Why Bareun adopted the 47 POS of the National Institute of Korean Language
When attaching POS to morphemes, Bareun uses the 47-tag POS tag set based on the National Institute of Korean Language (NIKL).
This is a system aligned with the Korean morpheme tag set standard of the Telecommunications Technology Association (TTA).
We adopted this standard instead of creating our own tags for consistency and compatibility.
The 47 POS at a glance
The 47 POS are organized by broad group as follows.
| Broad group | POS tags |
|---|---|
| Substantives | NNG common noun, NNP proper noun, NNB dependent noun, NP pronoun, NR numeral |
| Predicates | VV verb, VA adjective, VX auxiliary predicate, VCP positive copula, VCN negative copula |
| Modifiers | MMA descriptive determiner, MMD demonstrative determiner, MMN numeral determiner, MAG general adverb, MAJ conjunctive adverb |
| Independent words | IC interjection |
| Particles (josa) | JKS subject, JKC complement, JKG adnominal, JKO object, JKB adverbial, JKV vocative, JKQ quotative, JX auxiliary, JC conjunctive |
| Endings (eomi) | EP pre-final, EF final, EC connective, ETN nominalizing, ETM adnominalizing |
| Affixes and roots | XPN substantive prefix, XSN noun-derivational suffix, XSV verb-derivational suffix, XSA adjective-derivational suffix, XR root |
| Symbols | SF period, question mark, exclamation mark; SP comma, middle dot, colon, slash; SS quotation marks, brackets, dash; SE ellipsis; SO hyphen (attachment mark); SW other symbols; SL foreign word; SH Chinese character; SN number |
| Estimated / unanalyzable | NF estimated noun, NV estimated predicate, NA unanalyzable |
UNK and unanalyzable
Besides the table above, there is UNK, which means unclassified. Also, when the model estimates from the form that something is a noun or predicate but finds it hard to confirm, it assigns NF (estimated noun) or NV (estimated predicate), and when analysis is ultimately difficult it assigns NA (unanalyzable). These tags act as a safety net when new words or typos are encountered.
What following a standard means
graph LR STD[NIKL / TTA standard tag set] --> BAREUN[Bareun 47 POS]; BAREUN --> CORPUS[Compatible with training corpus tagging]; BAREUN --> TOOL[Comparable with other analyzers]; BAREUN --> USER[Lower learning cost for researchers and developers];
Why a standard tag set
- Consistency: Large corpora built with the standard tag set, such as the Sejong Corpus, can be used for training as they are.
- Compatibility: The tags are the same as in other morphological analyzers and research results, so results can be compared and transferred directly.
- Learning cost: It is a system already widely known in academia and industry, so there are no new tags to learn.
Creating your own tag set may feel easier at first, but you would have to re-tag existing corpora and could not check results against other tools. By following the standard, Bareun removed this cost from the start.
Analysis example
나는 학교에 간다 → 나/NP + 는/JX + 학교/NNG + 에/JKB + 가/VV + ㄴ다/EF
(Roughly: "I go to school".) The pronoun (NP), auxiliary particle (JX), common noun (NNG), adverbial particle (JKB), verb (VV), and final ending (EF) are all assigned as standard tags, unchanged.
Frequently asked questions
Q. How many tags are in Bareun's POS tag set?
It uses the 47 POS based on the National Institute of Korean Language. In addition, auxiliary tags such as UNK (unclassified) and estimated noun (NF), estimated predicate (NV), and unanalyzable (NA) are used.
Q. Why didn't you create your own tag set?
For consistency and compatibility. With a standard tag set, existing corpora such as the Sejong Corpus can be used for training as they are, and the tags are the same as in other analyzers and research results, so results can be compared directly.
Q. What tag is attached to a word that cannot be analyzed?
If it is estimated to be a noun, NF is attached; if it is estimated to be a predicate, NV; and if it is ultimately hard to decide, NA (unanalyzable). These tags are a safety net when new words or typos are encountered.