What is Korean morphological analysis, and why do you need it?

Korean morphological analysis is a natural language processing technique that splits a sentence into morphemes, the smallest units that carry meaning, and attaches a part of speech (POS) to each one. For example, the eojeol (a space-delimited word unit) "바른을 사용해서" ("because I used Bareun") is analyzed as 바른/NNP + 을/JKO + 사용하/VV + 아서/EC. Bareun (bareun) is a Korean morphological analyzer that performs this task with 99.6% POS tagging accuracy.

What exactly is a morpheme?

A morpheme is "the smallest piece of language that carries meaning." A single eojeol (a unit separated by spaces) contains several morphemes. "먹었다" ("ate") consists of three morphemes: 먹/VV + 었/EP + 다/EF. Bareun classifies morphemes with the 47-tag POS system based on the National Institute of Korean Language (NIKL) standard, and groups them into 8 major morphological units: substantives (N), particles (J) (called josa in Korean), predicates (V), endings (E) (called eomi), adverbials (A), determiners (M), interjections (I), and symbols (S). To learn about the POS system, see Why 47 POS tags and The 8 major morphological units. For the tag list, see Tag information.

Why is morphological analysis especially hard for Korean?

Because Korean is an agglutinative language. In an agglutinative language, grammatical elements such as particles and endings attach one after another to a root and form a single eojeol. You cannot split words by spaces alone, as you can in English, and conjugation, contraction, and omission occur constantly.

Why agglutination is tricky

"갔어요" ("went") is analyzed as 가/VV + 았/EP + 어요/EF, but on the surface the vowel of 가 is contracted and does not show. Bareun restores such conjugation, contraction, and omission, and it also tracks precisely which position in the original syllables each restored morpheme came from.

Ambiguity is another major obstacle: the same surface form can be interpreted in several ways. Depending on the context, "가는" can be analyzed as a form of 가늘다 (to be thin), 가다 (to go), or 갈다 (to grind or to replace). Bareun found 12,175 words that share the same surface form but have different analyses, and it tells them apart with word embeddings that include morphological information (가#J, 가-#V, and 가#N are embedded as different tokens). We have published this ambiguity data. We validate the ambiguity resolution performance on an evaluation set of 35,396 sentences and 8,285 surface forms. The composition of the ambiguity data is described in detail in Ambiguity dataset.

Where is morphological analysis used?

Morphological analysis is the starting point of almost all natural language processing for Korean.

Search engines and indexing: If you index eojeol as they are, "사용해서" and "사용하고" are treated as different words. Extracting the root 사용하 with morphological analysis greatly improves search accuracy.
Keyword extraction and topic analysis: Pick out only the substantives (nouns) to extract core keywords. Bareun makes it easy to extract only nouns with nouns() and only predicates with verbs().
Sentiment analysis and text mining: Use the POS information of predicates and adverbs to classify emotions and attitudes precisely.
Speech recognition and spoken-language processing: Bareun correctly analyzes spoken-language endings such as 감사합니당 and 알겠어용.
LLM preprocessing and RAG: Use it as a preprocessing step to improve retrieval accuracy and token efficiency.

Is Bareun's morphological analysis free?

Yes. Bareun's morphological analysis is free for individuals, companies, schools, and news organizations when used for research purposes. On the cloud (api.bareun.ai), you can also try it free for up to 50,000 eojeol. To apply it to a commercial service, you need a paid license.

Frequently asked questions

Q. What is the difference between morphological analysis and a morphological analyzer?

Morphological analysis refers to the technique or task, and a morphological analyzer is the program that performs it. Bareun is both a Korean morphological analyzer engine and a Korean spell checker.

Q. Is morphological analysis the same as tokenization?

They are similar but different. Simple tokenization splits text only into characters or eojeol, whereas morphological analysis also tells you the POS and the source of each piece (dictionary entry or out-of-vocabulary word). Bareun also provides segmentation (word splitting) results separately, which split text into the 8 major morphological units without POS tagging.

Q. Which POS system does it use?

It uses the 47-tag POS system based on the NIKL standard. It marks nouns, verbs, adjectives, particles, endings, and so on as detailed tags. In the segmentation stage before that, it splits eojeol into the 8 major morphological units using 115 segmentation rules (14 of them for error detection) discovered by analyzing the conjugation characteristics of Korean. To try calling the analysis yourself, see Morphological analysis with the API.

Q. How does it tell apart the ambiguity of the same surface form?

It distinguishes 12,175 words that share the same surface form but have different analyses by using word embeddings that include morphological information. For example, it embeds the particle 가#J, the predicate 가-#V, and the noun 가#N as different tokens.