Conversational AI Rule Subsumption via Directed Acyclic Graph
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing rule-based systems for dialog management in bots require reordering or rewriting existing rules when adding new rules, making them brittle and difficult to modify, and struggle to integrate with machine learning effectively.
Innovation Solution
The implementation of subsumption logic to arrange rules into a directed acyclic graph (DAG) based on specificity, allowing more specific rules to override general ones, enabling the addition of new rules without modifying existing ones and facilitating integration with machine learning.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If traditional rule-based systems are used for dialog management, then the system can handle conversational tasks, but adding new rules requires reordering or rewriting existing rules making the system brittle and difficult to modify
Solution Approach 1:
The patent segments the rule system into a hierarchical structure with general rules at higher levels and specific rules at lower levels. This segmentation allows new specific rules to be added without affecting general rules, eliminating the need to reorder or rewrite existing rules when adding new functionality.
Solution Approach 2:
The patent introduces a hierarchical dimension to the rule system, organizing rules across multiple levels of generality. This dimensional transformation converts a flat, linear rule structure into a multi-level hierarchy, enabling independent addition of rules at any level without disrupting the overall system structure.
2Reliability
If more rules are added to handle specific conversational scenarios, then the system becomes more robust, but the difficulty of managing and reasoning over the whole set of rules increases
Solution Approach 1:
By segmenting rules into hierarchical levels, the patent reduces the cognitive load on developers. Instead of reasoning over the entire flat rule set, developers only need to consider rules at their specific level and how they interact with parent and child rules, making rule management more manageable as the system scales.
Solution Approach 2:
The hierarchical structure acts as an intermediary layer between general rules and specific rules. This intermediate hierarchy provides a structured framework that automatically manages rule interactions, reducing the need for developers to manually reason about complex rule conflicts and interactions.
3Adaptability or versatility
If rule-based systems are used for dialog management, then conversational tasks can be handled, but integration with machine learning tools is difficult
Solution Approach 1:
The hierarchical rule structure serves multiple functions: it maintains traditional rule-based dialog management while providing a structured interface for integrating machine learning models. The hierarchy can accommodate both deterministic rules and probabilistic ML predictions at different levels, creating a universal framework that handles both approaches.
4Ease of manufacture
If traditional flat rule structures are used, then implementation is simple, but the system becomes brittle when modifications are needed
Solution Approach 1:
The hierarchical segmentation provides structural stability during modifications. By organizing rules into independent hierarchical levels with defined parent-child relationships, the system maintains stability when new rules are added or existing rules are modified, as changes at one level do not require restructuring the entire rule set.
Data Source
AI summary
Techniques are provided for building a dialog-state specific contextual language understanding system using subsumption logic. Information establishing conversational rules identifying the conversational dialog is received to present in respective dialog states. Each rule has a Boolean trigger expression of predicates for testing the conversational state together with logical connectives to identify when the rule is applicable. Subsumption logic is used to arrange the rules into a directed acyclic graph (DAG) where more specific rules are preferred to more general rules. During a conversation, the DAG is used to filter the triggered rules to only the most specific triggered rules from which a rule to run is selected. This structure makes it easier to build conversational systems because rules can be added or removed without having to change or reason over other rules. The rules also act as a constraint to help machine learned selection systems converge with less data.