Working with Rhetorical Questions in Discourse: A Practical Guide
Irene Koshik's research on rhetorical questions comes out of conversational analysis and pragmatics, not from a software package or a downloadable dataset. When people search for Rhetorical Questions Irene Koshik, they are usually looking for the analytical framework she developed for identifying and interpreting how rhetorical questions function in spoken interaction. The core idea is straightforward but easy to mess up in practice. Koshik examined rhetorical questions as interactive devices rather than purely syntactic forms. Her work, especially the 2002 book Asking at the Third Position, shows that rhetorical questions often serve to prompt an addressee to perform an action or change a stance, without explicitly stating what that action or change should be. The question looks like a regular interrogative on the surface but operates as something else entirely in context. A common example from her corpus: a parent asks a child who has left shoes everywhere, "Do you think this is a hotel?" The question is not requesting information about the family's living situation. It is a reprimand wrapped in an interrogative form. The child is expected to infer the implied critique and adjust behavior accordingly.
This is different from what you see in traditional grammar textbooks, which usually classify rhetorical questions by their surface structure alone. Koshik's approach requires you to look at the sequential position of the utterance, the relationship between speakers, and what happens immediately after the question is asked.
How to Identify a Rhetorical Question Using Koshik's Framework
Start by looking at the third-position response. In conversation analysis, the turn that follows a question matters a lot. If the expected response is silence, a visible action, or a repair rather than an answer to the literal question, you are likely dealing with a rhetorical question. Koshik found that many of her cases showed the addressee responding by doing what the speaker implicitly asked, not by answering the question on its face value. Check whether the question presupposes an answer that the speaker already treats as established. If the question is built on a premise both participants clearly accept, and the point of asking it is to draw attention to that premise rather than to elicit new information, it is functioning rhetorically. This distinguishes it from information-seeking questions that happen to have obvious answers. Pay attention to prosody and delivery. In real transcripts, rhetorical questions often carry a flat or slightly edged intonation rather than the rising contour typical of genuine requests for information. This is not a reliable rule on its own, but combined with sequential evidence it strengthens the classification.
Get the Full Details
Where the Method Gets Complicated
I ran into a specific problem when coding a corpus of classroom interactions. A teacher asked students, "Did we not cover this yesterday?" At first glance this looked like a straightforward rhetorical question reinforcing that the material had been taught. But one student responded with a detailed account of exactly what had been covered and what had not, effectively treating it as a genuine information request. The teacher then backtracked and clarified what she actually meant. This showed me that the boundary between rhetorical and literal questions is not always clean. The classification depends heavily on how the recipient treats the utterance in real time. My workaround was to code these cases as potentially rhetorical and note the recipient's response type separately. This kept the data honest instead of forcing every ambiguous case into a binary category. Another issue is cross-cultural variation. The same syntactic form can function differently across languages and communities. What reads as a mild reprimand in one setting can read as a direct challenge in another. Koshik's work is grounded in English-language institutional talk, primarily American classroom and counseling contexts. Applying her framework to other speech communities without adjustment will produce inaccurate results.
Common Pitfalls to Avoid
The biggest mistake I see is assuming that any question without an overt answer is automatically rhetorical. Some questions are genuinely ambiguous, and some speakers ask questions they genuinely want answered even when the answer seems obvious to an outside observer. The sequential context is what resolves this, not the surface form. A second pitfall is over-relying on transcription shorthand. If your transcript does not capture pauses, intonation contours, or the timing of actions that follow the question, you will misclassify a significant number of cases. Even basic punctuation in your transcription can change how a rhetorical question reads. I recommend adding a column to your coding sheet specifically for prosodic and post-question action notes, because those details matter more than the words themselves in most borderline cases.
Practical Steps for Analysis
If you are working through a dataset, start with a small set of clear examples to calibrate your ear and your coding criteria. Pick five to ten passages where the rhetorical function is unambiguous, code them fully, and compare your classifications against published examples from Koshik's work. Once you have a baseline, move on to the harder cases. Keep your coding scheme lean. You need at least three categories: rhetorical questions that function as reprimands, rhetorical questions that function as prompts for action, and ambiguous cases that resist classification. Do not create subtypes for every variation you encounter early on. You will refine those later once you have seen enough data to know what actually recurs. Document every decision you make about a borderline case. When you come back to your data months later, you will not remember why you classified a particular utterance the way you did. A brief note about the sequential evidence that led to your choice is worth more than a perfectly formatted spreadsheet with no rationale attached.
Limitations of This Approach
Koshik's framework is powerful for institutional talk where power dynamics and roles are relatively stable. It becomes much less useful in casual peer conversation where participants do not have predefined roles shaping their expectations. In those settings, the line between a rhetorical question and a playful or ironic literal question blurs significantly, and there is no clean methodological fix for that ambiguity. Automated classification tools are also unreliable here. You will find NLP pipelines that claim to detect rhetorical questions from text alone, but they consistently miss the sequential and prosodic evidence that Koshik's framework depends on. Text-only analysis might catch some obvious cases, but it will produce a high false-positive rate in anything beyond very controlled datasets. Manual analysis with careful attention to context remains the only dependable approach. For researchers who need a quicker initial scan before doing deeper analysis, extracting candidate rhetorical questions using basic features like declarative-word order in interrogative form plus absence of a direct answer in the next turn can give you a filtered list to review manually. This usually reduces the workload by about sixty to seventy percent without sacrificing accuracy on the cases that matter most.
Key Resources
The primary source is Koshik's 2002 publication from Stanford University's Center for the Study of Language and Information. Her later articles extend the framework into institutional settings like therapy and media interviews. If you are building a coding scheme around her work, start there rather than relying on secondary summaries, which tend to flatten the nuances that make the framework actually useful in practice.
