Must every request have a presequence?

No. Requests can be made directly; presequences are one way of organizing an interaction.

Can a presequence occur without the main request following?

Yes. The preliminary response may make the projected request unnecessary or unsuitable.

Is a clarification question after a request the same thing?

Not usually. It occurs within the handling of an already initiated request, rather than before that request's first part.

AI Glossary / Natural language and speech / Beginner

Presequences

Definition

Presequences are preliminary conversational exchanges that prepare for a possible subsequent action, such as a request, invitation, offer, or announcement. They are often short adjacency pairs: an initial question or other move followed by a response. Their role is to establish whether or how the projected next action should proceed.

What it means

A person asking whether someone is available may be preparing to invite them somewhere. Someone checking whether a service can perform a task may be preparing to request that task. The preliminary exchange has an immediate function, but participants can also recognize its relevance to what might come next.

The later action need not actually occur. A response can make it unnecessary, inappropriate, or impractical. Calling something a presequence therefore does not mean that a completed main request must appear later in the transcript. The interpretation rests on how the exchange functions in its conversational setting.

How it works

An initial move projects a possible activity and makes a response relevant. The response can encourage continuation, indicate a difficulty, or even supply an offer that removes the need for the anticipated request. Conversation analysts examine the sequence of turns and participants' displayed understanding rather than treating every opening question as the same kind of presequence.

A presequence differs from a clarification inserted after a request has already been made. For example, asking which file someone means after they request a file is part of handling an existing request. Checking whether files can be shared before making a particular request can instead be preliminary to that request.

For dialogue-system design, a useful consequence is to preserve the distinction between a capability question and an instruction to act. A system can answer the preliminary question and invite the user to specify what they need. It should use subsequent context to update its interpretation rather than assuming that an anticipated action has already been authorized or fully specified.

Three practical examples

1. Customer support: A user checks whether an assistant can locate an archived invoice before identifying the invoice they want retrieved.

2. Scheduling conversation: One person asks whether a colleague has time later, and the reply determines whether an invitation to review a draft follows.

3. Dialogue annotation: Analysts label the preliminary exchange separately from the main request so that evaluation does not confuse answering a capability question with completing the requested task.

Think of it this way

Think of a presequence as checking whether a conversational doorway is open before stepping through it. The reply helps shape the next move. The analogy is limited because conversation is not a fixed gatekeeping procedure: people can infer, redirect, offer, or decline in ways that change the anticipated activity.

Worked example

Consider this original fictional four-turn exchange:

User: “Can this assistant find older invoices?”
Assistant: “Yes, it can search the archived invoices available to your account.”
User: “Please find the invoice for order 317.”
Assistant: “I found the matching invoice; here is the document.”

Turns 1-2 form the preliminary capability exchange. Turns 3-4 form the main request-and-response pair in this imagined successful interaction. There are four turns and two adjacency pairs; the capability question is not itself the specific request for order 317.

Now change turn 2 to “No, this assistant only searches current invoices.” The user might then stop or choose another route, so turns 3-4 need not follow. This alternative demonstrates that the presequence can affect whether the main action occurs. Neither transcript records an actual tool call, invoice retrieval, or service capability. The examples illustrate sequence structure without claiming that a real assistant performed the actions.

Illustrative case study

Fictional case: Juniper Support evaluates a chatbot that treats every question about a feature as a command to use it. Reviewers observe that capability inquiries receive premature task-completion messages even when the user has not specified the relevant item.

The team separates capability responses from action requests in its test dialogues. It includes both conversations that proceed to a task and ones that end after a limitation is explained. The case illustrates a design correction and an evaluation distinction, not a measured improvement or a report about a real product.

Common misconceptions and mistakes

Do not equate a presequence with any sentence appearing early in a conversation. Its relation to a projected action matters. A standalone factual question can remain just that, and an invitation can be made directly without a preliminary exchange.

Also avoid inferring an unspoken request with complete certainty from a short availability question. Context and the response matter. In an AI system, recognizing a possible next action should help the conversation progress, not justify claiming that a task was requested or completed before the necessary information arrived.

Business use

For business assistants, test whole conversational paths rather than isolated intent labels. Include capability checks, responses that permit continuation, and responses that lead users to change direction. Evaluate whether the assistant answers the current move appropriately while remaining ready for the likely next one, and whether its completion claims correspond to actions actually performed.

Check your understanding

In the invoice example, when does the user make the specific request for order 317?

Answer: In turn 3. Turns 1-2 establish capability; they do not identify that order or constitute a completed retrieval.

Related terms

Adjacency pairs
Dialogue act
Conversation analysis
Adjacency pair
pre-expansion
clarification question
dialogue management
conversational context
turn-taking

Learn next

Learn adjacency pairs, then compare preliminary exchanges with clarifications inserted inside a main exchange. Annotate complete short dialogues, including cases where the projected action never occurs. Finally, test how a dialogue system distinguishes recognizing a possible intention from receiving a sufficiently specified request.

Adjacency pairs

Sources

Emanuel A. Schegloff, Sequence Organization in Interaction, chapter 4, publisher-provided pre-expansion extract consulted: https://www.cambridge.org/core/books/sequence-organization-in-interaction/276CD30E23D3444114A90E5E2B24D55F/listing?pageNum=1

Requests, Blocking Moves, and Rational (Inter)action in Survey Introductions, primary research discussion of blocking moves: https://pmc.ncbi.nlm.nih.gov/articles/PMC3116201/

Jurafsky and Martin, Speech and Language Processing, archived dialogue chapter: https://web.stanford.edu/~jurafsky/slp3/old_sep21/24.pdf

The invoice dialogue and company case are original fictional illustrations. The complete Schegloff book was not inspected.

Previous: Presentation Bias Next: PRESS statistic