define

define — Contextual Definition & Translation

A word's true meaning lives in its context. "Bank" by a river is not "bank" that holds money; Thai "เอา" shifts sense with the verb beside it; an idiom means nothing if you read it word by word. This skill takes a target (a word or phrase) plus the context it sits in, resolves the sense the context selects, and explains it the way a good language teacher would — then, on request, translates just that target contextually into another language.

This is a comprehension tool, not a localization tool. For editing translation files, adding locales, or syncing translation keys, use the sibling i18n skill. The two share a cultural knowledge base (see Locale & cultural nuance) but do different jobs: i18n produces localized strings; define explains a word in context.

---

Input contract

Field	Required	Default
target	yes	the word or phrase to define/translate
context	yes (for a true contextual answer)	the surrounding sentence or paragraph
source language	no	auto-detect from the context
explanation language	no	the language the user is writing in
target language (for translation)	no	only if the user asks to translate
learner level	no	inferred; difficulty reported as a CEFR band

If the context is missing or too thin to disambiguate, do not silently pick one sense. State the most likely sense, name the alternative(s), and ask for more context only when it would materially change the answer.

---

Core principle

Resolve the sense the context selects — not the dictionary-default sense, and not a word-for-word translation. The disambiguating cue is part of the answer: a learner learns more from why a word means what it does here than from the gloss alone.

---

Workflow

1. Identify the target + context (+ source/explanation/target languages, + level if given).
2. Disambiguate the sense using the context:
   • Polysemy — when the target has several dictionary senses, pick the one the context supports.
   • Idioms / collocations / phrasal verbs / fixed expressions — detect multiword units and read them as a unit; warn against the literal reading.
   • Figurative vs literal — metaphor, hyperbole, irony/sarcasm (literal meaning ≠ intended meaning).
   • Domain / jargon — the legal, medical, technical, or slang sense when the context is in that domain.
   • Named entities — identify a proper noun as a referent; do not force a common-word definition onto it.
   • Morphology — lemmatize inflected forms; note what the inflection contributes grammatically.
3. Produce the learner gloss (include only the fields that earn their place):
   • Contextual meaning — plain-language definition of the sense as used here. Lead with this.
   • Why this sense — the specific cue in the surrounding text that selects it.
   • Part of speech / grammatical role in this sentence.
   • Register & connotation — formal/informal, polite/blunt, positive/negative, archaic/slang, regional.
   • Etymology / roots — only when it illuminates the meaning or aids memory (morphemes, origin).
   • Related senses & near-synonyms — senses it does not mean here; near-synonyms and how they differ.
   • Difficulty — CEFR band (A1–C2); "n/a" for proper nouns.
   • Contextual translation (only if a target language is requested) — the sense-appropriate rendering, not literal; flag when no clean equivalent exists.
   • Example — one fresh sentence reusing the same sense.
4. Locale & cultural nuance — see below.

---

Locale & cultural engine (shared with i18n)

define and i18n run the same cultural engine — the per-locale teams and facts in ../i18n/references/locale-knowledge.md. The only difference is the surface: i18n works on translation files in a coding project; define answers inline, in chat. So when a target is culturally loaded, dialectal, honorific, classical, or classifier-bearing, load that reference and use the target locale's team — the @native market copywriter, @cultural consultant, dialect / sociolinguistic / historical advisors — exactly as i18n would. They are what make a contextual translation read natively and a gloss culturally true. Don't restate or fork it.

Examples where it matters:

• A Japanese honorific verb (召し上がる vs 食べる) → the register difference is the answer.
• A Thai classifier in the context → name it; explain why the count needs it.
• A th-bupphe word (เพคะ, ออเจ้า) → flag the classical/Ayutthaya register, not a modern gloss.

---

Review lenses

Two layers, reconciled toward the sense the context actually supports:

1. The shared locale team (cultural engine). For the target locale, simulate its team from locale-knowledge.md — the same personas i18n uses (@native market copywriter, @cultural consultant, and any dialect / sociolinguistic / historical advisor the locale needs). This guarantees cultural authenticity and a native-sounding contextual translation.

2. Comprehension lenses (define's layer on top). What the translation team doesn't specialize in — the learner-facing gloss:

• @lexicographer — sense inventory; picks the right sense; distinguishes near-synonyms.
• @etymologist — roots and morphology that make the meaning stick.
• @register & pragmatics analyst — formality, tone, connotation, social appropriateness.
• @idiom & collocation detector — flags multiword expressions; prevents literal misreading.

---

Output format

Default to a concise structured gloss: lead with the one-line contextual meaning, then the relevant fields trimmed to what matters for this target. Expand any field on request. Don't pad — a clear A1 word needs three lines; a polysemous idiom in a sarcastic sentence needs the full treatment.

---

Edge cases & heuristics

• Thin context → give the most likely sense, name the alternative(s), and ask for more only if it changes the answer.
• Genuinely two senses → present both, ranked, each with the cue that would select it.
• Sarcasm / irony → state that the literal meaning differs from the intended one.
• Slang / neologism → flag volatility with [shelf-life: short] (same convention as i18n's Gen Z variants).
• Code-switching / loanwords → identify the source language, then gloss in context.
• Multi-word target → treat a fixed expression as a unit; otherwise gloss head + modifiers.
• Proper nouns / named entities → identify the referent; don't define them as common words.

---

What define does NOT do

• Editing, creating, or syncing translation files; adding locales; translation keys → use i18n.
• Full-document or article-length translation → out of scope.
• Context-free dictionary dumps → context is the whole point; ask for it.

---

Cross-skill integration

Skill	Relationship
i18n	Shares references/locale-knowledge.md. i18n produces localized strings and translation files; define explains a word/phrase in context. Disjoint triggers — file/locale work → i18n, "what does this mean here" → define.
team-composer	For multi-perspective discussion when a definition or translation choice is contested or strategically loaded, rather than a single contextual gloss.