Reference Data Model Lexical

Lexical Resources — Layer 4

Layer 4 is the dictionary: the canonical inventory of signs and the lexical database of lemmas and senses.

Key tables

Signs (from OGSL)

• signs — 3,367 cuneiform signs. Fields: Unicode codepoint, canonical name, sign type (simple, compound, modified), GDL structural definition.
• sign_values — ~15,000 reading values for signs, with sub-indices. A single sign like KA can have readings: ka, gu, dug, inim, zu.
• sign_variants — Variant forms and glyph families.

Lexical entries (from ePSD2 and ORACC glossaries)

• lexical_lemmas — 61,435 dictionary headwords. Fields: citation form (cf), guide word (gw), part of speech (pos), language code, source (epsd2, ogsl, oracc), attestation count.
• lexical_senses — 155,491 meaning distinctions per lemma. One lemma can have multiple senses with different glosses and usage contexts.
• glossary_entries — ORACC glossary entries for specific projects.
• glossary_forms — Variant written forms attested in the corpus for a glossary entry.
• glossary_senses — Sense data from ORACC project glossaries.

OGSL as sign authority

The ORACC Global Sign List is Glintstone's canonical sign inventory. OGSL assigns internal sign IDs that serve as the primary key in signs. MZL (Borger's Mesopotamisches Zeichenlexikon) numbers are mapped via Unicode codepoints for cross-reference with CompVis and other systems.

~200–400 CompVis MZL labels remain unresolved after automatic matching — these are sign annotations without a confirmed OGSL mapping.

ePSD2 and ORACC glossaries

ePSD2 is the primary Sumerian lexicon (1.8 GB of dictionary data). It is imported as an ORACC project (epsd2) and provides comprehensive Sumerian entries with attestation counts. ORACC project glossaries (gloss-sux.json, gloss-akk.json, and others) provide lemma data for specific corpora.

The lexical_lemmas table merges entries across sources. Duplicate detection uses citation form + guide word + language code as the composite key.

The normalization bridge

The path from a written form in ATF to a dictionary entry:

written_form (e.g., "lu₂-gal")
  → normalized_form (e.g., "lugal")
    → lexical_lemma (cf="lugal", gw="king", pos="N", lang="sux")
      → lexical_senses (gw="king", sense="ruler", ...)

This bridge is implemented as the dictionary/lookup API endpoint and powers the interpret_token MCP tool. Candidates are ranked by attestation frequency.

Subscript handling differs by language. In Sumerian, du₃ ("to build") is a different word from du ("to go") — subscripts carry semantic content. In Akkadian, subscripts are sign disambiguation only — du₃ normalizes to du.

Language coverage

Source	Languages covered
ePSD2	Sumerian (sux)
ORACC glossaries	sux, akk, akk-x-stdbab, akk-x-oldbab, qpn (proper nouns)
OGSL	Sign level only (language-agnostic)

Elamite and Hittite lexical coverage is minimal.