华语. Serious corpus work requires variant awareness.

The fourth danger is raw count worship. A rare word is not always wrong. A common word is not always appropriate. Some words are rare because they belong to narrow domains; others are common but vague. Some phrases are frequent because of one repeated news formula, not because native speakers casually use them in daily life.

Worked example: 方便, 便利, 便捷

If you search these near-synonyms, raw frequency may not answer the learner's real question. The useful question is: what nouns and verbs does each word appear with?

The corpus is useful because it shows environment, not because it prints a magic answer.

Learner traps and repairs

Practice protocol

Choose one word pair: 问题/议题, 方便/便捷, 认为/表示. Search in a corpus or a controlled source set. Record five examples from news, five from social media or conversation if available, and five from technical/official text. Write a register note before writing an English gloss.

Additional practice and repair

Corpus misuse diagnostics

Query log template

Before/after repair set

The corpus worksheet should require users to save at least three concordance lines with source metadata before writing a usage note. Add warnings for duplicate boilerplate, named-entity noise, and one-source overrepresentation. Include a simplified/traditional variant reminder.

Practice visualization

Build a corpus-query worksheet with fields for query form, corpus/source, genre, date, region, segmentation mode, raw frequency, top collocations, example quality, and caution note.

Check corpus-tool claims against BCC, CCL, Unicode text segmentation, and ICU dictionary-based boundary analysis. Be clear that corpus evidence is empirical but not self-interpreting.