Semantic Drift Classification

knowledge skill

Classify semantic drift in a given text pair by identifying the type of conceptual change between an original and a modified statement. The agent produces a drift classification report. This capability is distinct in its focus on conceptual change types.

Worked examples

• {"input": {"original_text": "The new policy affects all employees.", "modified_text": "The new policy now affects all employees and contractors.", "context_domain": "HR Policy"}, "output": {"drift_type": "Extension", "confidence_score": 0.92, "key_difference": "Addition of 'contractors' to affected parties"}}
• {"input": {"original_text": "Smoking is allowed in the building.", "modified_text": "Smoking is strictly prohibited in the building.", "context_domain": "Facility Rules"}, "output": {"drift_type": "Contradiction", "confidence_score": 0.99, "key_difference": "Reversal from allowed to prohibited"}}

Input

• original_text: string, Original textual statement
• modified_text: string, Textually modified statement
• context_domain: string, Domain of discourse (e.g., medicine, law)

Output

• drift_type: enum[No Drift, Extension, Shift, Overlap, Contradiction], Type of semantic drift detected
• confidence_score: float[0-1], Agent's confidence in drift classification
• key_difference: string, Brief summary of the drift's nature

Details

• Skill type: knowledge skill
• Safety level: safe_public_research
• Version: 1.0.0