Language is a central component of human interaction. We use language for everyday conversations, for writing emails, discussing scientific research, composing poetry, and much more. Crucially, the use of language does not occur in a vacuum; language is always produced in a given context, and is always interpreted in a given context. Context includes everything from the speaker’s tone of voice to the entire relationship history between the speaker and listener – anything beyond the words in the actual sentence. The consideration of this context in developing a model of what language means is called pragmatics.
Language is also intentional: speakers produce utterances, or “speech acts”, for a reason. People say things to do things (Austin, 1962). A classical division of the different functions of speech acts is:
- Assertions: statements about the world with the intent of communicating information, e.g. “My name is Sean.”
- Interrogatives: questions addressed to the speaker with the intent of eliciting an informational response, e.g. “What is your name?”
- Imperatives: commands addressed to the speaker with the intent of eliciting an action, e.g. “Close the door.”
The specific reason itself may vary dramatically – perhaps two interlocutors are exchanging gossip to build a social rapport, or perhaps a politician is giving a speech to appease his constituency – but apart from a few exceptions, there is a reason. Somehow, people communicate these intentions with relatively few breakdowns in communication. Thus, the question I’m interested in is as follows: how are speakers able to so frequently communicate their intentions to listeners, and how are listeners so frequently able to infer those intentions? More specifically, how can we represent the computational mechanism behind this inferential process?
Indirect Speech Acts
Remarkably, listeners usually manage to understand speakers’ intentions even when they speak indirectly. An indirect speech act is one in which the literal interpretation of the sentence is not the same as the intended meaning (Searle, 1975). These are particularly common when speakers make requests; one study that elicited over 2000 requests from different participants in different scenarios found that almost 95% of these requests were indirect in some way (Gibbs, 1980). Generally speaking, a request is a speech act intended to cause the listener to execute some action. Thus, a request is a type of imperative. Consider the following ways in which a speaker could request a glass of water:
- Give me a glass of water.
- Please bring me some water.
- Could you get me a glass of water?
- Do you have any water?
- Would you mind getting me a glass of water?
- I’d really like a glass of water.
- Would you get me a glass of water?
- I was wondering if you could get me a glass of water.
- You know, I could really do with a glass of water.
- I sure am thirsty.
All ten sentences are ways in which a speaker in some context could request a glass of water. (1) and (2) are constructed as direct requests, and are clearly commands – there is no ambiguity about their meaning.
(3) through (10), however, are all different types of indirect requests. Some of them, like (3), are so commonly used that they hardly seem indirect at all. But if you consider the literal meaning of (3), the interpretation should be equivalent to: Are you able to get me a glass of water? We know that this probably isn’t what the speaker meant, because a simple “yes” or “no” response would likely not satisfy the intentions of their speech act, and might even irritate them. Similarly, (4) is not simply asking about whether or not the listener possesses any water, but rather making a request for water. Sentences (5) through (8) are also fairly common, but (9) and especially (10) are considerably less direct. Here, we see that indirect requests vary along a scale of conventionality. Many indirect requests are so conventional that we hardly register them as indirect requests, whereas other, less conventional ones involve more nuances involving the context and the relationship between the speaker and listener (Gibbs, 1986).
Why Produce Indirect Requests?
Given that people regularly produce indirect requests, the question that immediately arises is: why? After all, if speakers want to be as informative and efficient as possible with their speech acts (which many models of communication assume they do), it makes more sense to avoid ambiguity, such as the possibility of interpreting an utterance as either a question or a request, as in (4) above. So why use indirect requests?
One reason is politeness. Sentences like “Give me a glass of water” can come off as rude or overly demanding. While politeness may not seem like a relevant first-order goal in communication, it is important to recall that all interactions are situated within some social context. Thus, if a speaker anticipates seeing the listener again sometime in the future – or anticipates that the nature of their interaction will somehow influence future interactions with others, through gossip chains, etc. – then being polite will give off a better impression. Of course, the motivation to be polite is highly dependent on culture, the power dynamic between speaker and listener, and even individual differences between speakers. For example, a boss might feel comfortable giving a direct command to a subordinate, whereas two acquaintances might use indirect requests to achieve the same result. One study actually investigated how polite participants judged different forms of requests to be (Blum-Kulka, 1987); as expected, direct requests were judged as less polite than indirect requests. Interestingly, there was also considerable variation in politeness judgments among types of indirect requests. Conventional requests like “could you bring me some water?” were consistently judged as polite, whereas non-conventional requests like “I’m really thirsty” had much more variation in politeness judgments. This suggests that the conventionality of a request is in some way tied to perceived politeness; coercive “hints” can be seen as impolite because they are so indirect.
I mentioned above that motivations for politeness vary significantly across speakers. One area in which we know this to be true is across genders. Robin Lakoff (1973) famously enumerated the ways in which women’s speech is significantly differently than men’s. Among other things, women are more likely to use hedges (“sort of”, “it seems like”), more likely to apologize when stating an opinion (“I’m sorry, but I think that …”), and most relevantly, use indirect requests; gender differences in indirect request production was confirmed by more recent research (Macaulay, 2001), which also found interesting differences in the types of indirect requests produced across genders (to be covered in a later post, perhaps). Lakoff argues that these differences arise largely out of societal power structures, such as power imbalances between the genders, as well as cultural pressures for women to be more submissive and polite. If a person is consistently taught – whether explicitly or implicitly – that her role is hierarchically below that of a man’s, and that appeasement of others is more important than satisfying her own needs, then it stands to reason that they will feel uncomfortable producing direct commands. Instead, they might produce indirect requests, either of conventional or non-conventional forms, to avoid potential refusals.
This leads into another more general motivation for indirect requests. When a speaker requests something, there is always the possibility that the listener could refuse to fulfill the request. Thus, a speaker produces an indirect request to avoid the possibility of explicit refusal. The difference between “Give me a glass of water” and “Do you have any water?” is that in the latter case, the listener can implicitly refuse the request by treating it as a question, whereas the first offers no such ambiguity. In general, refusal or fulfillment of a request is not necessarily a function of the listener’s willingness to do something. It can be tied to willingness, but it can also be tied to the listener’s ability (e.g. “can you…?”) or whether they possess the item (e.g. “do you have…?”). The construction of an indirect request to address the possible grounds for refusal is called the obstacle hypothesis (Gibbs, 1986), and is a very powerful and general description of the type of indirect request that a speaker uses. Essentially, the argument is that speakers “formulate their requests to anticipate the potential obstacles…which hinder addressees in complying with requests” (Gibbs, 1986).
A similar idea is proposed by Pinker et al (2008), who argues that indirect requests are a form of optimal strategy for rational communication because they allow for plausible deniability. Consider the following scenario:
A driver is pulled over by a police officer for running a red light. As the police officer walks up to the car, the driver wonders whether he can bribe the officer to avoid a more costly ticket. Unfortunately, he doesn’t know whether the police officer is honest or corrupt. If he offers a bribe to an honest officer, he’ll be arrested; but if he offers a bribe to a corrupt officer, he’ll go free. His other option is obviously to not bribe the officer at all; then, regardless of the officer’s honesty, he’ll receive a traffic ticket.
Pinker suggests that the driver can produce an indirect speech act, such as “Maybe we can take care of this here”. This speech act does not overtly mention a bribe, so Pinker argues it is plausibly deniable – an honest cop cannot arrest the man, whereas a corrupt officer now has the opportunity to infer the intention and accept the bribe. Pinker notes that even though this speech act carries a strong implication of a bribe, it still has a nonzero uncertainty value about the driver’s intentions, whereas an overt attempt at a bribe has a zero uncertainty value; thus, if the driver wants to produce a bribe, the cost-benefit trade-off will be better if he produces an indirect bribe than a direct one. Plausible deniability – or more generally, nonzero uncertainty of intentional content – is useful even in non-legal situations, such as navigating ambiguous relationships.
Note that none of these possible motivations – politeness, the obstacle hypothesis, the plausible deniability – are mutually exclusive. To an extent, all describe the same phenomena at different levels of abstraction. For example, politeness is one explanation as to why speakers might address the possible obstacles to compliance rather than giving a direct command. Pinker’s “Rational Briber” proposal is quite similar to the obstacle hypothesis, formalized in a more game-theoretic model. Notably, while Pinker’s model is more formal, it does not explain the mechanism of production. The obstacle hypothesis makes specific predictions about the types of utterances that someone will produce in a given context, whereas Pinker’s model does not. Again, this does not mean they are incompatible; they simply explain different levels of abstraction. Pinker’s model aims to explain indirect speech acts in a game-theoretic, rational framework, and the obstacle hypothesis aims to explain the motivation and content for producing an indirect request in a given context. Pinker’s model is better-suited to models in the economic, rational-agent tradition, and the obstacle hypothesis is better-suited to models for pragmatics, as well as artificial intelligence (e.g. dialogue agents).
So, to summarize, we know that people regularly use indirect or non-literal language when they communicate, and that listeners are still able to infer their intentions. We know that there are a few different possibilities to explain why speakers speak indirectly – politeness, addressing grounds for refusal, plausible deniability – and that these possibilities are all conceptually related.
What we still don’t know is this: given some utterance, how does a listener infer the speaker’s intentions?
This is the question I hope to address going forward. It is essential for both models of human communication, as well as artificially intelligent agents that communicate with humans through language.
Austin, J. (1962). How to do things with words (Vol. 1988). http://doi.org/10.1057/ajcs.2015.8
Gibbs, R. (1980). Your Wish Is My Command: Convention and Context in Interpreting Indirect Speech Acts.
Gibbs, R. W. (1986). What makes some indirect speech acts conventional? Journal of Memory and Language, 25(2), 181–196. http://doi.org/10.1016/0749-596X(86)90028-8
Searle, J. (1975). Indirect Speech Acts. In Speech Acts: An Essay in the Philosophy of Language (pp. 6–6). http://doi.org/10.1145/1045416.1045423
Macaulay, M. (2001). Tough talk: Indirectness and gender in requests for information. Journal of Pragmatics, 33(2), 293–316. http://doi.org/10.1016/S0378-2166(99)00129-0
Lakoff, R. 1973. Language and woman’s place. Language in Society 2: 45-80
Blum-Kulka, S. (1987). Indirectness and politeness in requests: Same or different? Journal of Pragmatics, 11(2), 131–146. http://doi.org/10.1016/0378-2166(87)90192-5
Pinker, S., Nowak, M. A., & Lee, J. J. (2008). The logic of indirect speech. Proceedings of the National Academy of Sciences of the United States of America, 105(3), 833–838. http://doi.org/10.1073/pnas.0707192105