Must a discourse-old referent be mentioned using the same words?

No. A name, description, or pronoun can access the same established referent.

Is everything the listener already knows discourse-old?

No. Familiarity to the listener and prior introduction into the discourse are different dimensions.

Can a referent be old in one analysis and new in another?

Yes. If one analysis includes the earlier conversation and another begins later, the available discourse context differs. The chosen boundary should be explicit.

AI Glossary / Natural language and speech / Intermediate

Discourse-old

Definition

Discourse-old describes a referent that has already been established in the current discourse and is accessed again by a later expression. For example, after introducing a particular report, a speaker may refer to it as “the report” or “it.” The status belongs to the referent in context, not to the age or familiarity of the word.

What it means

A conversation builds a shared working representation of what it is discussing. Later references can reuse those established entities without reintroducing every detail.
The wording may change while the identity stays the same. “Maya Chen,” “the project lead,” and “she” can all refer to one person if the context supports that interpretation. Conversely, a repeated phrase such as “a customer” can introduce a different person and therefore need not be discourse-old.
Discourse-old also differs from hearer-old. A listener may know a famous organization before it is mentioned, but its first mention can still be new to this discourse. The distinction makes information status more precise than a general notion of familiarity.

How it works

A system tracks referents across successive mentions and decides whether each expression links to an entity already represented. Evidence can include names, grammatical features, semantic compatibility, and the surrounding conversation.
When a link is justified, later mentions can be grouped in a coreference chain. This preserves identity across different descriptions and pronouns. The system should not create a new entity merely because the surface phrase changes.
The analysis also depends on scope. A referent may be old within a complete conversation but appear new when a single message is analyzed in isolation. Inferable references and ambiguous mentions require annotation conventions rather than automatic treatment as explicitly old.

Three practical examples

A meeting summarizer connects “the revised plan” to the plan introduced earlier, so it does not list two unrelated plans.
A support assistant interprets “it still fails” using the previously established device or operation, while checking whether more than one candidate is available.
A document annotation team links a person's name and later role description when both identify the same individual.

Think of it this way

Think of retrieving an existing card from a conversation's index instead of creating a duplicate. The label used to find the card may differ from its original title. The analogy captures continuity of identity, but actual discourse understanding can involve ambiguity, inference, and changing descriptions that a simple card index cannot resolve by itself.

Worked example

Consider this invented note: “A technician inspected a printer. The device needed a replacement roller. The technician ordered the part. Another technician arrived later.”
The first sentence introduces two referents: one technician and one printer. In the second sentence, “the device” naturally refers to that printer, so its referent is discourse-old despite the changed wording.
In the third sentence, “the technician” accesses the previously introduced person. “The part” can access the replacement roller introduced in the preceding sentence. These links connect the actions to the appropriate participants and objects.
The final phrase, “another technician,” introduces a different person. The repeated word “technician” does not make that person discourse-old. A naive system that groups all identical occupational nouns would merge two individuals.
The result is an identity-aware reading: one printer, one replacement roller, and two technicians. This count follows the stated toy passage; more complicated texts can leave identity unresolved and should not be forced into a single interpretation without evidence.

Illustrative case study

Fictional case: ServiceThread generates handover notes for repair conversations. Its first extraction method creates a separate record for every noun phrase, producing both “printer” and “device” as unrelated assets. A later pronoun is then attached inconsistently.
Reviewers identify the established asset and link the later description to it. They also test passages with two printers so the system cannot succeed by merging every device reference indiscriminately.
The team records both missed links and incorrect merges as separate errors. This fictional case illustrates why continuity and distinction both matter; it does not claim an observed reduction in repair time or a guaranteed solution to reference resolution.

Common misconceptions and mistakes

Discourse-old does not mean old-fashioned language or an old real-world object. It describes whether a referent has already entered the discourse.
Do not identify old referents through exact string matching alone. Different descriptions can name the same entity, and identical descriptions can name different entities.
Do not assume a definite article or pronoun guarantees an unambiguous antecedent. If several established referents fit, the system may need more context or clarification.

Business use

The concept helps with coherent summaries, conversation memory, and records assembled from repeated mentions. It explains why a system must preserve established identities while avoiding accidental merges.
For business workflows, review examples should include renaming, role descriptions, pronouns, and multiple similar entities. A handover note that loses identity can assign an action to the wrong customer or device even when its individual phrases sound reasonable. Clear discourse boundaries and explicit uncertainty handling make these errors easier to diagnose.

Check your understanding

A paragraph first introduces “a consultant” and later says “another consultant joined.” Should both phrases automatically be assigned to the same discourse-old person?

Answer: No. “Another” signals an additional person in this example. Repeated vocabulary does not establish shared identity.

Related terms

Discourse-new
Coreference chain
Antecedent
Discourse model
information status
referring expression
hearer-old

Learn next

Trace all references to one person and one object through a short report. Replace ambiguous pronouns with explicit descriptions, then check whether the revision preserves identity without creating duplicates or merging distinct entities.

Sources

Sources: Jurafsky and Martin, Speech and Language Processing, Chapter 24, referring expressions and Section 24.1.2 consulted directly: https://web.stanford.edu/~jurafsky/slp3/24.pdf. Established referents, alternative descriptions, inferables and discourse boundaries were independently reviewed. The printer-and-technicians example is original.

Previous: Discourse-new Next: Discourse parsing