Must every field contain a value?

No. A faithful record may leave fields missing or uncertain when the source does not support them.

Can a template be filled from several sentences?

Yes, provided the values and references are correctly linked to the same situation.

Is this the same as generating an email from a form?

No. Here the direction is from source language to a structured representation.

AI Glossary / Natural language and speech / Intermediate

Template filling

Definition

Template filling is an information-extraction task that converts descriptions of recurring situations or events into structured records with predefined fields. A system identifies the relevant situation and supplies values for roles such as participant, location, amount, or date. Values may be extracted text spans or normalized concepts supported by the source material.

What it means

Many documents describe similar kinds of situations in different wording. A template defines the information an application wants to collect consistently, even when that information appears in different positions or across several sentences.
For example, a delivery template might distinguish the sender, recipient, item, and arrival date. Recognizing all the names in a document is not enough; each value must be assigned to the correct role and event.
This meaning of template filling is about interpreting source text. It differs from filling blanks in a prewritten message to generate prose, even though both activities use the everyday word template.

How it works

Define the event or situation schema and the types of values allowed in each field. Determine which source passages describe instances of that situation.
Extract candidate values and assign them to roles. Resolve references when different mentions denote the same participant, and keep separate records when the document describes different events.
Normalize values such as dates only when the necessary context is available. Preserve the distinction between an explicitly stated value, an inference, and missing information.
Validate the completed record against the source and schema. A template can be partially filled when information is absent. The expectation that a typical event usually has a participant or date does not justify inventing one to make the record look complete.

Three practical examples

An event directory extracts a seminar’s title, speaker, venue, and date from announcements with varied layouts.
A logistics analysis tool converts delivery reports into records that distinguish sender, recipient, goods, and timing.
A research project organizes descriptions of organizational partnerships into consistently defined participant and activity fields.

Think of it this way

Think of completing a structured case card from a narrative. The fields make comparison easier, but the card must remain faithful to the evidence. The analogy is limited because a narrative can describe several overlapping situations, use indirect references, or omit details that a form appears to expect.

Worked example

Consider this fictional note: “On 8 May 2026, Arlen Labs sent two sensors to Brook School. The shipment arrived on 10 May. A separate parcel of cables will be sent next week.”
For a shipment template, the first record can contain sender = Arlen Labs, recipient = Brook School, item = sensors, quantity = 2, dispatch date = 2026-05-08, and arrival date = 2026-05-10. “The shipment” links the second sentence to the sensor shipment.
The cables belong to a separate planned shipment. The text does not explicitly provide its quantity or establish that it has already been sent. “Next week” also needs a reference date before it can be normalized confidently.
Copying the sensor quantity and arrival date into the cable record would create unsupported information. The exercise shows that role assignment, event separation, reference resolution, and event status are all part of producing a faithful structured result.

Illustrative case study

In a fictional project, Cedar Events compiles workshop announcements into a searchable calendar. Some announcements mention an earlier session while promoting a future one, and a simple extractor occasionally combines the earlier date with the new speaker.
The team checks each filled record against the event it represents and keeps evidence links for the selected fields. Missing venue details remain unresolved rather than being copied from a previous event without support.
The imagined workflow prioritizes coherent records over superficially complete ones. It does not claim that a predefined schema captures every useful detail in an announcement.

Common misconceptions and mistakes

Template filling is not merely named entity recognition. Finding a person and a location does not determine their roles in an event.
It is not enough to collect values anywhere in a document; the values must belong to the same intended record.
An unfilled field is not necessarily a system failure when the source omits the information.
Finally, normalized or inferred values should not be presented as direct quotations. Their derivation and supporting context need to remain distinguishable.

Business use

Template filling supports searchable databases, document triage, reporting, and repeatable analysis of recurring information. It can reduce the effort of turning varied prose into comparable records.
Evaluate complete records as well as individual fields. A collection of individually correct values can still be wrong when values from different events are combined.
Document the schema, allowed inferences, normalization rules, and treatment of uncertainty. Preserve enough source context for review, particularly where a mistaken date, participant, or status would change a business decision. The objective is dependable structured information, not filling every available box at any cost.

Check your understanding

A report describes two shipments but provides a date only for the first. Should the extractor copy that date into both records?

Answer: No. It needs evidence that the date applies to the second shipment; otherwise that field should remain unresolved.

Related terms

Learn next

Separate event recognition from role assignment, then inspect a document containing two similar events. Trace every filled field to its supporting passage and identify which values require normalization or remain unknown.

Normalization

Sources

Source: Jurafsky and Martin, Speech and Language Processing, section21.8 on template filling and role-filler extraction: https://web.stanford.edu/~jurafsky/slp3/21.pdf . Independently consulted sections 21.8–21.8.1 on structured slots, extracted or inferred fillers, recognition, role extraction and coreferent mentions.

Previous: Template Next: Template matching