Must a factoid answer be one word?

No. A short phrase or sentence may be needed to express the fact clearly and preserve important qualifiers.

Are all “who,” “where,” and “when” questions automatically straightforward factoid tasks?

No. They may be ambiguous, seek a list, or require clarification about the intended context.

Is returning a relevant document enough?

A question-answering task typically expects an answer; relevant documents can provide evidence, but relevance alone does not identify the requested fact.

AI Glossary / Natural language and speech / Intermediate

Factoid questions

Definition

Factoid questions seek a specific factual answer that can usually be expressed briefly, such as a person’s name, a place, a date, or a quantity. In question-answering research, the term describes an answer type, not a guarantee that the question is easy, unambiguous, or supported by the available evidence.

What it means

A question such as “Which room hosts the training session?” typically asks for a short factual response. By contrast, “Why was that room chosen?” may require an explanation involving several considerations.
The distinction concerns what information would satisfy the request. A factoid answer can be concise even when finding it requires resolving names, comparing sources, or interpreting a date.
The word “factoid” has other uses outside natural language processing. Here it is the established question-answering term for a short factual information need. It should not be read as permission to provide a plausible-sounding detail without checking whether the source supports it.

How it works

A system first interprets what the question asks, including the target entity, relationship, and any time or scope restrictions. It then seeks evidence in an appropriate source.
In a retrieval-based approach, the system finds relevant passages and identifies or produces the answer from them. In a knowledge-based approach, it maps the question to a query over structured records. Other systems may generate answers using a language model, with or without retrieved evidence.
Regardless of implementation, the proposed answer must satisfy the actual question. A nearby date may describe an announcement rather than the event being asked about. If the source lacks the answer or the question is ambiguous, acknowledging that limitation is more reliable than inventing a specific value.

Three practical examples

An internal assistant answers “What is the room number for Thursday’s orientation?” from an approved event record.
A catalog search tool answers “Who created this exhibit?” using the named exhibit’s curator-approved description.
A reading-comprehension exercise asks students to identify the date of a fictional opening from a short passage containing several dates.

Think of it this way

Think of locating the right field on the right record card. The answer may fit in a small box, but choosing the correct card and field can require careful work. The analogy is limited because real questions often require interpreting prose, context, and changes over time rather than reading a single database cell.

Worked example

Suppose a fictional notice says: “The West Hall exhibition was announced on 3 March. It opened to visitors on 18 April and closed on 30 June.”
For “When did the West Hall exhibition open?”, the answer is 18 April. The notice’s first date, 3 March, is relevant to the same exhibition but answers a different question. The date 30 June describes closure, not opening.
For “In which year did the exhibition open?”, the passage does not provide enough information. A system should not infer a year merely because the day and month are available.
For “Why did the exhibition close?”, none of the three dates supplies the requested explanation. This question is not answered by selecting the closure date.
The example shows three separate checks: identifying the event, selecting the correct relation, and ensuring the requested level of specificity is supported. A short answer can be wrong even when every word in it appears in the source.

Illustrative case study

In a fictional project, Maple Learning adds a question-answering feature to its course handbook. Readers ask brief questions about course locations, session dates, and named coordinators.
The team tests questions against specific handbook versions and asks reviewers to check both the answer and its supporting passage. Questions about a later schedule change are left unresolved when the available handbook does not contain that update.
The imagined workflow illustrates evidence-aware factual answering. It does not claim that short responses are always safer or that adding a citation automatically proves the cited passage answers the question.

Common misconceptions and mistakes

A factoid question is not necessarily a trivia question. It can concern a practical detail in a workplace or personal document.
Short does not mean simple. Ambiguous names, temporal changes, and multiple candidate facts can make a brief-answer task difficult.
A factoid question does not always have a usable answer in the given collection. The world may contain the fact while the system’s evidence does not.
Finally, an answer that is correct somewhere can still be unsupported by the cited source or wrong for the question’s intended time, entity, or scope.

Business use

Factoid question answering can reduce the effort of locating specific details in manuals, catalogs, schedules, and knowledge bases. It works best when the system can identify reliable sources and preserve the boundaries of what those sources establish.
Evaluation should test correct answers, plausible distractors, ambiguous questions, and cases where the answer is absent. Checking answer correctness separately from evidence support helps reveal unsupported guesses.
For operational use, make the source and relevant version visible where helpful. A concise answer should retain necessary qualifiers, such as which branch, product version, or period it applies to, even if that makes the response slightly longer.

Check your understanding

A passage says a workshop was scheduled on Monday for an event taking place Friday. The user asks when the workshop takes place. Which day answers the question?

Answer: Friday, assuming the passage is the relevant evidence. Monday describes the scheduling action, not the workshop’s occurrence.

Related terms

Question answering
Information retrieval
Entity linking
reading comprehension
knowledge-based question answering
answer extraction
evidence grounding
unanswerable question
temporal reasoning

Learn next

Study the difference between retrieving a passage and answering a question from it. Then practice with passages containing several names or dates, including questions that cannot be answered from the text. Compare exact answer matching with human checks for meaning, qualification, and supporting evidence.

Sources

Jurafsky and Martin, Speech and Language Processing, archived question-answering chapter, definition and introductory paradigms: https://web.stanford.edu/~jurafsky/slp3/old_sep21/23.pdf. NIST TREC, original question-answering collection description and supported-answer judgments: https://trec.nist.gov/data/qa.html. Independent review checked the archived chapter’s factoid definition, retrieve-and-read approach and unanswerable cases, plus NIST’s original collection description.

Previous: Fact table Next: Factor