How this data is prepared and published

The TITUS corpus and its permitted use

Direct permission is recorded for storage, processing, and non-commercial public presentation of the TITUS Old Persian textual corpus in the Parsiandej reference. Parsiandej is not an official TITUS product and is responsible for its processing, labels, and presentation.

Source project names and links remain attached to related results. Public bulk downloads and programmatic third-party access are not provided; the complete machine-readable package remains private for academic review.

Source text is not the same as added data

1. Source text: cuneiform, transliteration, transcription, inscription identifiers, and line locations are kept without semantic rewriting.
2. Technical normalization: spacing, dividers, and searchable forms are stored separately and never replace the source text.
3. Tokenization: words, logograms, numerals, dividers, and editorial markers are distinguished.
4. Dictionary links: automatic, inferred, and reviewed matches retain their own method and confidence.
5. Generated English: line glosses, catalog descriptions, evidence notes, editorial notes, citation notes, and image descriptions are explicitly labelled Machine-generated and unreviewed.
6. Sign table: only assigned Unicode Old Persian characters are treated as valid script characters.

Generated English is not a reviewed critical translation

English content generated from the active authorized corpus snapshot is stored in an offline, version-bound package with provenance “machine-generated”, review status “generated”, a source-text fingerprint, and the generator batch. The visible label “Machine-generated and unreviewed” remains attached until a human-reviewed replacement is explicitly approved.

Generated line glosses and supporting prose help discovery and reading but do not claim a human-reviewed interpretation of syntax, grammar, history, or literary style. Source cuneiform, identifiers, scholarly references, and official Unicode names are not translated or relabelled as generated content.

Damage, reconstruction, and special marks

Damage, editorial reconstruction, and text completeness are not conveyed by colour alone: each published line includes a textual status. Numerals, logograms, and dividers retain their sign type.

Exact matches before phonetic approximation

The converter searches complete forms in published data first. A close suggestion is not the same as a documented attestation, and a phonetic approximation is not historical evidence or a translation.