Semantic Parser Adaptation Using Unlabeled Program Scripts

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Existing semantic parsing technologies struggle with adapting to new programming language domains due to the high cost and impracticality of providing large amounts of labeled training data, especially when interacting with third-party tools like SQL databases.

Innovation Solution

A method using unlabeled programmatic scripts and inverse semantic parsing to generate training data, with optional seed labeled data, to train a semantic parser for new language domains, reducing human involvement and resource intensity.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Reliability

If large amounts of labeled training data are provided to train a semantic parser for new programming language domains, then the accuracy and reliability of the semantic parser improves, but the cost, time, and human involvement required increases significantly

Engineering Contradiction:

Improvesemantic parser accuracyVSAvoidtraining data preparation time

Core Design Contradiction:

ReliabilityVSLoss of time

Solution Approach 1:

The system performs preliminary actions by using an inverse semantic parser to generate candidate natural language descriptions from unlabeled programmatic scripts before actual training occurs. This pre-generation of training data candidates eliminates the need for manual labeling during the training phase, significantly reducing the time and human resources required while maintaining parser accuracy

Inventive Principle:

Principle #10Preliminary action

Solution Approach 2:

The system enables self-service by allowing the semantic parser to automatically generate and validate its own training data through the inverse parsing process. The validation mechanism automatically determines which generated training examples are high-quality, eliminating the need for human reviewers to manually curate training datasets

Inventive Principle:

Principle #25Self-service

2Manufacturing precision

If manual labeling of programmatic scripts is performed to create training data, then the quality and precision of training data improves, but the ease of manufacture and resource intensity worsens

Engineering Contradiction:

Improvetraining data qualityVSAvoidtraining data generation ease

Core Design Contradiction:

Manufacturing precisionVSEase of manufacture

Solution Approach 1:

The inverse semantic parser serves as an intermediary that automatically generates candidate natural language descriptions from programmatic scripts. This intermediary process produces high-quality training data candidates that can be automatically validated, eliminating the need for manual labeling while maintaining data quality through the structured generation and validation process

Inventive Principle:

Principle #24Intermediary (Mediator)

Solution Approach 2:

The system implements feedback through a validation process that automatically evaluates the quality of generated training examples. High-quality examples are selected and added to the training set, creating a feedback loop that continuously improves training data quality without requiring manual intervention, thus easing the manufacturing process while maintaining precision

Inventive Principle:

Principle #23Feedback

3Productivity

If a semantic parser is trained on existing programming language domains, then the productivity for those domains improves, but the adaptability to new language domains worsens due to lack of labeled data

Engineering Contradiction:

Improveparsing speedVSAvoidlanguage domain adaptability

Core Design Contradiction:

ProductivityVSAdaptability or versatility

Solution Approach 1:

The system performs preliminary actions by generating training data candidates from unlabeled scripts in new programming language domains before actual training occurs. This allows the semantic parser to be quickly adapted to new domains without requiring manual labeling, maintaining both productivity through automated processes and adaptability to new language domains

Inventive Principle:

Principle #10Preliminary action

Solution Approach 2:

The system achieves universality by creating a training data generation mechanism that works across multiple programming language domains without domain-specific customization. The inverse semantic parser and validation process are domain-agnostic, allowing the same system to adapt to any programming language domain, thereby improving both versatility and maintaining productivity through automation

Inventive Principle:

Principle #6Universality (Multi-functionality)

Data Source

PatentUS12511489B2Semantic parsing with pre-trained language models

Publication Date: 2025.12.30 MICROSOFT TECHNOLOGY LICENSING LLC

AI summary

Implementations of semantic parsing using pre-trained language models are provided. One aspect includes a computing system for semantic parsing of natural language. The computing system comprises processing circuitry and memory containing instructions that, when executed, cause the processing circuitry to receive a request comprising a natural language utterance and generate a formal meaning representation using the natural language utterance and a language model comprising a semantic parser that has been prompted with training data generated by providing a dataset comprising a set of unlabeled programmatic scripts and a seed programmatic script, generating a set of parsed natural language descriptions by inputting the set of unlabeled programmatic scripts into an inverse semantic parser, generating a set of re-parsed programmatic scripts by inputting the set of parsed natural language descriptions into the semantic parser, and determining a set of labeled programmatic scripts by validating the set of re-parsed programmatic scripts.