Nisaba — Features

Overview#

Nisaba is the Ancient Text Analysis and Cross-Tradition Scholarly Research platform. Named after Nisaba (𒀭𒉀), the Sumerian goddess of writing, learning, grain, and reeds — credited in ancient tradition as the inventor of cuneiform script, the earliest writing system in human history — the domain provides the world's most comprehensive, AI-powered scholarly workbench for ancient texts across all world traditions. Her temple, the É-zagin ("House of Lapis Lazuli"), was conceived as the prototype of every library and archive that would follow. Nisaba likewise gathers every tradition's texts under a single scholarly roof.

The domain supports the complete scholarly lifecycle: reading a Sumerian cuneiform tablet or Dead Sea Scrolls fragment in its original script; morphological analysis; translation comparison; textual criticism; paleographic dating; etymological and intertextual research; and finally producing publication-ready digital critical editions compliant with international scholarly standards. Nisaba's 22 libraries cover ancient languages spanning cuneiform scripts (Sumerian, Akkadian, Hittite, Elamite, Ugaritic), Egyptian scripts (hieroglyphic through Coptic), Semitic scripts (Hebrew, Aramaic, Arabic, Ge'ez), South Asian scripts (Sanskrit, Pali, Tibetan), Greco-Roman languages (Greek, Latin, Avestan), Classical Chinese, and Old Church Slavonic — alongside canonical reference systems for every major scriptural tradition, textual criticism tools, digital critical edition publication, AI scholarly assistance, and connections to open-access digital humanities corpus platforms.

Domain Libraries#

The table below lists all 22 @nisaba/* packages and their responsibilities. See architecture.md for the dependency graph and per-package internals.

Ancient Language Engine (@nisaba/languages)#

Writing System Support#

Nisaba provides first-class support for reading, displaying, and analyzing texts in their original scripts — not merely transliteration. This means rendering cuneiform signs, polytonic Greek, right-to-left Hebrew and Arabic, Devanagari, Chinese characters, and dozens of other writing systems with scholarly-grade Unicode and font handling.

Semitic Scripts#

• Biblical Hebrew: Full Unicode support for consonants, Tiberian vowel points (all nine classes: qamets, patach, tsere, segol, hiriq, holem, qibbuts, shureq, qamets hatuf), all cantillation marks (te'amim), dagesh lene and forte distinction, shin/sin dot, maqqef, sof pasuq, and ketiv/qere notation. Also supports the alternative Babylonian and Palestinian vocalization systems — scholarly traditions distinct from the dominant Tiberian text.

• Aramaic: Imperial Aramaic; Biblical Aramaic (using Hebrew square script with Aramaic-specific features); Syriac in all three script variants — Estrangela (the oldest, used in classical literature), Serto (West Syriac/Jacobite tradition), and Eastern/Madnhaya (Nestorian tradition) — each with their respective diacritical vowel mark systems.

• Classical Arabic: Full Unicode for consonants, short vowels (fatha/kasra/damma), tanwin (nunation endings), sukun (quiescence marker), shadda (gemination), and all five hamza positions. Full Quranic orthographic conventions including tajweed markers — idgham (assimilation), iqlab (transformation), ikhfa (concealment), ghunna (nasalization) — and all pause marks (mandatory, permissible, prohibited) and sajda indicators.

• Ge'ez/Ethiopic: Full syllabary (base consonants multiplied by 7 vowel orders), Amharic extensions, the traditional numeral system, and all standard punctuation including the word divider, full stop, comma, colon, preface colon, and question mark.

• Phoenician, Paleo-Hebrew, Moabite, Samaritan: Unicode block mappings and bidirectional transliteration for early Northwest Semitic scripts and inscriptions.

Cuneiform Scripts#

• Sumerian cuneiform: Logographic and syllabic readings; determinative classification — the semantic classifiers (dingir/god, ki/place, lu2/person, giš/wood, etc.) that mark word type without being pronounced; full sign list integration with Borger's MZL (Mesopotamisches Zeichenlexikon), aBZL (Assyrisches Zeichenlexikon), and CDLI sign values; sign variant tracking; composite sign decomposition.

• ATF (ASCII Transliteration Format): Full parser and generator for the CDLI and ORACC standard. ATF is the scholarly format for representing cuneiform in ASCII — it encodes sign names, sign values, determinatives, glosses, damage notation (broken, effaced, illegible), editorial marks, and line numbering. Round-trip fidelity between ATF and Unicode cuneiform is guaranteed.

• Akkadian: Old Babylonian, Standard Babylonian, and Neo-Assyrian dialect conventions, with syllabic value disambiguation and logogram-to-Akkadian reading resolution using context heuristics.

• Hittite and Elamite: Language-specific sign value tables and determinative conventions for two of the most important ancient Near Eastern languages preserved on clay tablets.

• Ugaritic alphabetic cuneiform: The 30-letter alphabet written in cuneiform — one of the world's earliest alphabets — with word divider and transliteration support.

Egyptian Scripts#

• Hieroglyphic: Gardiner sign list integration (all categories A–Z and Aa), Manuel de Codage (MdC) encoding for scholarly input, JSesh-compatible rendering output, cartouche boundaries, direction indicators (left-facing/right-facing), and support for Middle Egyptian, Late Egyptian, and Ptolemaic sign extensions.

• Hieratic: The cursive script derived from hieroglyphics used for administrative and literary texts on papyrus; period-specific paleographic conventions and hieratic-to-hieroglyphic conversion tables.

• Demotic: Unicode support and transliteration following standard conventions for this highly cursive late-period Egyptian script.

• Coptic: Full support for all five major dialects — Sahidic (the primary literary dialect), Bohairic (the liturgical dialect of the Coptic Orthodox Church), Akhmimic, Fayyumic, and Lycopolitan — including dialect-specific letters and supralinear strokes.

South Asian Scripts#

• Sanskrit/Devanagari: Full Unicode for consonants, vowels, conjunct consonants (ligatures), virama (consonant killer), anusvara, visarga, chandrabindu, avagraha, and all Vedic accent marks (udatta, anudatta, svarita). Multiple transliteration schemes: IAST (primary scholarly standard), Harvard-Kyoto (ASCII-only), SLP1 (compact computational format), Velthuis (legacy TeX), and ISO 15919.

• Pali multi-script: Devanagari, Roman (PTS transliteration conventions), Sinhala, Thai, Burmese, and Khmer script representations, all interconvertible while preserving semantic identity.

• Prakrit: Ardhamagadhi (the Jain canonical language), Shauraseni, and Maharashtri Prakrit with dialect-specific phonological rules.

• Tibetan: Wylie transliteration support, proper consonant cluster stack construction, tsheg (syllable separator), and shad (phrase/sentence delimiter).

• Indic inscriptional scripts: Tamil Brahmi, Grantha, and Kharoshthi for early South Asian epigraphy and manuscript traditions.

Greco-Roman and Other Ancient Scripts#

• Ancient Greek: Full polytonic accentuation — acute, grave, circumflex, rough/smooth breathing marks, iota subscript/adscript, diaeresis, coronis — along with dialectal orthographic variants (Attic, Ionic, Doric, Aeolic, Koine) and both Attic and Ionic/Milesian numeral systems.

• Latin: Classical orthographic conventions (including inscriptional conventions without U/V and I/J distinction), Medieval Latin extensions, Ecclesiastical Latin conventions, and abbreviation expansion for the most common medieval scribal abbreviations.

• Avestan and Old Persian cuneiform: The language of the Zoroastrian Gathas and the later Avestan corpus; the cuneiform syllabic script of the Achaemenid Persian inscriptions.

• Old Church Slavonic: Both Glagolitic (the oldest Slavic alphabet) and Early Cyrillic; runic scripts for Elder Futhark, Younger Futhark, and Anglo-Saxon Futhorc; Linear B syllabic script for Mycenaean Greek.

• Classical Chinese: Traditional characters (not simplified), ML-assisted word segmentation for Literary Chinese texts that have no word boundary markers, punctuation insertion heuristics for unpunctuated classical texts, and variant character tracking (異體字).

Morphological Analysis#

Morphological analysis — breaking a word down into its grammatical components (root, stem, inflectional ending, pronominal suffix, etc.) — is the foundation of reading ancient texts. Nisaba includes paradigm-based parsers, not mere lookup tables, meaning it can analyze forms that don't appear in any dictionary.

• Hebrew: All seven major verb stems (Qal, Niphal, Piel, Pual, Hiphil, Hophal, Hithpael) plus less common derived stems; all conjugations (perfect, imperfect, imperative, infinitive construct, infinitive absolute, participle); weak root detection for all major root types (I-Nun, I-Yod, I-Aleph, II-Waw/Yod, III-He, Geminate); complete pronominal suffix decomposition (both subject and object suffixes); construct chain analysis; waw-consecutive identification; prefix stripping for inseparable prepositions, conjunctive waw, definite article, and relative pronoun.

• Greek: Full nominal paradigm (all three declensions, all cases including dual); complete verbal paradigm (all six tenses: present, imperfect, future, aorist, perfect, pluperfect; all four moods: indicative, subjunctive, optative, imperative; plus infinitive and all participle forms; all three voices); augment detection (syllabic and temporal); contract verb resolution for alpha, epsilon, and omicron contracts; participle agreement parsing; enclitic and proclitic handling.

• Latin: All five nominal declensions including irregulars; four conjugations plus all irregular verbs (sum, eo, fero, volo, possum, etc.); deponent and semi-deponent detection; ablative absolute identification; gerund/gerundive distinction; supine recognition.

• Sanskrit (Rust/WASM implementation for performance): External sandhi decomposition — sandhi is the phonological fusion that makes Sanskrit text appear as a continuous stream without word breaks; all vowel sandhi, consonant sandhi, and visarga sandhi rules; internal sandhi; compound analysis with classification of all five compound types (tatpurusha, dvandva, bahuvrihi, avyayibhava, dvigu); vibhakti (case) parsing for 8 cases × 3 numbers; verb conjugation for all 10 present classes and all tenses and moods; secondary derivation identification.

• Arabic: Root-pattern morphology — extracting the trilateral or quadrilateral root from any derived nominal or verbal form; Form I–X (and uncommon) verb form identification with semantic description per form; broken plural detection and classification; idafa (construct state / annexation chain) analysis; dual number support; passive voice detection.

• Akkadian: G-stem (basic), D-stem (intensive), Š-stem (causative), N-stem (reflexive/passive), plus Gt, Dt, Št, and Ntn iterative forms; stative conjugation; ventive suffix detection; subordination marker handling.

• Sumerian: Verbal chain analysis covering dimensional prefixes, modal prefixes, conjugation prefixes, pronominal elements, stem, and pronominal suffixes; nominal chain parsing (noun + possessive + case marker); finite/non-finite verb distinction.

• Pali: Sandhi decomposition; nominal declension (six groups); verb conjugation (present, aorist, future, imperative, optative, conditional, gerund, infinitive, participle); compound analysis.

• Coptic: Tripartite conjugation system (First Present, Second Present, Circumstantial, and other patterns); prenominal/prepronominal state handling; bound group analysis.

• Configurable token granularity: Tokenize at morpheme, word, phrase, or clause level depending on the analytical task.

• Position-preserving tokenization: Byte offsets are preserved so every token can be precisely anchored to the source text for standoff annotation alignment.

• Dictionary-form lemmatization: Returns the canonical dictionary form (lemma) of any analyzed word, with contextual disambiguation when multiple lemmas are possible.

Lexicon Integration#

Rather than directing scholars to external websites, Nisaba embeds direct access to the major scholarly lexica within the platform through a unified LexiconService API that provides a single lookup(word, language, options) interface returning structured entries from all available lexica for that language.

• Biblical Hebrew/Aramaic: Brown-Driver-Briggs (BDB), the standard scholarly reference; lexicon extension API supports licensed resources such as HALOT (Koehler-Baumgartner) via a plugin interface.
• Classical Greek: Liddell-Scott-Jones (LSJ), the comprehensive Greek lexicon (via Perseus, public domain); BDAG for New Testament Greek (license-aware plugin interface).
• Latin: Lewis & Short (via Perseus, public domain); principal parts cross-referencing.
• Sanskrit: Monier-Williams Sanskrit-English Dictionary (public domain); Devanagari/IAST dual lookup; Vedic/Classical usage distinction.
• Akkadian: Chicago Assyrian Dictionary (CAD), all 21 volumes (public domain / open access).
• Sumerian: Pennsylvania Sumerian Dictionary (PSD) and ePSD2 (the updated electronic edition).
• Arabic: Lane's Arabic-English Lexicon (public domain), the most comprehensive classical Arabic reference.
• Pali: PTS Pali-English Dictionary (Pali Text Society, public domain).
• Cross-language features: Semantic domain grouping following the Louw-Nida model; cross-language cognate linking via Proto-Semitic and Proto-Indo-European root reconstruction; frequency data per corpus showing how common or rare a word is within its tradition.

Transliteration Engine#

Transliteration converts a text from one writing system to another — typically from a non-Latin script to a romanized academic representation. Nisaba supports bidirectional transliteration with no data loss, using whichever scholarly convention is standard for each language.

• Hebrew: SBL academic, simplified (no diacritics), general-purpose with full vowel representation, and Library of Congress (ALA-LC) romanization.
• Greek: Round-trip with proper handling of rough/smooth breathing and polytonic accents; both scholarly (accent-preserving) and simplified ASCII modes.
• Sanskrit: IAST, Harvard-Kyoto, SLP1, Velthuis, and ISO 15919 bidirectional conversion.
• Arabic: DIN 31635 (primary scholarly standard), ALA-LC (Library of Congress), and Buckwalter (computational format used in NLP pipelines).
• Cuneiform: ATF conventions for sign values, determinatives (superscript notation), and damage notation.
• Round-trip fidelity: A text can be transliterated and then reverse-transliterated back to the original without data loss — validated for every language and scheme.

Translation Comparison Engine (@nisaba/translations)#

Ancient texts have been translated many times over centuries, and comparing those translations is central to scholarly work. Nisaba provides multiple display modes and automated analysis of translation differences.

Display Modes#

• Side-by-side: Parallel columns showing two or more translations simultaneously, synchronized by verse, line, or paragraph anchors. Configurable number of columns (2–6); independent scrolling with sync-lock toggle; visual alignment indicators for corresponding segments.

• Synoptic view: Horizontal layout showing all versions of a passage simultaneously with color-coded alignment markers and expandable/collapsible segment groups — the standard format for synoptic Gospel study, generalized to any multi-witness comparison.

• Interlinear: The source language text is presented word by word, with a morphological parsing line (part of speech, stem, grammatical form), a lexical gloss line (word-by-word translation), and one or more smooth translation lines beneath each source word. Supports right-to-left source texts with left-to-right glosses — the most information-dense format for students learning to read ancient languages.

• Reverse interlinear: The translation text is the primary display; source language words and morphological data appear above each translated word — allowing readers to follow a translation while maintaining connection to the original.

• Parallel text: Sentence-aligned original and translation in flowing, readable format.

Alignment Algorithms#

• Statistical methods: Gale-Church sentence alignment; IBM Model 1–5 word alignment; Hidden Markov alignment models; phrase-level bilingual concordance alignment.
• Neural methods: Transformer-based cross-lingual alignment models fine-tuned for ancient language pairs; attention-based word alignment extraction from pre-trained multilingual models.
• Manual alignment: Scholar-driven alignment interface with accept/reject controls; exportable as a parallel corpus for computational use.

Divergence Detection#

The places where translations genuinely differ from each other — or from the source — are often where the most important scholarly questions live. Nisaba surfaces these divergences automatically.

• Semantic divergence scoring: Quantifying how different translations are at word, phrase, clause, and sentence levels.
• Theological divergence flagging: Detecting translation choices that appear to reflect doctrinal presuppositions rather than purely philological judgment — for example, "virgin" vs. "young woman" at Isaiah 7:14, or "peace" vs. "sword" at Matthew 10:34.
• Omission and addition detection: Material present in one translation but absent from another, which may reflect different source text traditions.
• Formal vs. dynamic equivalence scoring: Locating a translation on Eugene Nida's spectrum between word-for-word (formal equivalence) and meaning-for-meaning (dynamic equivalence) rendering.

Translation History Timeline#

• Chronological tracking: How a passage has been rendered across history — from the earliest translations (the Septuagint for the Hebrew Bible, the Vetus Latina for the New Testament, the Targumic tradition) through medieval and modern translations.
• Influence mapping: Detecting where later translations borrowed specific phrasing from earlier ones, building a genealogy of translation language.
• Cultural context annotations: Explaining the historical, theological, and political circumstances that shaped major translation decisions.

Textual Criticism Engine (@nisaba/criticism)#

Textual criticism is the discipline of establishing the most accurate possible text of an ancient work by comparing all surviving manuscript witnesses. It is the foundation of all serious scholarly editions.

Manuscript Witness Registry#

• Full witness metadata: Siglum (the scholar's abbreviation), date range with confidence level, provenance (origin and custody history), script type, writing material (papyrus, parchment, paper, clay, stone, ostraca), physical dimensions, current repository and catalog number, digitization status and IIIF manifest URL.
• Witness classification: Majuscule (uncial, capital-letter manuscripts) vs. minuscule (cursive manuscripts); scroll vs. codex format; text family classification (for New Testament: Alexandrian, Byzantine, Western, and Caesarean traditions).
• Family group support: Manuscript families such as f1 (the Lake Group) and f13 (the Ferrar Group) in NT criticism, allowing family-level analysis rather than only individual manuscript work.
• Versional witnesses: Ancient translations (LXX, Vulgate, Peshitta, Coptic versions, Old Latin) treated as witnesses to their source text.
• Patristic citation tracking: Quotations in early church fathers, rabbis, and ancient commentators treated as indirect witnesses to the text they knew.
• IIIF manifest linking: Each manuscript record can link to its IIIF manifest, enabling side-by-side viewing of transcription and manuscript image from any compatible viewer.

Collation#

Collation is the systematic comparison of manuscript texts to identify every point of difference — called a variant unit.

• Automatic collation: a CollateX-compatible engine that performs Needleman-Wunsch token alignment, then constructs a directed acyclic variant graph of all witnesses to identify agreements and disagreements.
• Fuzzy matching: Handles orthographic variation common in manuscripts — itacism (confusion of similar-sounding vowels in Greek), matres lectionis inconsistencies in Hebrew — with language-specific normalization rules.
• Transposition detection: Identifying when the same words appear in different orders across witnesses.
• Block alignment: Handling large-scale structural variants where entire sections are in different positions.
• Manual collation interface: Accept/reject controls for cases where automatic collation needs refinement; custom variant unit boundary adjustment; uncertain reading markers.
• Output formats: Variant graph (JSON), alignment table (TSV/Excel), TEI-XML critical apparatus, and JSON interchange.

Variant Classification#

• Automatic ML-assisted classification into: orthographic, morphological, lexical, syntactic, semantic, substantial, or nonsense variant categories.
• Scribal error taxonomy: Haplography (accidental omission of a repeated sequence); dittography (accidental doubling); homoioteleuton and homoioarcton (eye-skip between similar endings or beginnings); metathesis (letter reversal); itacism; harmonization (assimilating to a parallel passage); lectio difficilior preference (the classical principle that the more difficult reading is usually original); deliberate theological alteration detection.

Stemmatic Analysis#

Stemmatics constructs a family tree (stemma codicum) showing how manuscripts are descended from each other, ultimately pointing toward the original.

• Phylogenetic methods: Maximum parsimony (Fitch algorithm); maximum likelihood estimation; Bayesian phylogenetics; NeighborNet for cases with conflicting signals suggesting manuscript contamination.
• Contamination detection: Consistency index analysis, split decomposition, and CBGM (Coherence-Based Genealogical Method) compatibility analysis for manuscripts copied from multiple exemplars.
• Lachmannian reconstruction: The classical method — identifying conjunctive errors (shared mistakes proving relationship) and separative errors (unique mistakes ruling out direct copying); archetype reconstruction.
• Interactive stemma visualization: Clickable stemma diagrams with confidence indicators; ability to propose alternative stemmata; export to Newick/Nexus format for phylogenetics software.

Critical Apparatus#

The critical apparatus is the system of footnotes documenting all significant textual variants and their manuscript support — the defining feature of a scholarly edition.

• Positive, negative, and mixed apparatus formats: Positive lists all witnesses supporting each reading; negative lists only witnesses diverging from the lemma; mixed uses both where appropriate (following Nestle-Aland conventions for NT criticism).
• Leiden conventions: The standardized notation system for epigraphic and papyrological texts.
• Siglum and abbreviation management: Complete management of manuscript abbreviations, versions, and scholarly sigla — with witness grouping for apparatus compactness.
• Automatic apparatus generation from collation data, with manual refinement for scholarly judgment calls.

Lacuna Tracking#

A lacuna is a physical gap in a manuscript — a hole, torn section, or area of damage — where text is lost.

• Lacuna registry: Extent of each lacuna, probable cause, and partial legibility notes.
• Proposed restorations: Scholarly conjectures about missing text, with confidence levels (certain, probable, possible, speculative) and supporting evidence.
• Lacuna visualization: Overlay of proposed readings on manuscript images.

Canonical Reference Management (@nisaba/canon)#

Canonical reference systems are the standardized ways of citing locations within sacred texts — systems that must be precisely understood for interoperability between databases, commentaries, and scholarly tools.

Biblical Canon Systems#

• Protestant canon (66 books), Catholic (73 books including deuterocanonicals: Judith, Tobit, 1–2 Maccabees, Wisdom, Sirach, Baruch), Greek Orthodox (additional: 1 Esdras, 3 Maccabees, Prayer of Manasseh), and Ethiopian Orthodox (including 1 Enoch, Jubilees, and additional books).
• Hebrew Bible (Tanakh) with traditional Torah/Nevi'im/Ketuvim ordering, which differs from Christian Old Testament ordering.
• Septuagint (LXX) with Rahlfs numbering, plus full versification mapping to the Masoretic Text — essential because chapter and verse numbers often differ (most famously in the Psalms, where LXX Psalm 9 = MT Psalms 9–10, producing a persistent off-by-one throughout the Psalter).
• Versification conversion engine: Automated conversion between MT, LXX, and standard English versification systems.

Quranic Reference#

• Surah/ayah numbering for all 114 surahs.
• Juz, hizb, rub: The traditional divisions of the Quran into 30 equal parts (juz') used for Ramadan recitation, and their subdivisions (hizb = half juz'; rub = quarter hizb).
• Qira'at variant tracking: The ten canonical readings of the Quran — systems of pronunciation and occasional textual variations representing different chains of oral transmission. The most common are Hafs (dominant in most Muslim-majority countries) and Warsh (dominant in North and West Africa). All ten are supported.
• Asbab al-nuzul: Linking verses to the traditional "occasions of revelation" — the historical contexts in which each verse was reportedly revealed.

Buddhist Canon Systems#

• Pali Canon: PTS (Pali Text Society) page numbering; Thai, Burmese, and Sinhalese edition numbering; standard sutta reference system (e.g., DN 22 = Digha Nikaya, Sutta 22).
• Chinese Buddhist Canon: Taishō Tripiṭaka with volume/page/column/line reference (the universal standard for Chinese Buddhist scholarship); CBETA reference system; Dunhuang manuscript numbering.
• Tibetan Buddhist Canon: Kangyur (the words of the Buddha) and Tengyur (Indian commentaries) classification; Derge edition numbering; Peking, Narthang, and Cone edition cross-referencing; Tohoku catalog numbers.

Hindu Scripture Systems#

• Vedic corpus: Rigveda mandala/sukta/mantra; Samaveda, Yajurveda (Black and White), Atharvaveda; Brahmanas, Aranyakas, and the major Upanishads.
• Epics and Bhagavad Gita: Mahabharata with both Critical Edition (Pune/Bori) and Vulgate numbering; Ramayana; Bhagavad Gita by chapter and verse.
• Sutras and Puranas: Yoga Sutras of Patanjali, Brahma Sutras, all 18 major Puranas, Dharmasutras.

Jewish Textual Systems#

• Talmud: Both Babylonian (Bavli) and Jerusalem (Yerushalmi) Talmuds with tractate/folio/amud (side: a = recto, b = verso) reference (e.g., Berakhot 2a).
• Mishnah and Tosefta: Tractate/chapter/paragraph.
• Midrash collections: Midrash Rabbah, Tanhuma, Pesikta de-Rav Kahana, and others.
• Kabbalah: Zohar with parasha/daf reference.

Classical Text Systems#

• Greek: Stephanus pages for Plato (the 16th-century pagination still used as the universal reference after 450 years); Bekker numbering for Aristotle; Diels-Kranz numbering for the Pre-Socratics; standard line numbering for Homer, Aeschylus, Sophocles, Euripides, Aristophanes.
• Other traditions: Avesta (yasna/ha/paragraph); Confucian Analects (book/chapter); Daodejing (chapter); I Ching (hexagram/line).

Cross-System Reference Resolution#

A core challenge in multi-tradition scholarship is that every tradition has its own citation conventions, and the same passage can be expressed in a dozen incompatible ways. Nisaba solves this with a canonical URI scheme that gives any passage a single stable identifier regardless of which convention was used to look it up.

• Canonical URI scheme: nisaba://tradition/work/reference (e.g., nisaba://bible/gen/1:1, nisaba://quran/2:255, nisaba://buddhist/pali/dn/22). A single stable identifier for any passage in any tradition.
• CTS/CITE protocol compatibility: Interoperability with the Perseus Digital Library and other digital humanities infrastructure.
• Bidirectional mapping: Convert any reference to any other reference system for the same passage (e.g., MT psalm numbering to LXX psalm numbering).

Philological Analysis (@nisaba/philology)#

Philology is the study of language in historical written sources — combining linguistics, literary criticism, and history to understand texts in their full depth.

Etymology and Semantic Analysis#

• Cross-language cognate chain tracking: Following word families across language boundaries — Proto-Semitic roots through Hebrew, Aramaic, Arabic, Akkadian, Ugaritic, and Ge'ez; Proto-Indo-European roots through Sanskrit, Greek, Latin, Avestan, and Old Persian; Sino-Tibetan root relationships; Egyptian-Coptic etymological continuity.

• Semantic drift visualization: Showing how word meanings changed across languages and centuries. For example, Hebrew ruach (wind/breath/spirit) → Greek pneuma → Latin spiritus → English "spirit" — with attestation timeline and usage context examples for each stage.

• Loanword detection: Identifying borrowed words across language boundaries — Sumerian loanwords in Akkadian; Persian loanwords in Biblical Hebrew; Greek loanwords in Rabbinic Hebrew; Sanskrit loanwords in Pali; Arabic loanwords in Persian, Turkish, and Urdu — using phonological correspondence rules and chronological filtering.

• Semantic domain mapping: Following the Louw-Nida semantic domain framework for NT Greek and the Semantic Dictionary of Biblical Hebrew approach; cross-linguistic semantic field comparison (e.g., the different conceptual ranges of Hebrew chesed, Greek agape/philia/eros, and Sanskrit prema/bhakti/karuna).

• Collocational analysis: Identifying statistically significant word co-occurrences and fixed phrases within each corpus; metaphor mapping following Lakoff-Johnson conceptual metaphor theory applied to ancient texts.

Hapax Legomena and Statistical Linguistics#

A hapax legomenon is a word that appears only once in the entire corpus — often the most challenging and disputed term to translate, since no other occurrences provide context.

• Hapax legomena detection: Automatic flagging of words occurring only once (hapax) or twice (dis legomena) in a corpus, with cross-corpus checking to determine whether a NT hapax appears in the LXX or Pseudepigrapha.

• Word frequency analysis: Type-token ratios, vocabulary richness metrics (Yule's K, Simpson's D, hapax percentage), and corpus comparison statistics.

• Stylometric authorship attribution: Using function word frequencies, sentence length distributions, vocabulary overlap, and other statistical features to analyze disputed authorship questions (e.g., Pauline epistles, Isaianic unity, the Documentary Hypothesis for the Pentateuch).

• Register analysis: Distinguishing prose from poetry, narrative from discourse, legal from prophetic — based on statistical profiles of syntactic patterns, vocabulary, and discourse markers.

Discourse Analysis#

• Chiastic structure detection: Automatic identification of chiasmus (A-B-C-B'-A' patterns) — a pervasive structural device in ancient Near Eastern and biblical literature — with configurable detection parameters and confidence scoring.

• Ring composition detection: Identifying Homeric ring structure (where a narrative opens and closes with the same theme) and inclusio (simpler framing where a passage begins and ends with identical or near-identical words).

• Discourse boundary detection: Using discourse markers, topic shifts, participant changes, temporal indicators, and formulaic expressions to automatically segment texts into discourse units.

• Narrative analysis: Character tracking across a narrative; speech/narration distinction; reported speech nesting; narrative perspective identification.

Intertextuality Detection#

• Quotation detection: Identifying explicit quotations of earlier texts in later ones (OT quotations in NT, Vedic citations in Upanishads, Quranic references to Biblical narratives) with source identification and citation format analysis.

• Allusion detection: Identifying probable but inexact references using shared vocabulary clusters, thematic parallels, and structural similarities with configurable sensitivity thresholds.

• Echo detection: For faint thematic or verbal resonances that may or may not represent conscious literary dependence — producing scored results with supporting evidence.

• Influence scoring: Quantifying probability and strength of intertextual connections; dependency direction analysis determining which text is source and which is dependent.

Paleographic Analysis (@nisaba/paleography)#

Paleography is the study of ancient handwriting — used to date manuscripts, identify scribal hands, and authenticate documents.

• Script classification from images: ML classifier identifying writing systems from manuscript images, even for partially damaged manuscripts.
• Period classification: Estimating a manuscript's date range from letterform shapes, ductus (the order and direction of strokes), and comparative paleographic databases.
• Regional classification: Identifying scribal schools by regional letterform characteristics — distinguishing Palestinian from Babylonian Hebrew scripts, or Alexandrian from Byzantine Greek hands.
• Letterform analysis: Quantitative measurement of stroke angle, aspect ratio, serif presence, baseline consistency, and ductus for systematic comparative work.
• Scribal hand identification: ML models trained on manuscript images for writer identification across documents — determining whether two manuscripts share a scribe.
• Collaborative vs. single-scribe detection: Identifying hand changes within a single manuscript, which may indicate different scribes, different copying sessions, or later additions.
• Multispectral imaging integration: Processing infrared, UV, XRF, and RTI (Reflectance Transformation Imaging) data to read faded, erased, or overwritten text — including palimpsests, manuscripts where the original text was scraped off and reused.
• AI-assisted letter reading: For damaged or partially legible characters, providing ranked candidate readings with confidence scores.

Cross-Tradition Comparative Analysis (@nisaba/comparative)#

Concept Equivalence Registry#

The concept equivalence registry is a curated scholarly database of analogous concepts across religious and philosophical traditions — carefully mapping where they converge, diverge, and resist comparison, without assuming equivalence.

• Curated cross-tradition concept mappings with similarity taxonomy (genetic, typological, analogical, contested) and evidence grading. Examples:

  • Ineffability: Tao ~ Brahman ~ Sunyata ~ Ein Sof ~ Theologia Negativa ~ Fana
  • Liberation: Moksha ~ Nirvana ~ Theosis ~ Salvation ~ Fana ~ Wu Wei
  • Universal ethics: The Golden Rule across Confucius, Hillel, Jesus, Muhammad, and Buddhist precepts
  • Cosmic order: Logos ~ Rta/Dharma ~ Ma'at ~ Tao ~ Torah
  • Self-transcendence: Anatta ~ Fana ~ Kenosis ~ Bittul HaYesh ~ Wu Wei

• Evidence requirements: Every mapping must cite primary sources, scholarly literature, and specify the level of scholarly consensus — preventing superficial syncretism.

• Version history: Tracking how concept mappings evolve as scholarship progresses.

Parallel Passage Detection#

• Semantic similarity algorithms: Detecting parallel passages across traditions using vector similarity of cross-lingual embeddings.
• Motif matching: Identifying shared narrative or thematic motifs (flood narratives, dying-and-rising deity figures, creation from chaos) across the Aarne-Thompson-Uther universal motif catalog.
• Structural analysis: Detecting structural parallelism in text organization (three-part narratives, wisdom instruction forms, lament patterns).

Influence Network#

• Directed influence graph with evidence-graded edges showing probable textual dependencies — mapping, for example, how Second Temple Jewish apocalyptic literature influenced early Christian texts, or how Stoic philosophy influenced Philo of Alexandria.

Open-Source Corpora Integration (@nisaba/corpora)#

Format Parsers#

The major digital humanities repositories use different text encoding formats. Nisaba provides parsers for all of them, converting each into a unified internal representation.

• TEI-XML (Text Encoding Initiative): Full TEI P5 support including teiHeader metadata, structural elements (div, p, l, lg, sp, stage), critical apparatus elements (app, lem, rdg, wit), named entities (persName, placeName, orgName), linguistic annotations, manuscript descriptions, and bibliographic data. Streaming parser for large files. Also TEI P5 generator with EpiDoc variant for epigraphic and papyrological texts.

• ATF (ASCII Transliteration Format): Full parser and generator for the CDLI/ORACC cuneiform standard.

• USFM/USX: United Bible Societies standard format for Bible software; OSIS XML (Open Scripture Information Standard); SWORD module format.

• OpenITI mARkdown: The format used by the Open Islamicate Texts Initiative for Arabic and Persian texts.

Corpus Connectors (13+ platforms)#

The connectors below bring external digital humanities corpora into the Nisaba ingestion pipeline, making their texts searchable and analyzable alongside all other connected materials. Each connector handles metadata ingestion, text retrieval, and mapping of the corpus's native citation scheme to Nisaba canonical URIs.

• Perseus Digital Library: Greek and Latin texts with morphological data, translations, and commentary.
• ETCSL (Electronic Text Corpus of Sumerian Literature): Complete Sumerian literary corpus with transliteration and translation.
• CDLI (Cuneiform Digital Library Initiative): Cuneiform tablet catalog with ATF transliterations.
• ORACC (Open Richly Annotated Cuneiform Corpus): Annotated cuneiform texts across specialized sub-projects.
• SuttaCentral: Pali, Chinese, Sanskrit, and Tibetan Buddhist texts with translations and parallel identifications.
• CBETA (Chinese Buddhist Electronic Text Association): Complete Chinese Buddhist canon (Taishō Tripiṭaka).
• Sefaria: Jewish texts (Tanakh, Talmud, Midrash, Halakha, Kabbalah, Liturgy, Philosophy) with Hebrew/Aramaic originals and translations.
• OpenITI: Arabic and Persian texts in mARkdown format.
• SARIT and GRETIL: Sanskrit and Prakrit texts in TEI-XML.
• CTP (Chinese Text Project): Pre-modern Chinese texts (Confucian, Taoist, Legalist, etc.).
• Dead Sea Scrolls Digital Library: Transcriptions and image references.
• ETCBC (Eep Talstra Centre for Bible and Computer): Morphologically analyzed Hebrew Bible.
• SWORD modules: Bible translations, original language texts, lexica, and commentaries.

Ingestion Pipeline#

The ingestion pipeline converts corpus content into Nisaba's internal representation. Chunking strategy is chosen based on text genre, because inappropriate chunk boundaries would break verse references, narrative pericopes, or synoptic alignments.

• Verse-aware chunking for poetry and scripture, respecting verse boundaries.
• Pericope-aware chunking for narrative texts, using discourse boundary detection.
• Section-aware chunking for prose treatises, following the author's structural divisions.
• Parallel-preserving chunking: Keeping synoptic parallels aligned across chunk boundaries for cross-tradition vector search.
• Corpus synchronization scheduler: Periodic re-ingestion of updated corpora with incremental update detection and conflict resolution.

Annotation and Commentary System (@nisaba/annotations)#

Standoff annotation stores annotations separately from the base text, referenced by character positions — rather than embedding markup within the text. This allows multiple annotation layers to overlap without conflict and prevents annotations from corrupting source text.

• W3C Web Annotation Data Model (WADM) storage with TextPositionSelector (character offset ranges) for multiple overlapping layers.

• Annotation type system: Textual notes (commentary, explanation, cross-reference); linguistic annotations (morphological, syntactic, semantic); translation annotations (gloss, paraphrase, formal translation); critical annotations (variant readings, emendation proposals); comparative annotations (cross-tradition parallel markers); structural annotations (discourse markers, chiastic boundaries); pedagogical annotations (difficulty level, prerequisite knowledge).

• Talmudic-style layered commentary model: Base text layer → first commentary layer (direct commentary on the base text) → second commentary layer (commentary on the commentary, as in the Talmud's Mishnah + Gemara + medieval commentators structure). Essential for corpora where the tradition of commentary is as important as the text itself.

• Annotation body formats: TEI-XML, Markdown, and plain HTML — scholars are not locked into a single format.

• Batch annotation operations: Create, update, or delete annotations across a passage range in a single operation.

• Export: W3C JSON-LD format and TEI-XML for use in publication and data interchange.

Digital Critical Edition Tools (@nisaba/editions)#

A digital critical edition is a scholarly publication combining a carefully established text, a critical apparatus, translations, and commentary — produced and distributed digitally.

• Edition project management: Full lifecycle management (draft → review → published); collaborator roles (editor-in-chief, contributing editor, translator, reviewer); milestone tracking.
• Apparatus attachment: Attaching and managing the critical apparatus per edition, with automated generation from collation data and manual refinement.
• Translation management: Managing multiple translations within a multi-lingual edition.
• Multi-format export: TEI-XML (the international standard for digital editions), EPUB, PDF, LaTeX (for print-quality publication), DOCX.
• Publication workflow: Scholarly review stages, revision tracking, and publication approval.

Temporal-Geographic Mapping (@nisaba/geotemporal)#

• Geographic timeline: Visualizing where manuscripts originated, where they traveled, and where they are held now — mapping transmission history cartographically.
• Cartographic visualization: Map-based display of manuscript origins, transmission routes, and scribal centers (Alexandria, Qumran, Constantinople, Nalanda, Babylon, Nineveh, etc.).
• Fuzzy date support: BCE/CE dates with approximate confidence ranges for undated manuscripts (earliest, latest, probable center, confidence level).
• Provenance chain documentation: Recording the ownership and custody history of manuscripts — essential for both scholarly research and repatriation and legal questions.

AI Scholarly Assistant (@nisaba/assistant)#

The AI assistant grounds all responses in the Nisaba corpus, providing citation-backed scholarly support rather than hallucinated generalizations.

• Translation commentary generation: AI-generated scholarly commentary on translation choices, citing parallel passages, manuscript variants, and lexical evidence.
• Textual variant evaluation: Reasoning through which variant reading is more likely original, applying standard text-critical criteria (lectio difficilior, intrinsic probability, transcriptional probability).
• Research question answering: Grounded in corpus evidence with explicit citations to source texts.
• Translation alternative generation: Offering alternative renderings of a passage with philological justification for each option.

Import/Export Standards (@nisaba/standards)#

• Full TEI-XML round-trip: Import and export in TEI P5 and EpiDoc.
• ATF: Import and export for cuneiform corpora.
• USFM/USX: For Bible software integration.
• OpenITI mARkdown: For Arabic/Persian text ecosystems.
• Citation export: Zotero RDF, BibTeX, and Endnote XML for bibliographic management.

Scholarly Workspace (@nisaba/workspace)#

The workspace is the scholar's primary interface — a configurable research environment bringing together all of Nisaba's capabilities.

• Unified cross-corpus search: Lemma-aware (search for a word and find all its morphological forms); morphology-aware (search for all nouns in the genitive case); Boolean; proximity; field-restricted (search only in manuscripts from a given century) — searching across all connected corpora simultaneously.
• Configurable split-pane reading environment: Multiple panes showing texts, translations, manuscript images, or annotations simultaneously; synchronized scrolling; font selection; light and dark themes.
• Personal library: Reading lists, bookmarks, highlights, and notes; citation manager export to Zotero, Endnote, and BibTeX.
• Research project management: Project lifecycle management; collaborator roles; shared resources; milestone tracking.
• Keybinding configuration: Power-user keyboard shortcuts for common scholarly operations.

Client SDK (@nisaba/client)#

• Zero-dependency TypeScript SDK (NisabaClient, built via NisabaClientBuilder) using native fetch.
• Typed resource operations covering manuscripts and witnesses, collation and stemma, annotations, editions, canonical references, corpus search, comparative concept maps, and the AI assistant.
• Automatic retry with exponential backoff for transient network failures.
• Request/response interceptor pipeline for authentication, logging, and transformation.
• In-memory response cache with configurable TTL for frequently-accessed reference data.
• Pagination support: Both offset-based and cursor-based pagination.
• Batch requests and WebSocket channels: batchRequests runs multiple operations together; connectWebSocket opens collaboration channels.
• React bindings: @nisaba/client also ships a reactHooks module with CRDT collaboration primitives (vector clocks, LWW sets, operational transforms).

A separate @nisaba/api-client package provides the generated typed client for the canonical V1 contract API (/api/v1/nisaba).

Infrastructure (@nisaba/schemas, @nisaba/database)#

Validation Schemas (@nisaba/schemas)#

Zod schemas for the domain entities that cross trust boundaries — request bodies, corpus ingestion payloads, and data arriving from external APIs. Each module exports schemas, inferred TypeScript types, and Create* (and most an Update*) variants.

• Manuscript schemas (manuscript.ts): ManuscriptSchema with siglum, material, script type, dimensions, provenance fields, dating (dateEarliest/dateLatest/dateConfidence), iiifManifestUrl; ManuscriptDimensionsSchema, WitnessClassificationSchema, ManuscriptQuerySchema.
• Variant and collation schemas (variant.ts): VariantReadingSchema, VariantUnitSchema, CollationConfigSchema, CollationResultSchema, CollationProjectSchema.
• Critical apparatus schemas (apparatus.ts): ApparatusEntrySchema, ApparatusReadingSchema, ApparatusConfigSchema, CriticalEditionSchema.
• Annotation schemas (annotation.ts): AnnotationSchema, AnnotationLayerSchema, AnnotationExportSchema, and the W3C selector schemas (TextPositionSelectorSchema, TextQuoteSelectorSchema, etc.).
• Canonical reference schemas (reference.ts): CanonicalReferenceSchema with tradition, canon, bookId/bookName, structure, versification; VersificationMappingSchema, ReferenceRangeSchema, ReferenceQuerySchema.
• Cross-tradition schemas (tradition.ts): ConceptMappingSchema, MotifSchema, InfluenceEdgeSchema with similarity taxonomy and evidence grading.
• Corpus and paleography schemas (corpus.ts): CorpusMetadataSchema, PaleographicAnalysisSchema, PaleographicFeatureSchema, DateEstimationSchema, ScribalHandSchema.
• Geotemporal schemas (geotemporal.ts): FuzzyDateSchema, ProvenanceEventSchema, ProvenanceChainSchema, HistoricalLocationSchema, TimelineEventSchema, GeospatialQuerySchema.

Database (@nisaba/database)#

• PostgreSQL via Prisma: typed schema for all domain entities, generated to database/src/generated/client; the datasource URL is NISABA_DATABASE_URL.
• 16 models: Manuscript, CollationProject, VariantUnit, CriticalEdition, Annotation, AnnotationLayer, ConceptMapping, MotifAttestation, InfluenceEdge, CanonicalReference, Passage, PaleographicAnalysis, ScribalHand, ResearchProject, PersonalLibraryItem, Bookmark.
• Seed data: database/prisma/seed.ts.
• Migration management: Prisma migration workflow; the Prisma client is re-exported via getNisabaClient() / createNisabaClient() / disconnectNisabaClient().

Cross-Domain Boundaries (@nisaba/cross-domain)#

Nisaba owns all manuscript, corpus, and textual-scholarship capabilities. Other Oshun domains consume Nisaba research outputs through a well-defined bridge rather than duplicating these concerns. The boundary exists so that the scholarly text layer stays independent of UI frameworks, meditation content, or productivity workflows — each domain can evolve at its own pace without coupling.

The data that crosses the boundary:

• Tara (contemplative practice): Nisaba supplies passage companions for meditation sessions — a scripture passage, its variants, and its translation history. Tara owns the practice flow; Nisaba owns the text objects.
• Arete (virtue and growth): Nisaba supplies structured study plans and canonical references for Arete's goal-setting and learning-path features. Arete owns the user's growth narrative; Nisaba owns the text curriculum.
• Veritas (truth and epistemology): Nisaba supplies source-depth signals — grounding claims in primary textual evidence with citations. Veritas owns the epistemic evaluation; Nisaba owns the corpus evidence.
• Nyx (cosmology and symbol): Nisaba supplies cosmology overlays — linking symbolic concepts to their textual attestations. Nyx owns the symbolic representation; Nisaba owns the primary sources.
• Shared concept graph: Nisaba maintains the authoritative concept graph nodes and edges; other domains read-link to them without duplicating the scholarly validation logic.

The bridge is implemented in libs/nisaba/cross-domain/src/nisaba-cross-domain.ts and verified by verifyNisabaCrossDomainCoverage. See specifications.md §8 for the complete signal interfaces and capability enum.

Scope#

This feature document is scoped to the 22 @nisaba/* packages under libs/nisaba/ plus the canonical V1 contracts in libs/contracts/src/nisaba/. It covers ancient text analysis, corpora, canon management, editions, annotations, languages, paleography, philology, criticism, comparative research, geotemporal analysis, translations, standards, the workspace and client surfaces, the cross-domain bridge, the mobile and study-plan surfaces, and assistant integration. Nisaba ships as libraries and generated API contracts; there is no apps/nisaba or services/nisaba.

Other domains may consume Nisaba research outputs, but Nisaba's manuscript, corpus, and textual-scholarship features stay here.

Autonomous Research Contribution (Phase 178)#

Nisaba is a co-owner of the autonomous research / agentic-scientist substrate (Phase 178, centered in Nous). Nisaba's side is scholarly authoring and philological grounding: the agentic scientist uses Nisaba's critical-edition, citation, and cross-tradition analysis surfaces when a research task touches ancient or primary-source texts, and Nisaba consumes the shared literature and novelty graph. Nous owns the agent loop and experiment ledger; Nisaba owns text-critical correctness.