Dialogue Discourse Trees for Rhetorical Flow in Chatbots
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current dialogue management systems, particularly chatbots, face challenges in maintaining proper rhetorical flow and logical consistency in multi-turn conversations, as they struggle to select responses that are both topically relevant and rhetorically cohesive, often failing to acknowledge prior turns and introducing new information inconsistently.
Innovation Solution
The use of Dialogue Discourse Trees (DDTs) is proposed to represent dialogue and manage rhetorical flow by generating trees based on historical utterances, allowing for the classification of candidate responses using machine-learning models to ensure proper rhetorical agreement and logical consistency.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If existing dialogue management solutions are used, then simple queries and replies can be supported, but the system fails to maintain proper dialogue logic and rhetorical flow in multi-turn conversations
Solution Approach 1:
The patent segments dialogue management into two distinct components: a discourse representation module that maintains rhetorical structure (using Rhetorical Structure Theory to create discourse trees) and a response selection module that operates on segmented candidate utterances. This segmentation allows the system to handle complex multi-turn conversations while maintaining logical consistency by processing different aspects of dialogue management separately.
Solution Approach 2:
The patent introduces a discourse representation module as an intermediary between the input dialogue history and the response selection process. This intermediary maintains a discourse tree that captures rhetorical relationships, serving as a mediator that ensures logical consistency is preserved while enabling versatile handling of multi-turn conversations through structured representation.
2Loss of information
If the system selects from topically suitable candidate responses, then response relevance is improved, but rhetorical coherence and dialogue logic are lost
Solution Approach 1:
The patent performs preliminary action by constructing a discourse tree representation of the dialogue history before the response selection process. This pre-established rhetorical structure serves as a foundation that guides subsequent response selection, ensuring that coherence is maintained without requiring exhaustive analysis during the selection phase, thus preserving both rhetorical coherence and selection efficiency.
Solution Approach 2:
The patent replaces the mechanical filtering process with a rhetoric-aware selection mechanism. Instead of simply filtering candidates based on topical relevance, the system uses the discourse tree structure to evaluate rhetorical coherence, substituting a more sophisticated but efficient mechanism that leverages pre-computed rhetorical relationships to guide response selection.
3Reliability
If the system acknowledges prior turns consistently, then dialogue logic is improved, but the complexity of response selection increases
Solution Approach 1:
The patent segments the complex response selection task into distinct phases: discourse analysis (building the tree), candidate generation (topical filtering), and rhetoric-aware selection (using the tree structure). This segmentation reduces overall system complexity by breaking down the monolithic selection process into manageable, specialized components that can be implemented independently.
Solution Approach 2:
The discourse tree serves as an intermediary data structure that captures rhetorical relationships without requiring the response selection algorithm to directly analyze complex dialogue history. This intermediary simplifies the selection process by providing a pre-processed, structured representation that encodes rhetorical agreement requirements, reducing the complexity of the selection system while maintaining reliability.
Data Source
AI summary
Disclosed systems, devices, and methods improve dialogue management through the use of dialogue discourse trees (DDTs). To determine whether a candidate response (CR) is appropriate in the dialogue, the CR can be added to other utterances that have been previously provided in a dialogue between two parties and a dialogue discourse tree (DDT) can be generated from the result. The DDT includes nodes corresponding to elementary discourse units (EDUs) representing text fragments of the utterances and the CR. The DDT may include nodes that indicate rhetorical relationships between EDUs. In some embodiments, the DDT comprises a node representing at least one dialogue-specific rhetorical relationship between two utterances. The DDT for the CR can be provided to a machine-learning model that has been trained to identify whether the rhetorical flow is maintained between the utterances of the DDT. If so, the CR can be provided in response to the request.