Argument Reasoning Detection Using Contextual Language Models
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing NLP systems struggle to accurately detect reasoning in arguments, particularly when explicit conjunctions are absent or equivocal, limiting their ability to understand implicit reasoning.
Innovation Solution
A machine learning classifier trained using a contextual language model to estimate the probability of reasoning in arguments, leveraging a training set of positively and negatively labeled examples generated by applying a masked language model to identify potential conjunctions and assign probabilities, followed by supervised learning.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional NLP methods are used to detect reasoning, then the system is simple and easy to implement, but the detection precision is low especially when explicit conjunctions are absent
Solution Approach 1:
The patent introduces an intermediary component - a contextual language model - that acts as a mediator between the input argument and the reasoning detection process. This language model provides contextual understanding and probability estimates that bridge the gap between simple keyword matching and accurate reasoning detection, enabling the system to handle implicit reasoning without requiring complex manual feature engineering
Solution Approach 2:
The system performs preliminary action by pre-training a machine learning classifier on a curated training set before deployment. This pre-training phase allows the system to learn patterns of reasoning from labeled examples, establishing a foundation of detected reasoning patterns that improves subsequent detection accuracy without increasing operational complexity
2Measurement precision
If a machine learning classifier with contextual language model is used, then the reasoning detection precision improves, but the training data requirement and system complexity increase
Solution Approach 1:
The patent applies partial action by focusing the training process on specific, critical features rather than attempting to process all possible linguistic elements. The system selectively processes arguments with equivocal conjunctions and those lacking explicit reasoning markers, concentrating computational resources on the most challenging cases where traditional methods fail, thereby achieving high precision without requiring exhaustive training data
3Productivity
If automated labeling is used to generate training set, then the labeling efficiency improves, but the reliability of labels may decrease
Solution Approach 1:
The system implements feedback by using the contextual language model to estimate probabilities and automatically label arguments, then using these automatically generated labels to train and improve the machine learning classifier. This closed-loop feedback mechanism allows the system to continuously refine its reasoning detection capabilities, with the classifier learning from its own predictions and improving label reliability over time while maintaining high productivity
Data Source
AI summary
Automated detection of reasoning in arguments. A training set is generated by: obtaining multiple arguments, each comprising one or more sentences provided as digital text; automatically estimating a probability that each of the arguments includes reasoning, wherein the estimating comprises applying a contextual language model to each of the arguments; automatically labeling as positive examples those of the arguments which have a relatively high probability to include reasoning; and automatically labeling as negative examples those of the arguments which have a relatively low probability to include reasoning. Based on the generated training set, a machine learning classifier is automatically trained to estimate a probability that a new argument includes reasoning. The trained machine learning classifier is applied to the new argument, to estimate a probability that the new argument includes reasoning.