POS Classification vs MeCab
MeCab and Suzume use different POS classification strategies, so the same token can carry a different label in each. Suzume applies its own feature and context rules to assign the public POS taxonomy (NOUN, VERB, ADJ, …). MeCab's granularity mirrors its dictionary, which is coarse for modern and colloquial usage — slang, newer pronouns, and na-adjective stems often land in plain 名詞 — while Suzume folds those into its compact tag set by context.
Looking for token boundary differences?
This page covers how tokens are labeled. For where token boundaries differ — merging, splitting, and normalization — see Differences from MeCab.
Reading the comparisons
In each comparison the MeCab row uses MeCab's Japanese POS names, and the Suzume row uses the public API tags (NOUN, VERB, ADJ, …) with the Japanese POS name shown next to each code. The MeCab output was collected with MeCab 0.996 and mecab-ipadic 2.7.0-20070801, without a user dictionary. See the comparison baseline for details; another dictionary can change both boundaries and POS labels.
Adjective-Derived よく
MeCab lexicalizes よく as an adverb. Suzume assigns ADJ as the continuative form of よい:
Ordinary adjective continuative forms such as 美しく are omitted because current MeCab also classifies them as adjectives.
Pronoun Recognition
MeCab classifies many pronouns as plain nouns. Suzume assigns PRON:
Applies to: あなた, あんた, みんな, みな, 皆, 某, 拙者, 我輩, 彼女, 彼氏, 奴, 我, わし, いくら (interrogative)
Suzume also assigns PRON to the colloquial pronouns どいつ, こいつ, そいつ, and あいつ. If a MeCab dictionary variant splits the surface, the normalization pipeline restores one token.
Na-Adjective Recognition
MeCab classifies na-adjective stems as nouns (形容動詞語幹). Suzume recognizes them as adjectives:
Examples include きれい, しずか, おだやか, げんき, しんちょう, ありきたり, 無限, and 滅多. Ambiguous nominal readings are resolved from context.
Note: Some 形容動詞語幹 intentionally remain NOUN: マジ, 不安, 不要, 乙, 不便, 公式, 可能, 容易, 積極, 健康, 傍若無人
て-Form Auxiliaries
After the connective て/で, subsidiary verbs lose their independent meaning and receive AUX. Suzume applies this consistently across the whole construction family, where MeCab labels them 動詞 (usually with the 非自立 subcategory).
Progressive いる and trial みる — the independent verbs いる and 見る remain VERB elsewhere:
Giving and receiving auxiliaries くれる, あげる, もらう:
Completive しまう and its contraction ちゃう:
Contracted progressive てる/でる — the lemma stays いる:
Humble いただく in polite requests (the negative ご確認いただけません follows the same pattern):
Resultative てある is the exception — ある keeps its existential meaning and remains VERB:
Purpose Expressions and Subsidiary ゆく
A hiragana verb stem before に plus a motion verb is classified as a nominal search unit expressing purpose. In literary "verb stem + ゆく/いく" expressions, the second element is treated as a verb.
Restoring Hiragana Verb Inflections
When MeCab analyzes a short pure-hiragana nasal sound change as a noun, Suzume restores the verb and its lemma from the following だ/で and the inflection pattern. A Godan-wa verb whose final う was split as an auxiliary is also restored to one token.
Verb Stems in Honorific Requests
In "お/ご + verb stem + くださる/いたす/いただく" constructions, a stem that is homographic with a noun is restored to a verb from the grammatical structure.
Suffix Recognition
Productive derivational suffixes receive the dedicated SUFFIX tag, and the stem keeps its own class and lemma.
Nominalizer さ turns an adjective into a noun; Suzume keeps the adjective stem visible:
X的 + な is treated as one na-adjective:
Suffixes that change token boundaries — 中, 抜き, 建て, and friends — are covered in Differences from MeCab instead.
Deverbal Nouns
A verb continuative used as a noun keeps NOUN, even where a MeCab dictionary labels it a verb form:
Other deverbal forms are context-dependent; see 違い and 推し in the per-word table below.
Context-Dependent POS
Suzume applies context-aware POS classification for several ambiguous words:
そう: ADJ before a copula (そうだ = hearsay), AUX after an auxiliary (しまいそう = appearance), and ADV otherwise.
でも: PARTICLE after a host, but CONJ at the start of a clause:
いかが: ADJ before a copula (いかがですか) and ADV otherwise.
大変: ADJ before the attributive copula な and before だ/です; it is ADV in adverbial use (大変良い) and in isolation.
どう: ADJ before a copula (どうだ) and ADV otherwise, including どうする and か…どうか:
よう: the volitional う after a 未然形 is AUX, while the formal noun よう in ように/ような is NOUN:
なら: the conditional stays a particle, but before a negative (ならない) and standalone it parses as the verb なる.
Particle Classification
MeCab classifies certain particles as nouns in some contexts. Suzume applies context-aware classification for 30+ particles:
The nominalizer の functions as a particle here, not a noun. Suzume classifies such cases as particles.
Colloquial Copulas and Particles
Colloquial copulas and particles that dictionary taxonomies handle unevenly are normalized to consistent tags:
| Example | Word | Suzume | Role |
|---|---|---|---|
| いいっすね | っす | AUX (lemma: です) | Casual copula |
| いいじゃん | じゃん | PARTICLE | Colloquial sentence-final particle |
| 夢みたい | みたい | AUX | Similative auxiliary |
| 私なんか | なんか | PARTICLE | Deprecatory/exemplifying particle |
| 英語はおろか | おろか | PARTICLE | "Let alone" particle |
| そうや | や | PARTICLE | Kansai sentence-final particle |
Suzume recognizes some regional predicate tails and particles in context. Examples include あかん and へん as AUX, ねん and さかい as PARTICLE, and the polite copula どす as AUX. These are grammatical labels, not a claim that every dialectal expression is covered.
Katakana Onomatopoeia
MeCab classifies katakana onomatopoeia (reduplication patterns) as nouns. Suzume recognizes them as adverbs:
Mimetics ending in っと are also kept whole as adverbs:
で+ある Copula Handling
Suzume applies context-aware classification for the copula である pattern:
ない Context-Dependent Classification
Suzume assigns ADJ rather than AUX to ない/なく/なかっ when they function as an existence adjective:
existence negation
negation auxiliary
lexical adjective
The continuative なく follows the same distinction. It is AUX after a verb or passive/causative stem, but ADJ after an adjective continuative or in an independent absence expression:
Per-Word POS Differences
The following words are classified differently between MeCab and Suzume:
| Word | MeCab | Suzume | Reason |
|---|---|---|---|
| なら | 接続詞 | PARTICLE / VERB (文脈依存) | Conditional particle by default; parsed as the verb なる before negatives or standalone |
| 違い | 名詞・ナイ形容詞語幹 | VERB / NOUN (文脈依存) | Standalone is VERB (lemma: 違う); 違いない, 違いがある, and modified nominal uses are NOUN |
| 推し | 動詞 | NOUN (名詞) | Modern noun usage |
| 嫌い | 動詞 | ADJ (形容詞) | Na-adjective |
| 大変 | 名詞・形容動詞語幹 | ADJ / ADV (文脈依存) | ADJ before な/だ/です; ADV in adverbial or isolated use |
| 超 | 接頭詞 | NOUN (名詞) | Modern usage |
| びっくり | 名詞・サ変接続 | ADV (副詞) | Adverb usage |
| なるほど | 感動詞 | ADV (副詞) | Adverb usage |
| たくさん | 名詞・副詞可能 | ADV (副詞) | Adverb usage |
| いずれ | 名詞・代名詞 | ADV (副詞) | Adverb usage |
| おめでとう | 感動詞 | ADV (副詞) | Adverb usage |
| じゃん | 接続詞 + 終助詞 | PARTICLE (助詞) | Colloquial sentence-final particle |
| よう | 感動詞 | AUX / NOUN (文脈依存) | Volitional う after 未然形 (見よ + う); formal noun in ように / ような |
| 時々 | 副詞 | NOUN (名詞) | Noun usage |
| 遥か | 副詞 | ADJ (形容詞) | Na-adjective |
| どう | 副詞 | ADJ / ADV (文脈依存) | ADJ before a copula; ADV otherwise |
| まじ | 助動詞 | ADJ (形容詞) | Adjective (katakana マジ stays NOUN) |
| っていう | 助詞 | DET (連体詞) | Determiner |
| という | 助詞 | DET (連体詞) | Determiner |
| といった | 助詞 | DET (連体詞) | Quotative determiner (DET_引用) |
| まして | 副詞 | CONJ (接続詞) | Conjunction |
| いわば | 副詞 | CONJ (接続詞) | Conjunction (言わば) |
| その後 | 名詞・副詞可能 | ADV (副詞) | Adverb usage |
| しどろもどろ | 名詞・形容動詞語幹 | ADV (副詞) | Adverb usage |
寒し is not a broad-POS difference: both analyzers label it as an adjective, and Suzume returns the lemma 寒い.
POS Granularity
Suzume's basic POS (pos) uses a simpler tag set than MeCab's detailed subcategories.
| MeCab | Suzume pos |
|---|---|
| 名詞,一般 | NOUN |
| 名詞,固有名詞,地域 | NOUN |
| 名詞,サ変接続 | NOUN |
| 名詞,副詞可能 | NOUN |
| 動詞,自立 | VERB |
| 動詞,非自立 | AUX (default) |
The 動詞,非自立 subcategory maps to AUX by default. A small set of subsidiary verbs (すぎる, くださる, あげる, くれる, もらう, いく, いる, …) is not blanket-mapped: they stay VERB in independent use, and context rules such as the て-form classification above assign AUX where they act as auxiliaries.
The extendedPos field provides Suzume's own finer-grained taxonomy. The following rows show rough MeCab feature analogues, but a MeCab label does not by itself cause the corresponding Suzume value; dictionary flags, structure, and context determine it.
| Rough MeCab feature analogue | Suzume extendedPos |
|---|---|
| 名詞,固有名詞 | NOUN_固有 |
| 名詞,固有名詞,人名 | NOUN_名 / NOUN_姓 |
| 名詞,数 | NOUN_数 |
| 名詞,サ変接続 | NOUN_転成 |
| 名詞,形式名詞 | NOUN_形式 |
| 動詞,連用形 | VERB_連用 |
| 動詞,未然形 | VERB_未然 |
| 動詞, 活用形=仮定縮約1 | VERB_仮定縮約 |
| 形容詞,連用形 | ADJ_連用 |
| 形容動詞語幹 | ADJ_NA |
| 助動詞, 活用型=文語・キ | AUX_文語過去キ |
| 助詞,格助詞 | PART_格 |
| 助詞,係助詞 | PART_係 |
extendedPos is not a drop-in replacement for MeCab dictionary features. For example, a proper noun or number can remain plain NOUN when Suzume has no dictionary or structural evidence for a narrower class. See the API Reference ExtendedPOS section for the full list.
Contracted Hypotheticals
Suzume recognizes colloquial contracted hypothetical forms as one inflected predicate while preserving the lemma:
Verb forms report VERB_仮定縮約 in extendedPos; adjective forms keep their adjective inflection category.