Multi-word search

A multi-word search matches a sequence of consecutive tokens. You do not select it: typing a space in the search box is what turns a word search into this one (see Search mode).

Because the corpus is stored token by token (tokenization process), each position in the sequence can carry its own pattern. That is what makes the feature useful for the things cuneiform scholarship actually looks for: formulae, syntactic constructions, and multi-word units whose meaning is not the sum of their parts.

Writing the query

Words are separated by a space, and each of them may carry its own regular expression. This lets you be strict where you know the form and permissive where you do not.

Plain sequences:

• a-na pa-ni (the prepositional phrase "before, in the presence of")
• ab-ba-nu DUMU IM-GAL (the personal name Abbanu with patronymic "son of Baʿlu-kabar")

With patterns (requires Regex search):

• ab-ba-n. DUMU .* retrieves any individual named Abbanu (with orthographic variants such as ab-ba-nu, ab-ba-ni) followed by the filiation marker DUMU and any patronymic (the .* pattern matches any character sequence)
• ^a-na pa-n. matches the syntagma beginning with a-na followed by any form of panum (e.g., pa-ni, pa-na, pa-nu)
• i-na .*-ti identifies prepositional phrases with ina followed by any noun in the genitive case (marked by the -ti ending)

Specifying the pattern position by position is what lets you balance precision and recall within one query: orthographic variation tolerated in the name, an exact match required for the patronymic — or the reverse. In the same way, a query for a syntactic construction can fix the prepositions and conjunctions while leaving the nominal or verbal slots open.

Reading the results

Results are aggregated and sorted alphabetically. Unlike the word search, they are organised by the complete sequence rather than by its individual terms. The Additional groupers control adds further criteria — genre, site, scribal school — which is the quickest way to see whether a formula belongs to one archive or runs across the corpus.

Attestations are reported at document level: the result lists the tablets in which the sequence occurs, not the individual lines as the word search does. This follows from what multi-word searches are usually for — locating the documents that share a formula, then reading them.

Clicking a siglum opens the text with every term of the sequence highlighted in yellow.

What it is good for

• Formulae — recurring administrative and juridical phrasing (a-na pa-ni PN, i-na ITI ŠE for dates in contracts).
• Prosopography — individuals identified by patronymic or title (PN DUMU PN₂, PN LÚ GN).
• Syntax — corpus-wide evidence for a construction: prepositional phrases, verbal chains, genitive constructions.
• Orthographic variation — how one multi-word expression is written across scribal traditions or periods.