Discourse Tree Analysis for Missing Text Fragments
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current Natural Language Inference systems are limited by datasets that represent only a fraction of human reasoning possibilities, leading to poor performance on datasets not specifically intended for them, and struggle to identify missing natural language expressions in input text.
Innovation Solution
The method involves generating a discourse tree from input text to identify rhetorical relationships, forming queries based on these relationships, and using a knowledge base to find candidate sentence fragments that match predefined rhetorical relations, thereby identifying potentially missing expressions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Speed
If existing NLI systems are used, then processing speed is improved, but the ability to identify missing information deteriorates
Solution Approach 1:
The patent introduces an intermediary component (the missing information detection module) that bridges the gap between existing NLI systems and comprehensive information analysis. This module acts as a mediator that processes the output of NLI systems and identifies missing information that would otherwise be lost, resolving the contradiction between processing speed and information completeness.
Solution Approach 2:
The system segments the information processing task into distinct components: the NLI system handles explicit information matching while the added module separately handles implicit information detection. This segmentation allows each component to specialize, maintaining the speed advantage of NLI systems while adding the capability to identify missing information through rhetorical relationship analysis.
2Measurement precision
If datasets are expanded to cover more human reasoning possibilities, then identification accuracy is improved, but system complexity deteriorates
Solution Approach 1:
The patent implements a universal rhetorical relationship framework that can handle multiple types of missing information scenarios through a single unified approach. Rather than creating separate systems for different reasoning types, the framework provides multi-functional capability to detect various forms of implicit information (causal relationships, presuppositions, entailments, etc.) using consistent methods, thereby improving accuracy without proportionally increasing complexity.
Solution Approach 2:
The system uses discourse tree structures that replicate the hierarchical organization of human reasoning in a computationally manageable form. By copying the structural patterns of natural language discourse into tree representations, the system captures complex reasoning relationships without requiring exponentially increased computational resources, maintaining a favorable complexity-accuracy balance.
3Measurement precision
If discourse analysis is performed to identify rhetorical relationships, then comprehension accuracy is improved, but processing time deteriorates
Solution Approach 1:
The system performs preliminary discourse parsing to identify rhetorical relationships before conducting full information analysis. By pre-processing the text to establish the discourse structure and rhetorical relationships in advance, the system prepares the groundwork for accurate comprehension while enabling more efficient subsequent processing, thus reducing overall processing time despite the added analytical depth.
Solution Approach 2:
The discourse analysis is applied selectively to local portions of text where rhetorical relationships are most likely to contain missing information, rather than uniformly analyzing the entire text. This localized approach concentrates computational resources on high-value areas, improving comprehension accuracy where it matters most while minimizing unnecessary processing time expenditure on less critical sections.
Data Source
AI summary
Techniques for identifying one or more missing fragments within input text are disclosed. A discourse tree (DT) is generated for the input text (IT) received, the IT having any suitable number of sentence fragments. An indication that the IT is likely missing one or more sentence fragments may be identified based on determining that one or more rhetorical relationships of the DT matches one of a set of predefined rhetorical relationships. A query is generated one or more sentence fragments of the IT and executed against a knowledge base to obtain a set of search results. A most-relevance search result can be utilized to identify a set of candidate sentence fragments. A subset of those candidate sentence fragments can be identified based on comparing them to the sentence fragments provided in the IT, each candidate sentence fragment of the subset being implied but excluded from the IT.