Does given always mean previously mentioned?

No. Information may be recoverable from shared context or inference, although the speaker can misjudge that availability.

Is new information always placed at the end?

No. That can be a useful pattern in some contexts, but it is not a universal grammatical requirement.

Does a pronoun automatically refer to the most recent noun?

No. The intended referent depends on grammatical and discourse evidence, and may remain ambiguous.

AI Glossary / Natural language and speech / Intermediate

Given-new

Definition

Given-new describes a distinction in how information is presented in discourse. Given information is treated as already available or recoverable for the listener or reader; new information is presented as the contribution being added. The distinction helps explain reference, sentence organization, and why a reply can be factually relevant yet awkward if it assumes the wrong shared background.

What it means

Suppose a conversation has established that a parcel arrived, but not who signed for it. “Lina signed for it” connects an available parcel and signing question with a new answer. The name is new in the relevant conversational role.

Different theories distinguish discourse-given, hearer-known, inferable, focused, and topical material in different ways. This entry uses given-new as an introductory information-structure distinction and identifies the classic given-new contract as one influential account, not a complete universal model of communication.

How it works

A speaker chooses wording based partly on what they believe the audience can identify. A listener connects that wording with available discourse, shared context, or an appropriate inference, then integrates the added information.

Clark and Haviland's given-new contract describes this coordination in comprehension: speakers should make the assumed background recoverable and contribute information that can be attached to it. If the assumed connection is missing or ambiguous, the listener may have to infer it, ask for clarification, or reinterpret the utterance.

For an assistant, this suggests tracking which entities and claims have been introduced and which assumptions are merely inferred. It does not justify pretending to know the user's private knowledge. A system should distinguish information in the conversation from information it possesses elsewhere and from what the user can reasonably be expected to recognize.

Three practical examples

1. Customer support: An assistant introduces a newly discovered component before using an ambiguous pronoun such as it, helping the customer connect the next instruction to the right object.

2. Report writing: A summary starts a sentence from an already established project and adds its new status, rather than assuming an unexplained label is familiar to every reader.

3. Dialogue evaluation: Reviewers check whether an answer addresses the information gap in the question, not simply whether it repeats related facts.

Think of it this way

Think of adding a note to an existing folder. The given part helps the reader find the intended folder; the new part supplies the note to place inside it.

The analogy is limited: people do not literally store conversation in labeled folders, and the needed connection can be inferred rather than explicitly mentioned. Different listeners may also have different background knowledge, so the speaker's assumption can be mistaken.

Worked example

Consider an original exchange. U1 says, “A replacement pump arrived this morning.” U2 asks, “Who tested the pump?” U3 answers, “Asha tested it.” The pump is available from U1, and U2 establishes the question of its tester. U3 supplies Asha as the answer while using it to refer to the already introduced pump.

There are three separate pieces of information to preserve: a pump arrived, a question was asked about testing, and Asha is asserted as the tester in the answer. U2 alone does not supply the answer Asha. Depending on context, its wording may also carry an assumption about testing that should not be confused with independently verified evidence.

Now place “Asha tested it” at the start of a conversation with no shared object in view. The pronoun's intended referent is not recoverable from the text alone. The sentence could be made clearer as “Asha tested the replacement pump,” provided that description identifies the intended object for the audience.

The example shows that the same words can be appropriate in one context and underspecified in another. Counting repeated words would miss the fact that pump and it can connect to the same entity.

Illustrative case study

Fictional case: A software help center sends a troubleshooting message that begins, “Restart it after clearing the queue.” New customers have not been told which service or device is intended. Staff understand it from their workflow knowledge.

Editors introduce the relevant component and explain the queue before giving the instruction. They test the revised message with readers who lack the staff's background. The change addresses an assumption about audience knowledge, rather than merely replacing a difficult technical word.

Common misconceptions and mistakes

Given information is not necessarily information already expressed with exactly the same words. A referent may be available from context, a synonym, or an appropriate bridging inference.

New does not mean recently invented or chronologically recent. An old historical fact can be new to the current exchange, while today's date may already be shared.

Topic, focus, and given-new should not be treated as interchangeable labels in every framework. Nor is given-before-new an exceptionless sentence-order law. Languages, constructions, contrast, and rhetorical aims can produce other arrangements.

Most importantly, presenting something as given does not make it true. Background assumptions can be mistaken or disputed, and an assistant must not treat presupposed material as independently checked fact.

Business use

For businesses, given-new offers a practical way to review instructions and explanations: identify what the reader is expected to recognize and what the message adds. This helps expose unexplained labels, unclear pronouns, and answers that fail to address the user's actual information gap.

Use audience-specific review rather than assuming that an internal expert's knowledge is universal. Preserve uncertainty where background assumptions are unconfirmed. In conversational systems, assess reference clarity and answer relevance alongside factual accuracy and grammatical fluency.

Check your understanding

Why is “Asha tested it” clearer after a conversation about a replacement pump than as an isolated opening sentence?

Answer: The earlier discourse supplies a recoverable referent for it. Without that context or a shared situation, the reader cannot determine the intended object from the sentence alone.

Related terms

Common ground
Anaphora
Bridging inference
Information structure
discourse-given information
focus
topic

Learn next

Learn next about common ground and referring expressions. Review a short instruction by marking its assumed background, new contribution, and unresolved references, then revise it for a reader who lacks internal project knowledge.

Common ground

Sources

Clark and Haviland's original 1977 chapter, “Comprehension and the Given-New Contract,” was read from Clark's author-hosted archive, particularly its account of the contract and recoverable antecedents: https://web.stanford.edu/~clark/1970s/Clark,%20H.H.%20_%20Haviland,%20S.E.%20_Comprehension%20and%20the%20given-new%20contract_%201977.pdf . The pump exchange and help-centre case are original illustrations. This historical account is not presented as collapsing every later distinction in information-structure theory.

Previous: Gittins index Next: Glejser test