AI Glossary / Natural language and speech / Advanced

Clefts

Definition

Clefts are sentence constructions that present a message in a divided structure, commonly highlighting one constituent against a supporting clause. In English, an it-cleft such as “It was Lena who repaired the scanner” highlights Lena. A wh-cleft, also called a pseudo-cleft, can express a related distinction: “What Lena repaired was the scanner.”

What it means

A cleft helps organize what a sentence is doing in its conversation. Compare “Lena repaired the scanner” with “It was Lena who repaired the scanner.” Both identify Lena as the person who performed the repair, but the second can be particularly appropriate when correcting someone who thought Omar performed it.
The highlighted constituent is often discussed as the focus. The remaining material supplies a background proposition with a slot to fill: someone repaired the scanner. Background here does not mean unimportant, and it does not guarantee that the listener already knew the information. Context affects how the sentence is interpreted.
For language systems, recognizing the event alone may therefore be insufficient. The system may also need to preserve which participant, object, place, or time the speaker is emphasizing or correcting.

How it works

A common English it-cleft combines it, a form of be, a focused constituent, and a following clause introduced by that or who. The focused constituent is understood in relation to a missing element in that clause. In “It was the scanner that Lena repaired,” the scanner supplies the object of repaired.
A common wh-cleft instead uses a clause beginning with what and a linking form of be: “What Lena repaired was the scanner.” Their detailed syntax depends on the construction and grammatical analysis.
A language system can connect the focused phrase to its role in the described event, then use surrounding dialogue to interpret the contrast. Surface word order alone is not a complete analysis. Intonation can contribute to focus in speech, and similar communicative functions can be expressed without a cleft. Languages also differ in how they mark these distinctions.

Three practical examples

• A question-answering system reads “It was on Tuesday that Lena repaired the scanner” and extracts Tuesday as the repair time rather than treating the sentence's initial it as the repaired object.
• A meeting summarizer preserves the correction in “It was the scanner, not the printer, that Lena repaired,” because naming only the repair event would lose the disputed distinction.
• A writing assistant changes a sentence's emphasis when the author wants to highlight who completed a task, while checking that the rewrite preserves the intended participants and action.

Think of it this way

Think of a cleft as placing a spotlight on one part of a scene. The rest remains necessary, but the spotlight directs attention. Moving the spotlight can answer a different question without changing the people and objects on stage.

Worked example

Consider this original teaching event: Lena repaired the scanner on Tuesday. Represent its basic content as an actor, an action, an object, and a time: actor=Lena; action=repair; object=scanner; time=Tuesday. This record omits tense details and conversational meaning.
Now consider three clefts:
“It was Lena who repaired the scanner on Tuesday.” The highlighted answer is Lena, appropriate to a question about who performed the repair.
“It was the scanner that Lena repaired on Tuesday.” The highlighted answer is the scanner, appropriate to a question about what she repaired.
“It was on Tuesday that Lena repaired the scanner.” The highlighted answer is on Tuesday, appropriate to a question about when the repair occurred.
All three recover the same four fields in the simplified event record. Their intended focus annotations differ: actor, object, and time respectively. An extraction test that checks only the four event fields would award all three the same result and would not test whether the system detected the focus.
Suppose a toy system correctly extracts all four fields from each sentence but labels the focus as actor in every case. Its event-field score is 12 correct fields out of 12, or 100%. Its focus accuracy is one sentence out of three, approximately 33.3%. These stipulated outputs illustrate different evaluation targets, not measured model performance.
Finally, compare “It was not Lena who repaired the scanner.” This denies the identification of Lena as the repairer; it should not be normalized into the positive event record above. A reader may accommodate a background assumption that someone repaired the scanner, but that assumption does not identify the actual repairer or independently verify that a repair happened.

Illustrative case study

Fictional example: BriefLens summarizes maintenance discussions. Its first design reduces every repair sentence to an actor-action-object record. In a correction-heavy meeting, this loses the distinction between an initial claim and a later speaker's correction about the equipment involved.
The team adds review examples containing object clefts, time clefts, and negated clefts. Reviewers check whether summaries preserve corrections and whether extracted records reflect the accepted statement in context. The scenario illustrates an evaluation need; it reports no improvement percentage or claim that one parsing technique solves the problem.

Common misconceptions and mistakes

Not every sentence beginning with it and be is a cleft. “It is raining” does not contain a focused constituent linked to a following cleft clause. Likewise, identifying a phrase as focused does not make it the grammatical subject of the repair event.
Another mistake is treating the background as necessarily old information or independently established fact. Clefts can present information as taken for granted without proving it. Removing cleft structure may preserve a basic event description while losing contrast, emphasis, or important effects of negation. Rewriting therefore requires more than deleting the opening words.

Business use

Cleft awareness is useful in summarization, question answering, translation, and correction handling. Evaluate whether the output preserves who did what, the relevant circumstances, and the distinction the speaker intended to resolve. Include examples where the focused element is a time or object, alongside actor examples, and inspect negated statements separately.

People also ask

Is focus always expressed through clefting?

No. Prosody and other constructions can also mark it.

Are all clefts it-clefts?

No. Wh-clefts are another familiar English pattern.

Does background information have to be true?

A construction can present an assumption without verifying it.

Check your understanding

Which phrase answers when in the worked example?

Answer: On Tuesday.

Why can event extraction be perfect while focus detection fails?

Answer: They measure different information.

Should the negated Lena sentence create a positive Lena repair record?

Answer: No.

Related terms

• Contrast
• Discourse
• Semantic role
• Focus
• information structure
• presupposition
• pseudo-cleft
• constituent

Learn next

Write actor, object, and time clefts for one event. Compare the questions they answer, then introduce negation and identify which parts of the interpretation change.

More in Natural language and speech

• Clarification - Clarification is a conversational act that seeks to resolve uncertainty about what another participant said or meant.
• Class-based n-gram - A class-based n-gram is a language model that predicts words using groups of words called classes.
• Classifier head - A classifier head is the part of a neural model that converts learned representations into scores for a classification task.
• Clitic - A clitic is a linguistic element that depends on a neighboring host for its pronunciation while contributing grammatical information.
• Closed class - A closed class is a grammatical word category whose membership is relatively stable and does not readily expand through ordinary creation of new words.
• Closure, stop - A stop closure is the interval during production of a stop consonant when the relevant articulators form a complete obstruction in the vocal tract.

Sources

King and Zaenen, F-structures, Information Structure, and Discourse Structure, LFG 2004 extended abstract, clefts and information packaging: https://web.stanford.edu/group/cslipublications/cslipublications/LFG/9/abstracts/lfg04kingzaenen-abs.html Cambridge University Press, Non-Canonical English Syntax, chapter 9, cleft types: https://www.cambridge.org/core/books/noncanonical-english-syntax/what-was-it-about-it-that-you-loved/432E3C10171545FBEA5C3958D8AB4121

Previous: Clause Next: CLEVER score