What’s a Pinpoint Document

Template?

Latest Update: 12-June-2023

Introduction

Journalist Studio/Pinpoint is introducing a new tool for extracting data from form-like

documents, named “Extract structured data”. It enables you to import form-based data,

and export the processed data as a spreadsheet. For example, if you have ten thousand

scanned auto accident reports that use a similar form, you can import the scans and export

a spreadsheet that enables you to group, sort, or filter accidents by date, automobile

So, What’s a “template”?

Intuitively, documents that share the same template look identical, except for the values of

some fields (which typically are what the user is interested in extracting). In other words,

they all look like they were printed by the same clerk. Minor differences due to scanning

(e.g. slight rotation) are also allowed.

More accurately, a template is a rendering of structured information with considerable

boilerplate (common) text that maintains a consistent reading order and a consistent

horizontal position. These concepts are explained below.

A collection of documents that merely have the same title, or that provide similar

information but in vastly different organization, are not a template. For example, a

1

Some examples of templates that aren’t supported:

● Top-to-bottom text (for example, traditional Japanese).

● Column format (as in some newspapers) where reading order flows from the bottom

of a column to the top of the next column.

● Cards (text boxes) that are laid out side by side (even if each card may have a valid

reading order) - example.

Contents of a template

A template can have boilerplate text, key/value pairs (for example, name and address

sections in a form), and tables.

It actually works well also for right-to-left if the document is written in a right-to-left

language.

The current release supports two types of values:

● String, inc. numerical values which are automatically converted to strings (so 1234 is

extracted as "1234")).

● Checkbox (with value extracted as either true or false).

Fixed Forms

The simplest example of a template is a fixed form: Each document has a fixed number of

pages, with the exact same layout across documents. For example, if the form has a

“Residence Address” field with an adjacent box where the address is expected to be filled in,

then both the key and the value box reside in almost exactly the same location in each of

the documents. The only difference between the documents is the values that are written in

these value boxes.

Non Fixed Forms

Unlike fixed forms, many form-like documents are printed from some structured data by a

computer system, and may include, besides the boilerplate and fixed key-value pairs, also:

● Optional fields or sections.

● A variable number of fields or sections.

● Variable-length tables.

In a valid template, all boilerplate text and key-value pairs are expected to keep the same

reading order. For example, if a template has a “Name” key followed by an “Address” key,

then in all documents “Name” must appear before “Address”. They are also assumed to have

a fixed horizontal position. So, if “Name” appears at the start of the line in one document, it

containing one or more key-value fields describing this entity. The section repeats multiple

times, each time using the same boilerplate and keys, but typically different values. The

repeated section can be viewed as a mini-template having a variable number of consecutive

instances within the full template.

A simple example:

Involved Persons:

As with the whole document, a repeated section in a template requires that:

● Reading order is fixed (that is, it is not allowed for one instance to swXitch the rows

between “First Name” and “Last Name”).

● Horizontal position of each boilerplate/key is fixed (that is, it is not allowed for one

instance to switch the position between “Badge #” and “Gender”).

structure is defined by the column layout: their horizontal positions and their column

headers (if any).

Despite their intuitive similarity, there is a fine line differentiating tables from repeated

sections:

● Repeated sections contain Key-value pairs, whereas tables contain values.

● Tables may (but aren’t required to) contain column headers, which represent the

“key” for all values in that column.

● Note that while grid lines are more common in tables, they don’t necessarily rule out

repeated sections (as seen in the first example).

In rare cases, it may become difficult to tell whether a template component is a table or a

repeated section, but in most cases following the above guidelines makes the choice clear.

occur. A collection of documents that are originally of the same template are still considered

to be of that template even if they are slightly distorted in this way.

Minor Changes

The tool is able to overlook small differences in documents, including:

● Font/style changes.

● Slight changes to boilerplate text.

● Addition of new boilerplate text or new key-value pairs.

Strictly speaking, such a collection is NOT of a single template, but of a number of close

templates (e.g. yearly versions of the same form). If the differences are small, the tool can

overcome these differences, but if the differences are large (especially changes to text or

garbage. But sometimes you may be surprised to get some good results. This is especially

true if there’s a segment of the documents (e.g the first page) that adheres to these rules

while the rest doesn’t. You may be able to get good extraction results from the “good”

segment.

Examples (from across the web)

● Country factbook dataset: Toy dataset built for demos. Includes key-value pairs,

tables and repeated-section. Great for learning how to use the tool.

● Medical Examination Report Form: Despite its complexity, if the form is fixed then it’s

a valid template. Some sections (e.g. “Testing”) may have a column layout, but as

long as it is fixed across documents, that’s still valid. Note that support for extracting