SSyntha
Pilotjsonlparquetcsv

Arabic Per-Turn Intent

11,000 retail-support conversations with an intent label on every customer turn.

11,000
rows
6
dimensions
3
formats
22
checks

Overview

Built for Arabic intent classification, including the context-versus-single-turn comparison. The label is drawn before generation and written onto the turn afterwards — the model is asked to express a label, never to assign one — so the corpus cannot quietly become a function of the validator's own regexes.

conversations
11,000
labelled turns
44,000
intent labels
15
retail verticals
5
registers
5
format
JSONL · Parquet

Row specimen

jsonl
{
  "id":           "uuid",
  "metadata":     { "vertical": "perfume", "register": "najdi" },
  "conversation": [
    { "role": "user",  "content": "...", "label": "price_inquiry",
      "label_subtype": null, "context_dependent": false },
    { "role": "agent", "content": "..." }
  ]
}

Field schema

Fields as they appear in every row of the file.

idstringStable row identifier
metadataobjectRetail vertical and register
conversationarrayTurns, each customer turn carrying its label
labelenumOne of 15 intents, drawn before generation
context_dependentboolTrue when the turn's meaning needs prior context

Inspection report

22 / 22
✓Turn count bounds✓Role alternation✓Minimum turn length✓No adjacent duplicates✓Every turn has content✓Levantine contamination✓Robotic support phrasing✓English inside dialogue✓Caricatured dialect✓Injected amount present✓Reference number present✓Brand named by agent✓Agent introduces itself✓Agent name not pre-known✓Sector vocabulary✓Sector-fit resolution✓Dialect hospitality marker✓Forbidden dialect slang✓Brand capability match✓Verification before resolution✓Dialect marker frequency✓Hospitality phrase repetition

Other datasets

50,000jsonlSaudi Arabic Customer Support$29940,000jsonlGulf Dialect SupportIn production15,000jsonlArabic Tool-Use TracesIn production