REFERENCE · GLOSSARY
Search, grammar, and measurement terms
Definitions of the search structures, Korean grammar, quality metrics, and performance units used throughout the kfind documentation.
Search input
- Lemmalemma
- The dictionary form without inflectional endings or particles. kfind compiles a lemma into searchable surface forms.
Example 걷다 has surfaces such as 걷고, 걸어, and 걸었다. - Part of speechPOS · part of speech
- A grammatical class, such as noun, verb, or adjective, that determines a word’s role and combination rules.
Example 먹다 is a verb, 맛있다 is an adjective, and 음식 is a noun. - Queryquery
- A search expression containing lemmas, optional POS tags, phrase structure, and options.
Example n:사용자 v:검증하다는 명사와 동사를 순서대로 찾는 검색 질의입니다. - Atomatom
- The smallest query unit: one lemma and an optional POS tag.
Example n:사용자와 v:검증하다는 각각 하나의 atom입니다. - Phrasephrase
- A query that joins two or more atoms under ordering and maximum-gap constraints.
Example n:사용자 v:검증하다는 사용자와 검증하다의 활용형을 정해진 순서로 찾습니다. - Inflectioninflection
- The formation of surface forms through particles, endings, or stem alternations without changing the part of speech.
Example 걷다 → 걷고, 걸어, 걸었다 - Derivationderivation
- The formation of a new lemma or part of speech by attaching a derivational affix.
Example 검증 → 검증하다, 문화 → 문화적
Korean grammar
- ParticleJ* · particle
- A grammatical element attached to a nominal to mark relations or meaning, including case particles and auxiliary particles.
Example 사람이의 이/JKS, 책을의 을/JKO, 나도에서 도/JX - EndingE* · ending
- A grammatical element attached to a predicate stem to express tense, modality, clause linkage, or sentence ending.
Example 먹었다 = 먹/VV + 었/EP + 다/EF - Prefinal endingEP · prefinal ending
- An ending between the stem and a final, connective, or transformative ending. It adds honorific, tense, or modal meaning, and multiple prefinal endings may be chained.
Example 먹었겠지만 = 먹/VV + 었/EP + 겠/EP + 지만/EC - Final endingEF · final ending
- An ending that closes a sentence and marks a statement, question, command, or another sentence type.
Example 먹습니다의 습니다/EF, 먹니의 니/EF - Connective endingEC · connective ending
- An ending that links a predicate to another clause and marks relations such as sequence, cause, contrast, or condition.
Example 먹고의 고/EC, 먹지만의 지만/EC, 먹으면의 으면/EC - Adnominal endingETM · adnominal ending
- An ending that turns a predicate into a modifier of a following nominal.
Example 먹는 사람의 는/ETM, 먹을 음식의 을/ETM - Nominal endingETN · nominal ending
- An ending that lets a predicate function as a nominal in a sentence.
Example 먹기의 기/ETN, 믿음의 음/ETN - Allomorphallomorph
- One of multiple forms with the same grammatical function, selected by the phonological context.
Example 받침 뒤의 은·이·을과 모음 뒤의 는·가·를 - Irregular conjugationirregular conjugation
- An inflection in which the stem or ending changes outside the regular combination pattern.
Example 걷다 → 걸어(ㄷ 불규칙), 돕다 → 도와(ㅂ 불규칙)
Morpheme labels
- General nounNNG
- A nominal that names a general person, object, or concept.
Example 학교/NNG - Proper nounNNP
- A nominal naming a specific person, place, organization, or other unique entity.
Example 서울/NNP - Dependent nounNNB
- A noun that normally requires a preceding modifier.
Example 할 수의 수/NNB - Counter nounNNBC
- A dependent noun used as a counter or measurement unit after a quantity.
Example 세 명의 명/NNBC - PronounNP
- A nominal that stands in place of a person or thing.
Example 나/NP - NumeralNR
- A nominal expressing quantity or order.
Example 셋/NR - VerbVV
- A predicate expressing an action or change and inflecting with endings.
Example 먹/VV - AdjectiveVA
- A predicate expressing a state or property and inflecting with endings.
Example 맑/VA - Auxiliary predicateVX
- An auxiliary verb or adjective adding aspect, modality, or result meaning after another predicate.
Example 먹어 보았다의 보/VX - Positive copulaVCP
- A form of the positive copula 이다 that turns a nominal into a predicate.
Example 학생이었다의 이/VCP - Negative copulaVCN
- A form of the negative copula 아니다.
Example 학생이 아니다의 아니/VCN - Subject case particleJKS
- A case particle marking a nominal as the subject.
Example 사람이의 이/JKS - Complement case particleJKC
- A case particle marking a complement before predicates such as 되다 or 아니다.
Example 학생이 되다의 이/JKC - Adnominal case particleJKG
- A case particle marking an adnominal relation to a following nominal.
Example 우리의 집의 의/JKG - Object case particleJKO
- A case particle marking a nominal as the object.
Example 책을의 을/JKO - Adverbial case particleJKB
- A case particle marking an adverbial relation such as place, direction, means, or recipient.
Example 학교에서의 에서/JKB - Vocative case particleJKV
- A case particle marking the addressee of a call.
Example 철수야의 야/JKV - Quotative case particleJKQ
- A case particle marking quoted content.
Example 좋다고의 고/JKQ - Auxiliary particleJX
- A particle adding meanings such as topic, contrast, limitation, or addition.
Example 나도에서 도/JX - Conjunctive particleJC
- A particle coordinating two or more nominals.
Example 사과와 배의 와/JC - Prefinal endingEP
- An ending between a stem and final ending that adds honorific, tense, or modal meaning.
Example 먹었다의 었/EP - Final endingEF
- An ending that closes a sentence and marks its sentence type.
Example 먹었다의 다/EF - Connective endingEC
- An ending connecting a predicate to a following clause or predicate.
Example 먹고의 고/EC - Nominal endingETN
- An ending that lets a predicate function as a nominal.
Example 먹기의 기/ETN - Adnominal endingETM
- An ending that turns a predicate into a modifier of a following nominal.
Example 먹는 사람의 는/ETM - Nominal prefixXPN
- A prefix attached before a nominal or root to form a nominal.
Example 풋사과의 풋/XPN - Noun-forming suffixXSN
- A derivational suffix that forms a noun.
Example 문화적의 적/XSN - Verb-forming suffixXSV
- A derivational suffix that forms a verb.
Example 공부하다의 하/XSV - Adjective-forming suffixXSA
- A derivational suffix that forms an adjective.
Example 자연스럽다의 스럽/XSA - RootXR
- A lexical root that combines with derivational affixes to form a word.
Example 아름답다의 아름/XR
Compilation and execution
- Query planquery plan
- The compiled representation that combines candidate programs, anchors, and verification conditions.
- Candidate programCandidateProgram
- An execution unit preserving a fixed anchor, core projection, suffix-consumption state, decision constraints, and provenance.
- Anchoranchor
- A fixed byte string searched first to locate possible morphological matches in the corpus.
- Boundary policyboundary
- A condition deciding whether a candidate must be a token, a structurally verified component, or an unrestricted substring.
- Provenanceprovenance
- Explanation data identifying the POS analysis and inflectional or derivational rules that produced a match.
Data and resources
- Morphological lexiconlexicon
- A resource supplying lemma POS, conjugation class, and morphological information during query compilation.
- Component resourcecomponent resource
- A resource supplying fine POS and aligned morpheme spans for local structural verification of a candidate token.
Quality and performance
- True positiveTP · true positive
- A correct item returned by the search.
- True negativeTN · true negative
- An incorrect item excluded from the search results.
- False positiveFP · false positive
- An incorrect item returned by the search.
- False negativeFN · false negative
- A correct item omitted from the search results.
- PrecisionPrecision = TP / (TP + FP)
- The fraction of returned results that are correct. The benchmark records zero when the denominator is zero.
Example With TP 8 and FP 2, precision is 8 / 10 = 0.8. - RecallRecall = TP / (TP + FN)
- The fraction of expected results that were returned. The benchmark records zero when the denominator is zero.
Example With TP 8 and FN 2, recall is 8 / 10 = 0.8. - F1 scoreF1 · F1 score · 2PR / (P + R)
- The harmonic mean of precision and recall. It falls when either input falls, and is recorded as zero when their sum is zero.
Example 정밀도 0.8, 재현율 0.8이면 F1은 0.8입니다. - Raw metricraw metric
- A quality metric that counts every fixture expectation in TP, TN, FP, and FN, regardless of the product contract.
Example Raw FN 4 may include cases outside the product contract. - Contract-adjusted metriccontract-adjusted metric · TPᶜ/TNᶜ/FPᶜ/FNᶜ
- A quality metric that counts only human-reviewed in-contract cases in TPᶜ, TNᶜ, FPᶜ, and FNᶜ. A fixture without review uses the raw confusion matrix and reports reviewed as zero.
Example Raw FN 4 with FNᶜ 0 means four misses were observed, all outside the contract, and no in-contract item was missed. - Latencylatency · p50/p95
- Time to complete one operation. p50 is the median, while p95 is the value at or below which 95% of observations fall.
- Throughputthroughput · cases/s · MiB/s
- Work completed per unit time. Units remain workload-specific, such as cases/s for fixture evaluation and MiB/s for corpus scanning.