REFERENCE · QUERY

Query and execution options

Expansion selects the generated forms. POS selects the grammatical rules, and the boundary policy selects admissible source candidates.

Default query plan

With no options, kfind expands inflections, infers POS from its lexicons, and applies the POS-specific smart boundary. It normalizes the query to NFC and permits 24 Unicode scalars between phrase atoms.

expand=inflection
pos=auto
boundary=smart
unicode-normalization=nfc
max-gap=24

Automatic analysis preserves the union of valid POS analyses for an ambiguous lemma. An explicit POS narrows that set; surface spelling alone never forces an unknown input into verb or adjective rules.

Morphological expansion

Value	Coverage	Example
`literal`	Input string only	`걸어 → 걸어`
`inflection`	Particles, endings, allomorphs, and irregular conjugation	`걷다 → 걷고 · 걸어 · 걸었다`
`derivation`	Inflection plus productive derived lemmas	`검증 → 검증하다 · 검증했다`

--literal is shorthand for --expand literal --pos literal. Derivation covers the productive suffixes -적, -하다, -되다, -시키다, -스럽다, -답다, -롭다, and -화. Predicate results are inflected in turn.

Boundary policy

Value	Decision	Use
`smart`	Locally verifies particle or ending consumption and fine-POS components	Precision-oriented default
`token`	Requires Unicode boundaries around the completed token	Independent tokens
`any`	Preserves morphological substring spans without side boundaries	Recall-oriented automation

For n:요리 in 중국요리, smart mode can admit 요리 as a verified noun component while rejecting 국요, which crosses component boundaries.

Part-of-speech system

--pos auto collects analyses from the core lexicon, enriched predicate metadata, user lexicon, productive suffix patterns, and full-POS lexicon. Explicit coarse POS values are:

noun (n:)          pronoun (pro:)     numeral (num:)
verb (v:)          adjective (adj:)   determiner (det:)
adverb (adv:)      particle (j:)      interjection (intj:)
literal (lit:)

Verbs and adjectives use different final, connective, and adnominal ending sets. Particles select allomorphs by the final sound of the preceding nominal.

Normalization, phrase distance, and alternatives

--unicode-normalization nfc compares normalized query and candidate forms. none preserves the input representation. Returned spans always address the original text.

Phrase atoms are matched independently and joined in source order. --max-gap limits the Unicode scalars between adjacent token spans and defaults to 24. A single-atom query can use global --pos; multi-atom queries require per-atom POS tags.

An unquoted, unescaped | finds any one of multiple atoms. Whitespace around it is optional, but CLI queries should be quoted so the shell does not interpret the operator as a pipe. Each alternative is one atom and cannot be mixed with phrase atoms. Write a literal pipe as \| or "|". --max-gap does not apply to a disjunction.

Input and output

Filesystem traversal respects ignore rules. Use --hidden, --no-ignore, --glob, and --type to select files, and --encoding for non-UTF-8 input.

kfind --glob '*.md' --hidden 검증 .
kfind --encoding euc-kr 걷다 legacy.txt
kfind --json --sort path 걷다 src
kfind --explain-query --pos verb 걷다

Default output is a bounded stream. --sort path buffers all file results for deterministic path ordering. --json emits JSON Lines with source spans and provenance. --explain-query prints the compiled plan without scanning a corpus.