Hierarchical Spoken Language Understanding via Clause Segmentation

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Spoken language understanding in human-computer dialog systems is hindered by grammatically incorrect utterances and unpredictable errors in speech recognition, leading to inefficiencies in semantic analysis and scalability issues, particularly in tasks requiring a large number of classes.

Innovation Solution

A hierarchical method that partitions speech recognizer output into self-contained clauses, identifies and qualifies dialog acts, and recursively determines domain objects and actions, using a combination of classifiers and domain-independent/domain-dependent modules to generate a detailed semantic representation.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Device complexity

If a single classifier is used to classify input directly into actions, then the system is simple to implement, but it does not scale well for tasks requiring a very large number of classes

Engineering Contradiction:

Improvesystem implementation simplicityVSAvoidscalability to large number of classes

Core Design Contradiction:

Device complexityVSAdaptability or versatility

Solution Approach 1:

The patent segments the classification task into multiple hierarchical levels. Instead of using a single classifier to handle all classes, the system divides the classification into domain-independent and domain-dependent stages, with further segmentation into dialog act classification and domain object/action identification. This hierarchical segmentation allows the system to scale to large numbers of classes while maintaining manageable complexity at each level.

Inventive Principle:

Principle #1Segmentation

2Ease of operation

If semantic analysis is based on syntactic structures of the language, then the analysis is systematic, but it fails when utterances are grammatically incorrect or contain unpredictable errors

Engineering Contradiction:

Improvesystematic analysis approachVSAvoidaccuracy with grammatically incorrect utterances

Core Design Contradiction:

Ease of operationVSReliability

Solution Approach 1:

The patent introduces an intermediary layer between speech recognition and semantic analysis. The domain-independent module acts as a mediator that processes recognized text through dialog act classification before passing to domain-specific analysis. This intermediary approach allows the system to handle grammatically incorrect utterances by focusing on functional dialog acts rather than relying solely on syntactic correctness.

Inventive Principle:

Principle #24Intermediary (Mediator)

Solution Approach 2:

The patent changes the parameter of analysis from syntactic structure to dialog act classification. Instead of relying on grammatical correctness, the system classifies utterances based on their functional meaning (dialog acts such as request, inform, question, etc.). This parameter change enables reliable semantic analysis even when the input contains grammatical errors or unpredictable variations.

Inventive Principle:

Principle #35Parameter changes

3Adaptability or versatility

If a hierarchical method with multiple classification levels is used, then scalability to large numbers of classes is improved, but the device complexity increases

Engineering Contradiction:

Improvescalability to large number of classesVSAvoidhierarchical module structure

Core Design Contradiction:

Adaptability or versatilityVSDevice complexity

Solution Approach 1:

The patent implements universality by creating a domain-independent module that handles general dialog act classification applicable across multiple domains. This universal layer processes common linguistic patterns and dialog acts that recur across different application areas, reducing the need for domain-specific customization and managing overall system complexity while maintaining scalability.

Inventive Principle:

Principle #6Universality (Multi-functionality)

Data Source

PatentUS8190436B2System and method of spoken language understanding in human computer dialogs

Publication Date: 2012.05.29 MICROSOFT TECHNOLOGY LICENSING LLC

AI summary

A system and method are disclosed that improve automatic speech recognition in a spoken dialog system. The method comprises partitioning speech recognizer output into self-contained clauses, identifying a dialog act in each of the self-contained clauses, qualifying dialog acts by identifying a current domain object and/or a current domain action, and determining whether further qualification is possible for the current domain object and/or current domain action. If further qualification is possible, then the method comprises identifying another domain action and/or another domain object associated with the current domain object and/or current domain action, reassigning the another domain action and/or another domain object as the current domain action and/or current domain object and then recursively qualifying the new current domain action and/or current object. This process continues until nothing is left to qualify.