Indexes

The indexes are the registers of the corpus: the lexicon, the onomasticon, the prosopography, the seal catalogue. Each of them is a list of entities, and each entity gathers its own attestations across the texts.

They are not compiled separately from the edition. An index entry acquires an attestation when an editor assigns a token to it; the list is a view over those assignments, never a document to be maintained by hand. This has a practical consequence worth stating at the outset: you do not add a word to the lexicon by editing the lexicon, you add it by lemmatising an occurrence.

1. Headword or named entity? A distinction that decides everything

The corpus stores every written word as a token. What an index registers is a type: the abstract entity of which the token is an instance. Two questions look alike at this point, and they are not the same question:

Of which word is this form an instance? — and — Which name does this spelling render?

The first is answered by lemmatisation, the second by named-entity recognition. In computational-linguistic terms they belong to different tasks, and DAPCA keeps them on separate planes because conflating them corrupts both.

1.1 The dictionary headword

A headword — a lemma — is the citation form of a lexeme: the class of all inflected forms that share one lexical meaning. i-ša-am, i-ša-am-mu and ta-aš-ta-ma are realisations of the lexeme šâmu "to buy".

Assigning a lemma is a paradigmatic statement. It says: this form belongs to that paradigm, and therefore to that dictionary meaning. The entry has a morphology, a root, a language, a guideword; it has a place in a lexicon that could be printed as such.

1.2 The named entity

A named entity is not a lexeme. It is a referring expression. pil₂-su-DINGIR-KUR does not mean anything in the way šâmu means "to buy" — it designates.

Normalising it as Pilsu-Dagān states that this spelling renders that name. It says nothing about who bore the name, and it does not place the form in an inflectional paradigm: it places it in an onomasticon. The four classes DAPCA uses — personal, divine, geographic, month name — are types of referent, the Assyriological counterpart of the categories of a named-entity recogniser.

1.3 Why the two planes never meet

A token — in a specific context — belongs to one plane or the other, never to both, and the plane is decided by the token's class, which is fixed in the transliteration by the marker prefixes:

A named token can only receive a name form; a lexical token can only receive a lemma. This is enforced, not merely recommended: an assignment that would cross the line is refused, or detached with a warning.

1.4 And a third act: identifying the individual

Normalising a name is a judgement about writing and language. Deciding that this attestation of Pilsu-Dagān and that one, in another cuneiform tablet, are the same man is a judgement about history.

Two men called Pilsu-Dagān share one name form and are two distinct persons; one man whose name is spelled three ways has three spellings and one identity. Keeping the acts apart is what allows you to be certain about a reading while remaining undecided about its bearer. See §3.

2. The four registers

Genealogy is not a fifth register but a layer derived from the third (Historical Persons), and the Global lemmatizer is the tool that feeds the first two (Lexicon and Named entities).

3. A historical person is not a marked filiation

Two operations look adjacent and are conceptually distinct. One is done on a tablet, in Mark genealogies; the other maintains a register.

Mark genealogies records evidence; a historical person is an interpretation. A tablet may state a filiation between two men neither of whom has been identified — the statement stands, and remains useful, without any person existing.

How evidence becomes a family tree is described in Genealogy. It is not automatic, and the step that is usually missing is not the marking of more filiations.

4. Adding entries and linking them to what exists

The single most important thing to understand about editorial work here:

The lexicon is the exception, and the reason is worth stating. The other three registers describe material that only this project holds: the names as this corpus spells them, the individuals who appear in it, the seals impressed on its tablets. Nobody else can supply them. An Akkadian dictionary is not in that position — it exists, it is maintained by projects devoted to it, and DAPCA aligns to it rather than duplicating it.

4.1 The three ways a link comes into being

1. At import, automatically. Loading a text can re-attach its tokens to entries that already exist, each within its own plane, never creating any (Adding a new tablet). It is a starting point, not a result: where a spelling points to several entries the importer takes the most frequently attested one and flags the word for review.
2. In bulk, in the lemmatizer. You select the occurrences that share one interpretation and assign the same entry to all of them at once (/lemmatizer/).
3. One at a time, from the entity's own page or from the tablet — a prosopographic attestation, a seal attestation. A user — with the correct role (administrator or advanced user) — can create and/or edit the link between a token and its normalized form from the /tablet/<id>/view page (/tablet/5/view) after clicking on the single transliterated word.

4.2 Nothing propagates backwards

The registers record editorial statements. They are revised by editors, not recomputed.

5. Who sees what