Can that be left out?

Often in English complement constructions, yes, but acceptability depends on the expression and context.

Is whether always equivalent to that?

No. Whether introduces an embedded alternative question, while declarative that has a different role.

Do all languages place these markers before the clause?

No universal position is assumed here; language-specific grammar and annotation conventions must be consulted.

AI Glossary / Natural language and speech / Intermediate

Complementizers

Definition

Complementizers are words that mark an embedded clause, especially a clause functioning as the complement of another expression. English that in “Rina believes that the shipment arrived” and whether in “Rina asked whether it arrived” are familiar examples.

What it means

A complement clause supplies content required or licensed by another expression. In “Rina believes that the shipment arrived,” the embedded clause expresses what Rina believes. The complementizer helps signal its connection to the larger sentence.

The spelling alone is not enough. That in “that shipment” is a determiner, while that in the belief sentence marks a clause. Whether normally introduces an embedded question about alternatives. If can introduce a question in “ask if it arrived,” but it can also introduce a condition in a different construction.

Linguistic theories differ in how broadly they use complementizer and how they represent these words. This entry focuses on clear clause-introducing examples rather than claiming that every grammar assigns an identical category or tree position.

How it works

A parser identifies the embedded clause and relates it to the expression that licenses it. In a dependency analysis of “Rina believes that the shipment arrived,” arrived can head the embedded clause, believes connects to that clause, and that marks its subordinate status.

Universal Dependencies uses SCONJ for subordinating conjunctions, a class that includes complement-clause markers but also other subtypes. Its ccomp relation identifies certain clausal complements, while mark links a marker to the clause head. Those labels describe distinct pieces of the analysis rather than interchangeable names for one object.

Some English complementizers can be omitted: “Rina believes the shipment arrived” still contains an embedded clause. A system that finds clauses only by searching for that will therefore miss valid examples. Accurate analysis must use the complete grammatical structure and preserve distinctions such as reporting, questioning, negation and uncertainty.

Three practical examples

An information-extraction tool recognizes that a reported claim belongs to the person making the claim rather than automatically treating it as an established fact.
A grammar tutor contrasts that in a noun phrase with that introducing an embedded clause.
A question-answering pipeline identifies the unresolved proposition in “Check whether the order has shipped.”

Think of it this way

Think of a complementizer as a sign at the entrance to an embedded room in a sentence. The sign helps show how that room connects to the surrounding structure.

The analogy does not imply that every room needs a visible sign. Clauses can lack an overt complementizer, and the word's interpretation depends on the construction rather than on a fixed label printed on its spelling.

Worked example

Preserving the scope of a reported statement

Take the invented sentence “The supervisor said that the delivery was delayed.” Its main reporting event is said, with supervisor as the speaker. The embedded content is “the delivery was delayed,” introduced by that.

A teaching representation can record report(speaker=supervisor, content=delayed(delivery)). It should not silently remove report and present delayed(delivery) as independently verified by the assistant. The word that helps locate the clause, but the reporting verb and wider context establish the attribution.

Now use “The supervisor asked whether the delivery was delayed.” The embedded content is a question, so summarizing it as “The delivery was delayed” would be especially misleading. Finally, omit that from the first sentence: “The supervisor said the delivery was delayed.” The reported-content relationship remains.

These three cases show why punctuation and keyword lists cannot replace structural interpretation.

Illustrative case study

A fictional logistics firm extracts incident information from short notes. Its first system treats every phrase containing “was delayed” as a confirmed delay. This merges customer questions, denied claims and statements attributed to other people.

The team adds test cases with reported and questioned clauses, including examples without an overt that. Reviewers inspect whether the summary retains the speaker and whether the embedded content is asserted, questioned or rejected.

The team uses parsing as one source of evidence rather than assuming a parse tree alone establishes factual truth. The practical lesson is that small grammatical markers can help expose information structure, but the whole construction determines what a document actually commits to.

Common misconceptions and mistakes

Every occurrence of that is a complementizer. Its grammatical role changes with context.
Every subordinate clause begins with an overt complementizer. Some permit omission.
Complementizer and clausal complement mean the same thing. One is a marker; the other is a clause with a grammatical role.
SCONJ contains only complementizers. The UD category also covers other subordinating markers.
Content inside a complement clause is automatically true. Reporting, doubt and questions can embed it without establishing it.

Business use

Complementizer-aware analysis is useful when extracting claims, requests and questions from documents. Preserve attribution and uncertainty instead of flattening nested sentences into disconnected facts. This is especially valuable in support histories, meeting notes and incident summaries.

Use examples that vary only a few words: said that, asked whether and denied that can lead to very different interpretations. Include omitted markers and ambiguous spellings. Evaluate whether the final structured record preserves the intended relationship, because a correct word-level tag is only an intermediate success.

Check your understanding

Why should “She asked whether the server failed” not be summarized as a confirmed server failure?

Answer: The embedded clause presents the content of a question. The sentence reports asking about a possible failure, not establishing that the failure occurred.

Related terms

Clause
clausal complement
subordination
SCONJ
dependency parsing
reported speech
syntactic scope

Learn next

Study clausal complements and reported speech. Annotate a small set of sentences containing that, whether and omitted markers, then compare what each sentence asserts with what it merely reports or asks.

Sources

Universal Dependencies, SCONJ: https://universaldependencies.org/u/pos/SCONJ.html. Complement-clause markers within a broader category.

Universal Dependencies, ccomp: https://universaldependencies.org/u/dep/ccomp.html. Clausal-complement and marker examples.

Jurafsky and Martin, Sequence Labeling: https://web.stanford.edu/~jurafsky/slp3/18.pdf. Contextual ambiguity of that. Examples and logistics case are original illustrations.

Previous: Complementation Rule Next: Complete assignment