Scholarly Attribution

職業

分類

學術

Disambiguation Priority

When matching a name to a known scholar, use this hierarchy:

1. Nasab chain (4+ generations) — nearly unique
2. Nisba + death date — strong disambiguation
3. Kunyah + nisba — moderate disambiguation
4. Full name (ism + nasab + nisba) — strong
5. Ism + laqab — WEAK — many scholars share this pattern
6. Ism alone — USELESS for disambiguation — never match on ism alone

DANGER: Common Shared Names

These names are shared by multiple prominent scholars. ALWAYS require additional disambiguation:

2. Author Signal Locations in Text

Signal Reliability Hierarchy (highest → lowest)

Extraction Patterns

Colophon patterns (look at last 2-3 pages):

• تم الكتاب بحمد الله / فرغ من تأليفه / وكان الفراغ منه
• على يد (scribe attribution — NOT author)
• كتبه + name (may be author OR scribe — disambiguate from context)
• في شهر [month] سنة [year] — composition or copy date

Introduction patterns (look at first 3-5 pages):

• أما بعد فيقول العبد الفقير + name — author declaration
• قال الشيخ الإمام + name — editorial attribution (lower reliability)
• سألني / طلب مني — commissioning context (identifies audience, not author)
• هذا كتاب + title + ألفه / صنفه + name

Shamela HTML patterns:

• <div class="nass">المؤلف: [name]</div> — check but verify
• Author name in URL path or database ID — very low reliability

3. Multi-Author Scenarios

Commentary Attribution

In a multi-layer text (matn + sharh + hashiyah):

• Each layer has a different author. The sharh author is NOT the matn author.
• The matn text is typically quoted in a distinctive style (often smaller font, or preceded by "قال المصنف")
• The sharh author's voice is the main narrative
• Attribution must track which author said what

Compilation vs. Authorship

• Compiler (جامع/مصنف): Selected and organized others' content. Example: البخاري compiled hadith.
• Author (مؤلف): Wrote original content. Example: الغزالي wrote إحياء علوم الدين.
• Editor (محقق): Modern critical edition editor. NOT the author. Track separately.
• Scribe (ناسخ): Copied the manuscript. NOT the author. Track separately.
• Annotator (معلق): Added footnotes to someone else's text. Separate layer.

KR Pipeline Handling

• author: the primary intellectual author/compiler
• editor: the tahqiq editor (modern, separate field)
• related_scholars: others mentioned (matn author if this is sharh, etc.)
• NEVER merge compiler and author attributions

4. Death Date Verification

Death dates are the single most powerful disambiguator. Two scholars rarely share both name AND death date.

Death Date Signal Sources

Verification Heuristic

1. Extract death date from LLM output
2. Compare against extraction metadata (Shamela)
3. If they MATCH: suspicious — LLM may have parroted extraction, not independently verified (M-4 methodology rule)
4. If they DIFFER: investigate both. If extraction has no date but LLM provides one → genuine inference, verify via web search
5. Cross-check: the death date must be PLAUSIBLE given the text's content (a text referencing events in 800 AH cannot be by an author who died in 600 AH)

Era Plausibility Checks

• Content references to specific events/scholars provide a terminus post quem
• Madhab references: a text following late Hanafi terminology post-dates ~4th century AH
• Language style: classical vs. late medieval vs. modern Arabic differ noticeably
• If death date contradicts content-implied era → FLAG for human review

5. Confidence Scoring

Confidence Levels for Attribution

Name-Only Match Cap

Per KR decision: name-only matches are capped at confidence 0.65. This prevents auto-linking a text to a scholar based solely on a shared name without additional evidence (death date, works list, or biographical confirmation).

6. Common Attribution Errors

Error: Confusing Tahqiq Editor with Author

Pattern: Shamela metadata lists the tahqiq editor (محقق) as the author. Detection: Author death date is modern (post-1350 AH) but text content is clearly classical. Fix: The modern name is the editor. Search the introduction for the original author.

Error: Attributing Compilation to Compiler's Source

Pattern: A mukhtasar (abridgment) is attributed to the original author instead of the abridger. Detection: Text structure is condensed; references "the original" or "the author said." Fix: The mukhtasar author is the abridger. Add mukhtasar_of reference to original.

Error: Confusing Namesakes Across Centuries

Pattern: ابن حجر attributed to d.852 when the text is actually by ابن حجر الهيتمي d.974. Detection: Content discusses Shafi'i fiqh in a late style → likely الهيتمي. Hadith commentary → likely العسقلاني. Fix: Use science + era + works list for disambiguation.

Error: Anonymous Works

Pattern: No author identified in text or metadata. Detection: Introduction uses passive voice ("جُمع هذا الكتاب"), no colophon attribution. Fix: Mark as attribution_status: "unattributed", confidence 0.0. Do NOT assign an author by inference — flag for human research.

Gotchas

1. Honorifics are NOT part of the base name — Strip الشيخ, الإمام, العلامة, الحافظ, شيخ الإسلام before matching. But PRESERVE them in displayed names — they carry scholarly status information.

2. كنية (kunya) is not the given name — أبو حنيفة is a kunya (father of Hanifa), not a first name. The ism (given name) is النعمان. Search engines and databases may index under either form.

3. نسبة (nisba) indicates origin, not always identity — الدمشقي means "from Damascus" and applies to hundreds of scholars. It's useful for disambiguation only in combination with other name components.

4. Shared names across centuries — At least: 3 scholars named ابن تيمية, 4 named ابن قدامة, 5+ named ابن حجر. Always require death date or specific qualifier for unique identification.

5. رحمه الله confirms the scholar is deceased — Presence of this formula after a name is useful for death date estimation: if the text is dated, the scholar died before the text was written.