Clause Detection

Detects and extracts clauses in natural language documents using semantic similarity.

Remarks

The Clause Detection class enables Grooper to locate and extract sections of a document that match a provided sample clause, even when the wording varies. This is accomplished using semantic similarity, leveraging vector embeddings to compare the sample text to passages in the document.

How It Works

• The user provides a sample clause and configures semantic matching options using the 'Detection Settings' property.
• During extraction, the document is split into overlapping segments, and each segment is compared to the sample using a language model's embedding.
• The most semantically similar segments are selected and extracted as section instances.

This approach is especially useful for:

• Locating boilerplate or legally significant language (e.g., confidentiality, indemnity, or termination clauses)
• Comparing contracts or policies for the presence and consistency of required language
• Extracting policy or regulatory statements that may be phrased differently across documents

Configuration Guidance

• Use the 'Detection Settings' property to define the sample clause and adjust semantic matching parameters.
• Select an embedding model that is appropriate for your document language and domain.
• Test with representative documents to ensure reliable detection, especially if clause wording varies significantly.

LLM Connector Requirement

This extractor requires a properly-configured LLM Connector on the repository Root to communicate with the LLM service. Ensure the connector is set up in your environment.

Using Clause Detection with AI Extract

One technique involves running an AI Extract operation on the section instance that Clause Detection produces. The basic idea is:

• Clause Detection is selected and configured for the Data Section's "Extract Method".
• An AI Extract specification is added to the Data Section's "Fill Methods" collection.
• First Clause Detection finds the most similar text chunk (or chunks) to the sample "Semantic Quoting Queries" (the sample clauses provided in Detection Settings).
• Then, AI Extract will run on the section of text Clause Detection returns.

Clause Detection was designed in such a way that Grooper expects AI Extract to execute only against the text section it returns. When using Clause Detection, you should not configure a Quoting Method for the Data Section's AI Extract specification.

• Users should keep AI Extract's "Document Quoting" setting as its default value. This will pass only the text returned by Clause Detection as the quote sent to the LLM.
• Users may experience unpredictable results when using other Quoting Methods, including DI Layout.

Notes

• Clause Detection is best used with documents that contain rich, natural language content.
• The accuracy of detection depends on the quality of the sample clause and the selected embedding model.
• For more information, see the documentation for Semantic and Section Extract Method.

Properties

Name	Type		Description
Detection Settings	Semantic	►	Defines one or more sample clauses and semantic matching settings for clause detection. EditorObject Properties Editor ConverterExpandable Converter The 'Detection Settings' property allows you to specify examples of the clause to be detected, as well as configuration options for semantic similarity matching. This includes the sample text, the embedding model to use, chunking and clustering parameters, and other advanced options. Adjust chunking and clustering parameters to control the granularity and sensitivity of matching. For more details, see the documentation for Semantic.

See Also

Semantic

Used By

Data Section