(Tentative)
| Definition | Functional Requirements | Syntax | Semantics |
Definition
Intent is a structured semantic directive inferred from User input. It captures both the high-level domain of the request and its specific class, along with contextual modifiers such as urgency, certainty, and spatial/temporal cues.
Functional Requirements
- Semantic Classification
- Must classify user input into a two-level hierarchy:
- Domain: High-level category (e.g., Informational, Procedural)
- Class: Specific intent type within the domain
- Must enforce domain-class consistency via conditional validation logic.
- Execution Modifiers
- Must support optional parameters that refine how the intent is interpreted or executed:
- Urgency: Priority of execution
- ConfidenceLevel: Degree of user certainty
- TemporalCue: Time-related context (e.g., “tomorrow”, “now”)
- SpatialAnchor: Physical or virtual reference point (e.g., “that object”, “left corner”)
- Confidence Scoring
- Must include a numeric confidence score representing the system’s certainty in the intent classification.
- Must support downstream filtering or fallback strategies based on this score.
- Traceability
- Must include a Trace object with:
- Origin: The module or subsystem that inferred the intent
- Timestamp: Time of intent creation (conforming to Time.json)
- Uniqueness and Referencing
- Must include a globally unique IntentID for runtime referencing, logging, and chaining.
Syntax
https://schemas.mpai.community/PGM1/V1.0/data/Intent.json
Semantics
| Label | Description |
| Header | Intent Header |
| – Standard-Intent | The characters “PGP-INT-V” |
| – Version | Major version – 1 or 2 characters |
| – Dot-separator | The character “.” separating version components |
| – Subversion | Minor version – 1 or 2 characters |
| IntentID | Unique identifier for this intent instance |
| Intent | Unified semantic intent object |
| – Domain | High-level category of intent |
| —— Informational | Seeking facts, summaries, or explanations |
| —— Procedural | Requesting actions or processes |
| —— Confirmational | Affirming, denying, or clarifying |
| —— Expressive | Sharing emotion or opinion |
| —— Spatial | Interacting physically or visually |
| —— Meta-Interaction | Engaging with the system itself |
| – Class | Specific intent type within the domain |
| —— ask_fact | “What’s the capital of Norway?” |
| —— trigger_Behaviour | “Start the cleaning routine.” |
| —— confirm_action | “Yes, go ahead.” |
| —— express_emotion | “I’m frustrated with this.” |
| —— move_object | [User points and says “Put this here.”] |
| —— request_help | “How do I use this feature?” |
| Parameters | Optional modifiers refining execution |
| – Urgency | Execution urgency (low, medium, high) |
| – ConfidenceLevel | User certainty (tentative, neutral, assertive) |
| – TemporalCue | Time-related signal (e.g. “now”, “later”) |
| – SpatialAnchor | Physical reference (e.g. “zone_3”, “object_A”) |
| ConfidenceScore | Confidence in intent classification |
| Trace | Provenance metadata |
| – Origin | Module or subsystem that generated the intent |
| – Timestamp | Time the intent was created |