Patent Text Generation Using NLP and Templating
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current systems for generating patent application text using natural language models often produce inconsistent output that fails to follow desired forms and are limited in extracting and utilizing semantic information, while regular expressions restrict output flexibility.
Innovation Solution
The integration of natural language models with programmable templating languages, such as Handlebars, and helper functions to translate template text into a format suitable for word processing environments, enabling precise and multi-dimensional text generation by extracting grammatical and legal parts of patent claims and inserting them into a structured model data object.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of operation
If natural language models are used to generate patent application text, then text generation capability is improved, but output consistency and adherence to desired forms deteriorates
Solution Approach 1:
The system segments the text generation process into distinct components: a natural language model generates candidate text, while a separate programmable templating language component ensures structural consistency. The templating system divides the output into structured sections (abstract, claims, specification) with defined formats, allowing each component to specialize in its strength.
Solution Approach 2:
A programmable templating language acts as an intermediary between the natural language model and the final patent document. The template receives generated text and systematically formats it according to patent requirements, mediating between the model's creative output and the rigid structural requirements of patent forms.
2Measurement precision
If regular expressions are used for text extraction, then extraction precision is improved, but output flexibility deteriorates
Solution Approach 1:
The system changes the parameter of extraction from fixed regular expression patterns to a more flexible programmable templating approach. The template language allows dynamic parameter substitution and conditional logic, enabling the same extraction framework to adapt to different patent types, sections, and formatting requirements while maintaining precision through structured validation.
Solution Approach 2:
The programmable templating language serves multiple functions: it extracts structured information, validates content format, performs text generation, and ensures compliance with patent requirements. This universal system replaces multiple specialized tools (regex extractors, format validators, text generators) with a single multi-functional framework.
3Reliability
If static templates are used for text generation, then output consistency is improved, but adaptability to different patent formats deteriorates
Solution Approach 1:
The system transitions from static templates to dynamic programmable templates that can adapt to different patent formats. The templating language includes conditional logic and parameter substitution capabilities, allowing the same template framework to dynamically generate different patent document types (utility models, design patents, invention patents) while maintaining consistent structural quality.
Data Source
AI summary
In some implementations, a device may identify a text from a document that includes structured information, wherein the structured information is a set of patent claims. The device may extract a first one or more parts and a second one or more parts, wherein the first one or more parts are one or more grammatical parts and the second one or more parts are one or more legal parts. The device may insert the first one or more parts and the second one or more parts into a structured model data object, wherein the structured model data object is a claim model data object. The device may generate structured output text from the structured model data object, wherein the structured output text corresponds to one or more patent document parts. The device may provide the structured output text.