Sentence-Level and Public Language Correction

Starting with version 3.4.0, the Bareun spell checker offers corrections that look at the whole sentence rather than a single word, along with corrections based on the public language guidelines of the National Institute of Korean Language. It is especially useful for polishing text written by public institutions, such as press releases, official documents, and notices.

The new features fall into two kinds.

• Always-on corrections: Expressions that clearly violate the norms. They are corrected without any configuration.
• Opt-in corrections: Expressions that the guidelines recommend changing but that are also common in normal writing, or whose correction changes the sentence structure. They run only in calls where the request option is turned on.

Always-On Corrections

Honorifics on Inanimate Subjects

The honorific '-시-' (-si-) honors a person. Attaching '-시-' when the subject is an object or an event violates the norms. Bareun removes '-시-' from the predicate in sentences whose subject is not a person.

(The first sentence means "The coffee you ordered is ready", the second "That item is sold out", and the third "The president has something to say".)

As the last example shows, for indirect honorifics where the subject is the words or thoughts of the person being honored, '계시다' is corrected to '있으시다'. Sentences with a person as the subject, such as '선생님께서 오셨습니다' ("The teacher has arrived"), are left as they are.

Incorrect Expressions in the Public Language Guidelines

(The first sentence means "Please keep this point in mind", the second "Our country's exports have grown".)

Opt-In Corrections

There are four options. The first three must be turned on to work. The last one turns off a feature that is always on.

Sentence-Level Check enable_sentence_check

Subject-predicate agreement: When the object particle '을/를' comes before a '-되다' or passive-verb predicate, it is corrected to the subject position.

(Meaning: "300 members of the National Assembly were elected" and "Illegal acts must be eradicated".)

Passive sentences that rightly take an object, such as '개가 내 손가락을 물렸다' ("My finger got bitten by a dog"), are not corrected. Whether a noun is a subject is judged from context. In installations that do not use the context judgment feature, the body text is not changed and the fix is offered only as a confirmation block.

Translationese passive '~에 의해 ~되다': For sentences with a clear agent, it suggests the active-voice form as a confirmation block. The body text is not changed.

(Meaning: "Joseon was founded by King Taejo, Yi Seong-gye" and "King Taejo, Yi Seong-gye, founded Joseon".)

If '~에 의해' marks a cause or basis rather than an agent (세금은 법에 의해 정해진다, "Taxes are set by law"), no suggestion is made. The context judgment feature decides whether it is an agent, so installations that do not use that feature do not produce this suggestion.

Public Term Simplification Suggestions enable_public_term_refine

For nearly 170 words, such as administrative terms and Japanese-style terms refined by the National Institute of Korean Language, it suggests plainer words as a confirmation block. Because the original is not wrong, the body text is not changed.

(Meaning: "We will let you know the schedule of the year-end party", "Submit by the next day", "An event will be held at the riverside park", and "We introduced AI technology".)

For words with several meanings, it suggests the one that fits the context. For 'AI', if the same sentence contains words such as '조류' (bird), '고병원성' (highly pathogenic), or '가금' (poultry), it suggests '조류 인플루엔자' (avian influenza). Otherwise it suggests '인공 지능' (artificial intelligence).

Contested-Usage Fixes enable_contested_fix

These are expressions the guidelines recommend changing but that are so widely settled that they appear in normal writing. Most are suggested as confirmation blocks, and only the overuse of the plural '-들' is corrected in the body text.

The plural '-들' is removed only right after an expression that already states a quantity, such as '많은', '여러', or '세 명의'. When there is no quantity expression, as in '학생들이 모였다', it is left as it is.

Turn Off Hate and Discrimination Expression Check disable_bias_check

Bareun suggests neutral expressions, as confirmation blocks, for words (about 1,400) that Urimalsaem defines as '낮잡아 이르는 말' (derogatory terms). This feature is always on. Pass this option as true only when you want to turn it off.

In writing that uses derogatory words on purpose, such as dialogue in a novel, you can use this option to turn the suggestions off.

What Is a Confirmation Block

A confirmation block is a correction result that shows candidates for a fix without changing the body text. In the response it is a correction block whose category is CONFIRM. The first item in the candidate list is the original, and suggestions start from the second item. It is the writer who should choose, based on context, so Bareun does not decide.

graph LR
  A[Original: 망년회 일정] --> B{Confirmation block};
  B -- Keep original --> C[망년회 일정];
  B -- Pick suggestion --> D[송년회 일정];

Sentence-level suggestions come as a block that covers one whole sentence as its original range, plus small blocks attached to each changed spot inside it. This lets Bareun show each change one by one, such as '이성계에 의해 → 이성계가' and '건국되었다 → 건국했다'.

How to Turn On the Options

In the Bareun web UI, turn them on with the switches in the correction settings panel (sentence-level check, public term simplification suggestions, contested-usage fixes, and turning off the hate and discrimination expression check). When calling the API, put the options in the config of the spell checking API. Turn on only the ones you need.

{
  "document": { "content": "국회의원 300명을 선출되었다. 망년회 일정을 알려 드립니다.", "language": "ko-KR" },
  "encoding_type": "UTF8",
  "config": {
    "enable_sentence_check": true,
    "enable_public_term_refine": true,
    "enable_contested_fix": true
  }
}

curl \
  -H "api-key: koba-ABCDEFG-1234567-LMNOPQR-7654321" \
  -H "Content-Type: application/json" \
  -X POST https://api.bareun.ai/bareun.RevisionService/CorrectError \
  -d '{"document": {"content": "국회의원 300명을 선출되었다.", "language": "ko-KR"}, "encoding_type": "UTF8", "config": {"enable_sentence_check": true}}'

If you pass no options, you get the same corrections as in 3.3 and earlier, plus the corrections for honorifics on inanimate subjects and incorrect public-language expressions. The meaning of the remaining options is summarized in CorrectError API options.

Frequently Asked Questions

Q. What changes if I do not turn on the options?

Only honorifics on inanimate subjects ('커피 나오셨습니다 → 나왔습니다') and incorrect expressions such as '염두하다' and '저희 나라' are always corrected. The sentence-level check, public term simplification, and contested-usage fixes work only when their options are on.

Q. Does a confirmation block change the body text?

No. The corrected text in the response stays identical to the original, and the correction block carries both the original and the suggestions. Let the user choose in your UI.

Q. Can I use it for text that is not from a public institution?

Yes. Note, however, that 'contested-usage fixes' also suggests expressions common in everyday writing ('좋은 하루 되세요'), so we recommend turning it on for text that must follow the guidelines, such as official documents and press releases.

Q. How do I turn off the derogatory-word suggestions?

Put "disable_bias_check": true in config. Other corrections are not affected.