Powered by Zoomin Software. For more details please contactZoomin

The Semaphore Fact Extraction Framework (FACTS)

Near Ordered Across Paragraphs In Document

  • Last Updated: July 8, 2026
  • 3 minute read
    • Semaphore
    • Documentation

Summary

The Near Ordered Across Paragraphs In Document context looks for facts and/or anchors, each in different paragraphs in the same document, where those paragraphs are ordered and separated by at most a specified number of words (set by the near count metadata).

Usage notes

The count on this extractor refers to the number of tokens (words, and, if you choose not to ignore punctuation, punctuation marks as well) between the paragraphs that each independently contain an anchor or a fact within the same document.

Fundamental patterns for model and content

If you have an extractor that is an instance of Near Ordered Across Paragraphs In Document with:

  1. anchor/fact: “phrase range 1”
  2. anchor/fact: “phrase range 2”
  3. anchor/fact: “phrase range 3”

and a near count metadata of 20, then the following content should fire:

This is “phrase range 1” that our anchor or fact is looking for in a separate paragraph from the other anchor or fact phrase ranges. Here are lots more words in the first paragraph which are NOT counted towards the “count”, as they are in the same paragraph that the evidence has been found in.

These are some words that will be skipped over as part of the “count”.

As will these.

This is “phrase range 2” that our anchor or fact is looking for in a separate paragraph from the other anchor or fact phrase ranges. Here are lots more words in the second paragraph which are NOT counted towards the “count”, as they are in the same paragraph that the evidence has been found in.

And these also.

This is “phrase range 3” that our anchor or fact is looking for in a separate paragraph from the other anchor or fact phrase ranges. Here are lots more words in the third paragraph which are NOT counted towards the “count”, as they are in the same paragraph that the evidence has been found in.


Near Ordered Across Paragraphs In Document Context Properties

Hierarchical Relationships

At least one of the following is required:

Label Range Constraints
fact Context Elements one or more allowed
not anchor Context Elements one or more allowed
anchor Context Elements one or more allowed
not fact Context Elements one or more allowed
optional anchor Context Elements one or more allowed
optional fact Context Elements one or more allowed

Associative Relationships

No associative relationships are mandatory.

Label Range Constraints
depended on by context Contexts only one allowed
depends on context Contexts only one allowed
extract fact as Fact Name only one allowed
group fact as Fact Name only one allowed
location Locations only one allowed
precluded by context Contexts only one allowed
precludes context Contexts only one allowed
punctuation Punctuation rules only one allowed

Metadata

A near count is mandatory.

Label Range Constraints
near count Integer only one allowed
context position from document end Integer only one allowed
context position from document start Integer only one allowed
fact position from context end Integer only one allowed
fact position from context start Integer only one allowed
field String only one allowed
return group fact ID Boolean only one allowed
return raw text also Boolean only one allowed
return value using regex String only one allowed
{1-10} greedy repeat Integer one or more allowed
{1-10} non-greedy repeat Integer one or more allowed
debug Boolean only one allowed
TitleResults for “How to create a CRG?”Also Available inAlert