AI Glossary / Natural language and speech / Intermediate

Discourse deixis

Definition

Discourse deixis is reference to some aspect of a stretch of discourse, often using expressions such as “this” or “that.” The referent may be a claim, description, proposal, or situation conveyed by earlier language, rather than a person or object named in it. Resolving the reference requires understanding the surrounding discourse.

What it means

Suppose someone says, “The supplier has cancelled the shipment. That is unexpected.” The word “that” naturally points to the reported cancellation, not simply to the supplier or shipment as an object.
This is different from a straightforward entity reference such as “The supplier called. It confirmed the order,” where “it” refers to the supplier. The distinction concerns what is being referred to, not a rigid rule about which pronoun appears.
Discourse-deictic references can package a substantial amount of prior content into a short expression. A system may need to recover both the relevant span and the aspect being discussed, such as whether the speaker is evaluating a proposition or commenting on how it was expressed.

How it works

An analysis first identifies an expression whose meaning depends on discourse context. It then considers candidate spans or interpreted content that the expression could target.
The surrounding predicate helps constrain the referent. “That was false” invites a truth-evaluable claim; “That was an awkward way to explain it” concerns a manner of expression. The closest preceding noun is not necessarily suitable for either interpretation.
The system must also preserve uncertainty when several readings remain plausible. A reference may concern one clause, a whole explanation, or a combined passage. Annotation schemes and theoretical accounts differ, so the scope and target representation should be stated explicitly.

Three practical examples

  • A meeting summarizer resolves “That would delay the launch” to the proposal being discussed, rather than attaching the comment to the last person named.
  • A support assistant interprets “That did not solve it” in relation to the preceding troubleshooting action and reported problem.
  • A document-review tool links “This is inaccurate” to the claim under review so that a correction does not appear to criticize an unrelated entity.

Think of it this way

Think of pointing at a whole paragraph on a page instead of pointing at one noun inside it. The gesture resembles how “this” or “that” can select prior discourse. The analogy is limited because the intended referent may be the paragraph's claim or implication, rather than the printed words themselves.

Worked example

Consider this invented exchange. A: “We can release the update on Monday if the accessibility review finishes on Friday.” B: “That is too risky.”
One possible target of “that” is the conditional release plan as a whole. Another plausible reading is the dependence on finishing the review so close to release. It is unlikely to mean that Monday, considered merely as a day, is intrinsically risky.
A system that must create an action item should not silently choose a highly specific interpretation without enough context. It could ask, “Do you mean the Monday release date or relying on Friday's review?”
Now change B's response to “That sentence is too long.” The target is explicitly the linguistic formulation. Change it to “That would leave no time for fixes,” and the likely target becomes the proposed timing arrangement and its consequence.
The exercise shows how the following words constrain the reference and why a nearest-noun rule fails. It also shows that a reasonable interpretation can remain uncertain until the conversation supplies more evidence.

Illustrative case study

Fictional case: MinuteMap generates decision notes from planning meetings. In one test transcript, a participant proposes outsourcing a task and another says, “That needs a separate approval.” The initial summary attaches the approval requirement to the contractor's name rather than the outsourcing proposal.
Reviewers revise the annotation to mark the proposed action as the referent. The team adds examples in which short demonstratives refer to a plan, claim, or explanation spanning more than one clause.
For unresolved cases, the summary preserves an explicit uncertainty note instead of inventing a precise target. This fictional case illustrates a reference-resolution failure and a review strategy; it does not report measured accuracy or establish a universal parsing method.

Common misconceptions and mistakes

Do not assume that “this” and “that” always indicate discourse deixis. They can also point to physical objects or refer to previously mentioned entities.
Do not force every reference onto a noun phrase. The target may be content expressed by an entire clause or passage.
Avoid confusing reference resolution with truth assessment. Identifying the claim meant by “that is false” does not independently prove the claim false; it explains which claim the speaker is rejecting.

Business use

Discourse deixis matters for meeting notes, conversational assistants, document feedback, and any workflow that converts vague references into explicit records. Errors can attach a criticism, approval requirement, or consequence to the wrong target.
A useful evaluation asks reviewers to identify the referenced content and the degree of ambiguity. It should distinguish clear cases from cases that require more context. That distinction helps teams choose when to rewrite a reference explicitly and when to ask for clarification rather than overstate the system's understanding.

People also ask

Does discourse deixis always refer to the previous sentence?

No. The relevant content can be a clause, several sentences, or another accessible discourse segment.

Is it the same as ordinary pronoun coreference?

It is related, but the target can be a proposition, description, or other discourse-level content rather than a previously named entity.

Can “it” participate in this kind of reference?

Yes. The phenomenon is defined by its referent and context, not by an exclusive list of demonstrative words.

Check your understanding

After a long proposal, a participant says, “That is impractical.” Why might linking “that” to the final noun in the proposal be wrong?

Answer: The participant may be evaluating the entire proposal or a particular aspect of it. The intended target must be inferred from the discourse, and ambiguity may require clarification.

Related terms

  • Anaphora
  • Coreference resolution
  • deixis
  • discourse segment
  • proposition
  • reference resolution
  • demonstrative

Learn next

Compare examples where “that” refers to an object, a reported event, and an entire claim. For each, write an explicit paraphrase of the referent and note what additional context would resolve any ambiguity.

More in Natural language and speech

  • Discount - In count-based language modeling, a discount reduces the probability mass assigned to observed events so that some mass can be allocated elsewhere...
  • Discourse - Discourse is connected language interpreted as a structured whole, such as a conversation, report, story, or article.
  • Discourse connectives - Discourse connectives are words or phrases that signal relationships between parts of a text or conversation.
  • Discourse-new - Discourse-new describes a referent that is being introduced into the current discourse rather than already being established in it.
  • Discourse-old - Discourse-old describes a referent that has already been established in the current discourse and is accessed again by a later expression.
  • Discourse parsing - Discourse parsing is the computational analysis of relationships and structure across parts of a text.

Sources

Sources: Bonnie Lynn Webber, Discourse Deixis: Reference to Discourse Segments: https://aclanthology.org/P88-1014.pdf ; Jurafsky and Martin, Speech and Language Processing, Coreference Resolution and Entity Linking: https://web.stanford.edu/~jurafsky/slp3/24.pdf

Previous: Discourse connectives Next: Discourse model

  • Home
  • Daily AI Updates
  • Blog
  • AIFBL Academy
  • AI Glossary
  • Products and Tools
  • Store
  • Free CFO AI Prompt Pack
  • About Shihan Sheriff
  • Privacy Policy
  • Terms of Use
  • Refund Policy
  • Cookie Policy
  • Earnings Disclaimer