How common are indirect requests in English? [research review]
As I’ve written about before, people often make requests indirectly.
But I didn’t put a number on it––not even an order of magnitude. And “often” might mean different things to different people. So how often is often?
Of course, the frequency of indirect requests likely depends on all sorts of factors, many of which have been well-studied: the formality of the setting, the relationship between speaker and listener, the “size” of the request, the urgency of the request, the local culture, the language or language variety being spoken, and so on. One might object that the situation is so complex that it’s almost beside the point to give a general, context-independent frequency estimate––that when it comes to something like human communication, there’s no such thing as “context-independent”.
I’m actually quite sympathetic to this view. At the same time, I think that you can learn something by trying to estimate the frequency of indirect requests, provided your confidence intervals are wide enough and your claims about generalizability are appropriately couched.
So the goal of this post is to report estimates from the relevant literature on indirect requests in English, and settle on some frequency interval. Note that what I’ll be estimating is essentially the probability that a given request is indirect, not the probability of encountering a request more generally. That is, I’m trying to estimate:
p(indirect | request), not
A range of estimates
Frequency estimates obviously vary, sometimes systematically (e.g., whether the requests are naturally-occurring or produced in the lab).
Additionally, the rate of request interpretations depends on how “generous” we are in terms of what counts as indirect. If we count conventional indirect requests like “Can you pass the salt?”, then it tends to be quite high; if we only count oblique, non-conventional indirect requests like “It’s cold in here”, then it’s considerably lower.
Note also that I’ve excluded from this review papers that don’t consider the rate of direct requests; thus, any papers comparing the frequency of different indirect forms are not included.
Gibbs (1981) presented a series of UCSD undergraduates (N = 28) with various written scenarios. Each scenario ended with the protagonist needing to make a request. Participants were asked to pretend they were the protagonist, and to produce five different utterances to get what they needed.
Altogether, subjects generated 2178 requests, each of which Gibbs sorted in one of 13 categories, based on the grammatical construction used. One category was Want/Desire (e.g., “I want…”), another was Ability (e.g., “Can you…”), another was Direct (e.g., “Give me a hamburger”), and so on.
There are a few ways one could determine how many of these requests were indirect. The minimally conservative approach is to count as indirect everything not categorized as direct. Only 5% of the requests were annotated as direct, so that leaves 95% of the utterances.
One objection might be that this estimate includes highly conventional utterances like “Can you…” or “I want…”. The maximally conservative approach would be to include only categories like State of world (e.g., “I’m dying for a cigarette”) and perhaps Embedded requests (e.g., “I was wondering if you could…”), which combined constitute about 13% of the requests produced.
That’s a pretty big confidence interval: 13%-95%. But describing it as an interval is also perhaps misleading, because the estimate depends entirely on what counts as an indirect request. If you want to know how often people used a non-imperative construction, it’s very often (95%); if you want to know how often they used something very indirect like “It’s cold in here”, then it’s a minority of the time, though perhaps still a non-trivial amount (13%).
The other necessary caveat is that this is a laboratory study, conducted on college undergraduates: it may not be representative of how people act in the wild, and the sample may not be representative of the population more generally.
Goldschmidt (1998) sampled 200 “spontaneous favors” from over 100 different people (in American English), in both face-to-face and telephone interaction. In two-thirds of these cases, the author was an “active participant” in these favors, while in one-third, they were an “inactive observer”. After collecting the favors, Goldschmidt annotated them with a variety of taxonomies, including the grammatical construction used, the surrounding conversational context (e.g., the “pre-favor” and its formulation), the relationship between the interlocutors, and so on.
My main interest is the favor requests themselves. Of the 200 requests, 79% used something other than an imperative construction. This included need statements like “I need a ride” (5%), embedded imperatives like “I was wondering if you can give me a ride” (24%), question imperatives like “Can you give me a ride?” (39%), and a couple other categories. Only 3.5% were hints, like “I have to get to the seminar later”.
So again, we have a very large range, which depends entirely on how we define “indirect request”: 3.5%-79%.
Flöck (2016) annotated both naturally-occurring and experimentally elicited requests (in both American English and British English). The naturalistic requests consisted of 520 utterances (260 for each English variety). The other requests were elicited using a “Discourse Completion Task” (DCT), in which participants produced responses to written scenarios (also 520 requests total, 260 for each English variety).
I’ll focus first on the naturalistic data. About 47% of American English requests and 53.5% of British English requests were indirect in some way––i.e., they used something other than an imperative construction. Of non-imperative constructions, interrogatives tended to be slightly more common than declaratives.
The DCT requests exhibited a strikingly different pattern. 97.3% of American English requests and 96.2% of British English requests were indirect in some way; the main difference between the language varieties is that a (relatively) larger portion of the British English indirect requests used a declarative form (despite interrogatives still being the most common in both).
The estimates from the DCT task are pretty aligned with the Gibbs (1981) result––namely, requests elicited in an experimental setting are extremely likely to be indirect (~95-97%). But the naturalistic requests are considerably less indirect (i.e., more direct) on average: between 47%-54% were indirect, depending on the language variety.
Note that it’s a little harder with this data presentation to determine the rate of non-conventional indirect requests. But if we define this as the rate of declarative utterances, then it’s between 26%-32% for naturalistic utterances, and between 2.7%-13.5% for elicited utterances.
Another estimate comes from a recent study I published with Federico Rossano (Trott & Rossano, 2020). I’ve written about this study before, and the main goal was not to assess overall indirect request frequency, but rather to estimate the impact of speaker entitlement on the type of request used. That said, we can still derive some overall frequency estimates across the different experiments used. (Of course, the rate varied a lot across conditions: direct requests were more likely when the speaker’s entitlement was higher.)
The first pair of experiments were very similar to Gibbs (1981): participants (N = 73, N = 72) read a series of passages which ended in a character needing to make a request, and were then asked to generate a request in a free response box. Averaging across the different conditions, 71% of requests were indirect in the first experiment, and 57% were indirect in the second. Very few requests (<3%) followed the “hint” pattern; most were either modal interrogatives (“Can you X”) or embedded modals (“I was wondering if you could X”).
In a third experiment (N = 82), participants selected a possible request from a set of candidates, which varied in the grammatical form used. Again averaging across conditions, the rate of indirect requests was 77%. A minority of requests (~12.5%) followed the “hint” pattern, though more than in the free response study.
This range (57%-77%) is less than the estimates from Gibbs (1981) and Flöck (2016) using similar DCT methods, but higher than the naturalistic requests identified in Flöck (2016). As with the other experimental studies, the obvious caveat is that these requests were not produced in a naturalistic setting.
Below, I’ve combined all these estimates into a single summary table. I’ve tried to include as much relevant information about each estimate as possible, including whether it’s based on naturalistic data, and whether the estimate is for the rate of non-imperative requests (very inclusive), or the rate of non-conventional requests (much less inclusive).
Based on these estimates, we can derive a couple tentative conclusions—with the necessary caveat, of course, that this is a limited sample. (Note that I plan to update the table if I find new estimates from other sources; if you know of any, please let me know!)
First, as you’d expect, the inclusive definition yields a much higher rate than the less inclusive definition. This is obvious, from one perspective—but the size of this difference is also illuminating: in many cases, the rate of non-imperative requests is higher by a factor of at least 7x (the exception is with Flöck (2016), and I’m less confident my operationalization of “non-conventional request” for that study).
|Source||Naturalistic?||Rate of non-imperative requests||Rate of non-conventional requests|
|Trott & Rossano (2020)||No||71%-77%||3%-12%|
* I’m less confident about these rates, because I defined as “non-conventional requests” all the declarative utterances; this would include things like “It’s cold in here”, but it might also include more conventionalized constructions (maybe even something like “I need a jacket”).
Second, in both naturalistic and non-naturalistic settings, the majority of English requests are indirect in some way—meaning they use something other than an imperative construction. Together with the first point, this suggests something like: people rarely use imperative constructions, and instead rely on highly conventionalized forms for making a request. These conventionalized forms are pretty transparent as requests: indeed, the default interpretation of “Can you open that window?” is probably as a request, not as a question.
This is also consistent with the view that conventional indirect requests are essentially grammatical construcitons (Stefanowitsch, 2003). Their purpose is not necessarily to “mask” the speaker’s intent or even signal politeness––rather, they’re simply the default. On this view, using an imperative construction is actually a deviation from the default, and thus signals impoliteness.
Finally, there’s a surprisingly large difference between estimates from naturalistic and non-naturalistic settings. I’m not sure why. If constructions like “Can you X” are the conventional form for making a request, then shouldn’t we see them occur “in the wild” just as much as in the lab? Does this mean that the lab studies are overestimating the degree to which “Can you X” is conventionalized? My intuition is that it should actually go the other way: requests made in the lab have very low stakes, so I wouldn’t expect speakers to feel the need to adhere to politeness norms. On the other hand, perhaps it is consistent with the “default” view of conventionalized indirect requests: if “Can you X” is the default, then maybe it’s just the first form that comes to mind in the lab––and so participants just produce the easiest thing to produce. The implication of that interpretation is that speakers in the wild are thus more likely to deviate from politeness norms by using an imperative construction. Again, I don’t know exactly why this would be: maybe it has something to do with the corpora––if it’s mostly requests made to friends or family members, then imperative forms might be more common.
A meta-point here is that it’s genuinely quite hard to make precise estimates about the rate of indirect requests. The whole point of scientific studies is to derive estimates of population parameters from sample statistics. But as we’ve seen, a lot depends on exactly how “indirect” is defined, and also on the context in which the requests were produced (i.e., in the lab or in the wild).
These estimates are probably reliable enough to make broader statements like “the majority of requests are indirect in some way”, or “a minority of requests use very non-conventional forms”. On the other hand, they’re probably not precise enough for the numbers themselves to be meaningful. For example, if you run a study and find that 85% of the requests were indirect in some way, it’s not clear that this is meaningfully different from the other studies––or if it is, how. It’s lower than some estimates of lab-produced requests (~95%), and higher than others (~77%). Of course, while the overall rates may not be that meaningful across studies, the relative differences can still be meaningful across conditions within a study: hence the benefit of manipulating some key variable (e.g., the relationship between the requester and recipient) and holding all else constant, and asking how that affects the relative likelihood of making an indirect request.
The final point I’d like to make is that, as noted earlier, these are conditional probability estimates: that is, what’s the likelihood of a given request being indirect? This is different from the rate of requests overall, i.e., of all the utterances that one might encounter, what proportion are requests? And it’s also different from the rate of indirect requests overall, i.e., of all the utterances that one might encounter, what proportion are indirect requests? Obtaining these latter estimates feels even harder, because the rate of requesting feels like it would vary hugely across contexts––perhaps moreso than the relative rate of direct vs. indirect requests would vary.
Aijmer, K. (2014). Conversational routines in English: Convention and creativity. Routledge.
Ervin-Tripp, S. (1976). Is Sybil there? The structure of some American English directives. Language in society, 5(1), 25-66.
Flöck, I. (2016). Requests in American and British English: A contrastive multi-method analysis (Vol. 265). John Benjamins Publishing Company.
Gibbs Jr, R. W. (1981). Your wish is my command: Convention and context in interpreting indirect requests. Journal of Verbal Learning and Verbal Behavior, 20(4), 431-444.
Goldschmidt, M. M. (1998). Do me a favor: A descriptive analysis of favor asking sequences in American English. Journal of Pragmatics, 29(2), 129-153.
Stefanowitsch, A. (2003). A construction-based approach to indirect speech acts. Metonymy and Pragmatic Inferencing, (1975), 105–126.
Trott, S., & Rossano, F. (2020). The Role of Entitlement in Formatting Preferences Across Requesters and Recipients. Discourse Processes, 57(7), 551-572.