Cross Reference Entry

When NOT to use prominent_see_also:

• Regular synonyms (use cross_references with type synonym)
• Words that happen to sound similar but aren't confusable in practice
• Words with different POS that a learner wouldn't confuse
• Words in very different registers or domains that wouldn't be confused in practice

Format:

"prominent_see_also": [
  {
    "target_id": "00754_shimaru",
    "reading": "しまる",
    "headword": "{閉|し}まる",
    "note": "intransitive"
  }
]

• Always include target_id when the target entry exists
• Always include a brief note in English (2-4 words) explaining the relationship
• Add references bidirectionally — both entries should point to each other
• For N/Nする pairs, use notes like "verb form" / "noun form"
• For transitive/intransitive pairs, use notes like "transitive" / "intransitive"
• For homophones, use a brief gloss distinguishing the words

2. cross_references — "Related Words" box at the bottom (STRUCTURED)

A structured array displayed in a "Related Words" box at the bottom of the entry page. These express lexicographic relationships between entries. Two-way linking is encouraged but not as critical as for prominent_see_also.

Cross-Reference Types (for cross_references)

antonym — Opposites (HIGH PRIORITY)

Use for direct opposites.

{
  "type": "antonym",
  "reading": "あける",
  "headword": "{開|あ}ける",
  "label": "to open"
}

Label: Brief gloss of target word

keigo — Honorific/Humble Forms (HIGH PRIORITY)

Use for formal speech equivalents.

{
  "type": "keigo",
  "reading": "めしあがる",
  "headword": "{召|め}し{上|あ}がる",
  "label": "honorific"
}

Labels: honorific or humble

Common keigo links:

• 食べる → 召し上がる (hon.), いただく (hum.)
• 行く → いらっしゃる (hon.), 参る (hum.)
• 言う → おっしゃる (hon.), 申す (hum.)
• 見る → ご覧になる (hon.), 拝見する (hum.)

synonym — Similar Meaning (MEDIUM PRIORITY)

Use for words with similar meaning but different nuance.

{
  "type": "synonym",
  "reading": "りかいする",
  "headword": "{理解|りかい}する",
  "label": "formal"
}

Label: Distinguishing characteristic (e.g., "formal", "written", "casual")

contrast — Easily Confused (MEDIUM PRIORITY)

Use for words learners often confuse.

{
  "type": "contrast",
  "reading": "が",
  "headword": "が",
  "label": "subject marking"
}

Especially important for:

• Particles: は vs が, に vs で, に vs へ
• Similar verbs: 聞く vs 聴く, 見る vs 見える vs 見せる

homophone — Same Reading, Different Meaning (MEDIUM PRIORITY)

Use for words that share a reading. Note: if the homophones are easily confused, prefer prominent_see_also instead.

related — Semantically Connected (LOW PRIORITY)

Use for derived words, compounds, or thematically related vocabulary.

{
  "type": "related",
  "reading": "たべもの",
  "headword": "{食|た}べ{物|もの}",
  "label": "food (noun)"
}

see_also — General Reference (LOW PRIORITY)

Use for general cross-references that don't fit other categories.

{
  "type": "see_also",
  "reading": "しょくじ",
  "headword": "{食事|しょくじ}",
  "label": null
}

pair — DEPRECATED

Do not use. Transitive/intransitive verb pairs should use prominent_see_also instead. Existing pair-type entries in cross_references should be migrated to prominent_see_also when entries are revised. The type remains technically valid in the schema but should not appear in new entries.

Format Requirements

Each cross_references entry requires:

*Headword is required for proper resolution. Without it, cross-references cannot be disambiguated between homonyms.

Each prominent_see_also entry requires:

Note: Valid cross-reference types are defined centrally in build/constants.py and shared across the schema, validation, and build scripts.

Hybrid Cross-Reference System

The dictionary uses a hybrid system that supports both:

1. Hard-coded target_id — Direct reference to an entry ID (unambiguous)
2. Forward references — References by reading/headword to entries that may not exist yet

Resolution Priority

When resolving a cross-reference:

1. If target_id present AND entry exists → resolved (use ID directly)
2. If target_id present AND entry missing → ERROR (stale reference)
3. If no target_id → resolve by reading/headword (may be pending if target doesn't exist)

When to Use target_id

Use target_id when:

• The target entry exists in the dictionary
• You want guaranteed, unambiguous resolution

Don't manually add target_id when:

• Creating forward references to entries that don't exist yet
• You're unsure which homonym is correct

Instead, use the harden_references.py script to automatically add target_id to resolvable references.

Example with target_id

{
  "type": "antonym",
  "target_id": "00754_shimaru",
  "reading": "しまる",
  "headword": "{閉|し}まる",
  "label": "intransitive"
}

Example forward reference (no target_id)

{
  "type": "antonym",
  "reading": "ひらく",
  "headword": "{開|ひら}く",
  "label": "to open"
}

Homonym Disambiguation

CRITICAL: Many Japanese words share the same reading but have different kanji (homonyms). The headword field is essential for correct resolution.

Example: The reading かんじょう has multiple entries:

• {感情|かんじょう} — emotion, feeling
• {勘定|かんじょう} — bill, calculation

Always include the headword to ensure cross-references link to the correct entry.

// CORRECT — specifies headword for disambiguation
{
  "type": "synonym",
  "reading": "かんじょう",
  "headword": "{勘定|かんじょう}",
  "label": "bill, calculation"
}

// INCORRECT — no headword, may link to wrong homonym
{
  "type": "synonym",
  "reading": "かんじょう",
  "label": "bill, calculation"
}

When Creating New Entries

When creating new entries (e.g., via prompts/newentries.md), add cross-references as part of entry creation:

1. Always add prominent_see_also for:

   • Transitive/intransitive pair (if both entries exist)
   • N/Nする pair (if both entries exist)
   • Obvious homophones that learners would confuse

2. Add cross_references for:

   • Direct antonyms
   • Keigo equivalents
   • Close synonyms with clear distinctions
   • Other relevant relationships

3. Check if the target entry exists using check_duplicate.py or the entries index:

   • If yes: include target_id and add a back-link on the target entry
   • If no: create a forward reference (reading + headword only)

4. For back-links on existing entries: When you add a cross-reference pointing to an existing entry, also add a reciprocal reference on that target entry pointing back to the new entry.

Priority Order

When adding references to entries, prioritize:

1. HIGH — Always add if applicable:

   • Transitive/intransitive pairs → prominent_see_also
   • N/Nする pairs → prominent_see_also
   • Easily confused homophones → prominent_see_also
   • Keigo equivalents → cross_references (keigo)
   • Direct antonyms → cross_references (antonym)

2. MEDIUM — Add when natural:

   • Close synonyms with clear distinction (synonym)
   • Particle contrasts (contrast)
   • Related compounds (related)

3. LOW — Add sparingly:

   • Thematic groupings
   • General see_also references

Extracting from Notes

The notes field often contains vocabulary that should be cross-referenced. Look for:

Patterns to Extract

1. Pair verbs:

   • "Pair: {閉|し}まる" or "PAIR VERB: ..."
   • "The intransitive counterpart is ..."

2. Antonyms:

   • "Opposite: {開|あ}ける"
   • "Antonym: ..."

3. Keigo:

   • "{召|め}し{上|あ}がる (honorific)"
   • "Humble form: いただく"

4. Related words:

   • Words in furigana notation within COMMON PATTERNS
   • Nouns derived from verbs: 食べる → 食べ物

Automated Extraction

Run the extraction script to find potential references:

# Dry run — see proposed changes
python3 build/extract_references.py

# Apply changes
python3 build/extract_references.py --apply

# Single entry
python3 build/extract_references.py --id 00396_taberu

Note: The extraction script now performs immediate resolution. When a target entry exists, the extracted reference automatically includes target_id.

Hardening References

The harden_references.py script scans entries and adds target_id to resolvable cross-references. This "hardens" forward references into direct ID-based references once the target entry exists.

# Dry run — see what would change
python3 build/harden_references.py

# Apply changes
python3 build/harden_references.py --apply

# Single entry
python3 build/harden_references.py --id 00485_shimeru

When to run:

• After adding new entries that are targets of existing forward references
• Periodically to ensure all resolvable references have target_id
• Before releases to maximize resolution coverage

Handling Non-Existent Entries

Important: You can add references to entries that don't exist yet.

• Use reading as the primary key (required)
• Include headword for display purposes
• The link will be marked as "pending" in the web interface
• When the target entry is created, the link automatically becomes active

This allows you to:

• Plan future entries
• Track vocabulary relationships before full coverage
• Show learners related vocabulary even if not yet in dictionary

Validation

After adding references, validate:

python3 build/validate.py --id {entry_id}

The validator checks:

• Required fields present (type, reading)
• Valid type values
• Reading is valid hiragana
• No self-references
• Homonym mismatches — warns when a headword is specified but doesn't match any existing entry with that reading
• Stale target_id — ERRORS when target_id points to a non-existent entry
• Hardenable references — warns when a reference could be hardened (target exists but no target_id)

Validation Messages

Quality Checklist

• Transitive/intransitive pair linked via prominent_see_also (for verbs)
• N/Nする pair linked via prominent_see_also (if applicable)
• Homophones linked via prominent_see_also (if easily confused)
• Keigo forms linked (for common verbs)
• Antonyms linked (if obvious opposite exists)
• References in notes are also in cross_references
• Each reference has correct type
• Reading is valid hiragana
• Headword uses proper furigana notation
• Labels/notes are concise and consistent
• No pair type in cross_references (use prominent_see_also instead)

Symmetry Requirements

Cross-references should be bidirectional for most relationship types. The table below summarizes when back-links are required vs. optional:

Checking symmetry

Use the asymmetry report to find one-way references:

python3 build/find_merge_candidates.py --asymmetry-only

Use the cluster linter to find incomplete semantic groups:

python3 build/check_semantic_clusters.py

Cluster processing

When fixing symmetry issues, process related entries together as a cluster rather than one at a time. This ensures both sides of a relationship are updated in the same session. See the "Cluster Mode" section in prompts/add_cross-references.md for the detailed workflow.