Undoing things with words

Abstract

Over the last five decades, philosophers of language have looked into the mechanisms for doing things with words. The same attention has not been devoted to how to undo those things, once they have been done. This paper identifies and examines three strategies to make one’s speech acts undone—namely, Annulment, Retraction, and Amendment. In annulling an act, a speaker brings to light its fatal flaws. Annulment amounts to recognizing an act as null, whereas retraction and amendment amount to making it null. Speakers employ retraction to cancel the deontic updates engendered by a given act. They instead use amendment to adjust its degree of strength. I will argue that annulling, retracting, and amending are second-order speech acts, whose felicity conditions vary with the type of illocution they operate on. Undoing is therefore conceived of as a form of doing. Furthermore, I claim that, in calling off our acts, we undo the conventional or illocutionary effects of our words while leaving intact their past causal or perlocutionary outcomes.

1 Introduction

Starting with the pioneering work of Austin (1962), several philosophers of language have turned their attention to speech acts and their conditions for success [cf., among many others, Strawson (1964), Searle (1969, 1975a, b), Bach and Harnish (1979), Searle and Vanderveken (1985), Alston (2000)]. In spite of providing an articulate framework for how to do things with words, the literature almost completely lacks discussion on how to undo what we have done in uttering certain expressions.¹ This paper sets out to identify and examine the most common strategies to make one’s speech acts undone. In particular, it is an examination of three ‘undoing strategies’—namely, the Annulment Strategy, the Retraction Strategy, and the Amendment Strategy.

Austin himself pointed out that speech acts may be ‘annulled’ if certain conditions are found not to be fulfilled. To understand what he had in mind, a simple example may help. Imagine that, after a two-step online interview, Bob is told that he got the job. Being very happy with that, he attends in person at the HR Department the next day, but they regretfully inform him that, unbeknownst to the recruiter, that job position had already been filled. Since the act of hiring someone is dependent on a specific position being actually vacant, the recruiter’s act—at first regarded as felicitous both by the speaker and the hearer—is now to be declared null and void. It is worth underlining at the outset that Austin’s use of the verb ‘undo’ is rather misleading, for it suggests that the speech act that gets ‘undone’ was once done, but annulments operate on fatally infelicitous acts that had never been really done. Annulling an act amounts to recognizing that its validity was only purported. However, as I will claim, speakers have at their disposal other strategies to undo things stricto sensu—i.e., to cancel or adjust acts they have carried out with success. Amongst them are what I call the Retraction Strategy—by means of which we can make the normative statuses introduced by our illocutions vanish—and the Amendment Strategy—which is employed to alter the illocutionary force of one’s previous acts (specifically, by tampering with its degree of strength).

Clearly, in order to give a full picture of ‘how to undo things with words’, one should not only single out the main communicative strategies to rescind one’s illocutions, but also clarify whether and to what extent the type of the involved act (be it verdictive, exercitive, or other) as well as further elements (e.g., the context, the relationship between the participants) bear on their success. For instance, whether or not a certain promise can be successfully taken back seems to depend (among other things) on how serious the promisor’s reasons for retraction are as well as on whether the promisee has granted her the right to retract commitment. Such constraints on retraction do not hold in the case of, say, assertions or orders. Given the preliminary character of my analysis, I will leave many of such complexities aside.

In what follows, I outline a number of considerations to initiate the exploration of the mechanisms to undo speech acts. Since the phenomenon of ‘undoing’ clearly stands as the flip side of Austin’s account, it is first necessary to outline the requisite background in speech act theory.

2 Background: speech acts, the doctrine of infelicities, and illocutionary effects

It is very well known that Austin distinguished between three broad kinds of speech acts, to wit, locutionary, illocutionary, and perlocutionary acts. An utterance of a meaningful sentence—i.e., the performance of a locutionary act—by the right speaker in the right context (e.g., “I pronounce you husband and wife”, uttered by the officiant during a wedding ceremony) counts as the performance of an illocutionary action (e.g., the act of marrying) and produces some perlocutionary upshots (e.g., the belief of the attendees that the parties got married, the joy of the groom’s mother, and so forth). Illocution is the core of speech act theory. While ‘locution’ and ‘perlocution’ capture what one says and what one causes (or may be held responsible for causing) by saying it, respectively, the notion of ‘illocution’ accounts for what one does, thereby enshrining the action potential of our utterances.

Austin introduced three sets of rules whose violation makes an illocutionary act infelicitous. An attempted illocution is void or misfires if the invoked conventional procedure does not exist or an existing procedure is invoked by inappropriate persons or in inadequate circumstances (A rules). Further fatal infelicities involve the execution of the procedure—as when it is carried out incorrectly or incompletely (B rules). The act is instead an abuse of the procedure (but still effective) when the flaws concern the speaker’s mental states (e.g., condoling with no regret) or her subsequent behaviors (e.g., advising the hearer to do $\phi$ and then reproaching her for doing it) ($\varGamma$ rules). As I remark later, felicitous and infelicitous acts mobilize different undoing strategies. In particular, the Annulment Strategy is employed to undo void acts erroneously held to be valid up to a certain point, while the applicability of both the Retraction and the Amendment Strategies presupposes that an illocutionary act was achieved (even though it was possibly achieved in some abusive ways).

In Austin’s framework, illocutionary acts also engender three kinds of effects, viz. they (i) produce the hearer’s comprehension of the meaning and the force of the locution (i.e., illocutionary uptake); (ii) ‘take effect’ in conventional (as opposed to natural) ways; and (iii) invite specific responses on the recipient’s part (as in the case of requests, which invite the recipient to grant or refuse them²). For my present purposes, it will suffice to focus on the second of those effects. To clarify what it consists of, Austin relied on the act of christening a ship, which brings it about that the ship is given a certain name and, thereafter, a number of acts (such as referring to it with a different name) will be improper (Austin 1962, p. 116). The christening of a ship brings off a shift in norms: as soon as the ship is felicitously baptized ‘Queen Elizabeth’, speakers will ought to refer to it with that particular name.

Two caveats are in order here. First, such an effect generalizes to all illocutionary types—that is, it is proper of both ritual acts (such as christening) and communicative acts (such as ordering or even asserting).³ It is essential to the notion of illocution that it ‘takes effect’ in the sense of reshaping the normative context it occurs in: if one is successfully ordered to do $\phi$, one forthwith gets a duty to do it which was absent from the context until a moment before; and once I assert that P, I commit myself to the truth of P (and to make manifest my reasons in favor of it in case I am asked to), thereby introducing new normative constraints in the world of utterance.

Second, Austin’s effect (ii) does not flow into the perlocutionary. Perlocutionary acts are the production of changes in the natural course of events, as opposed to illocutionary acts, which are the bringing about of deontic states of affairs. Consider the following example. The NYPD is investigating a complex case of political corruption. The police commissioner demands from the detective on the case the utmost confidentiality:

• (1) No comments, John. That’s an order.

Given that the police commissioner is in a position to issue orders to her subordinates, her illocution succeeds—and John ipso facto acquires an obligation to avoid certain comments. This does not mean that the intended perlocutionary effect will follow; after all, John may always neglect his duty and comment on the investigation. In order for the police commissioner to successfully accomplish the perlocutionary act of deterring John’s behavior, he will have to align to her will—which is clearly something over and above his recognition that her words impute a duty on him.⁴

To put it more generally, perlocutions are the causal [or ‘normal’, says Austin (1962, p. 116)] production of consequences on the addressee’s thoughts, feelings, and behaviors, which largely rest on external contingencies, for a speaker can never predict her audience’s exact reaction. Illocutionary acts, by contrast, amount to the assignment of new deontic statuses to the participants in the interaction. Illocutions conventionally bound the interactants together in new normative relationships comprising certain rights, duties, entitlements, commitments, and so on, which cannot fail to obtain unless the act misfires [cf. Sbisà (1984, 2014)].

The distinction between conventional or illocutionary effects and causal or perlocutionary consequences is central to our discussion. Indeed, in calling off our speech acts, we annul, retract, or amend the illocutionary effects of our words while leaving intact their past perlocutionary outcomes.⁵ One can, for example, take back a marriage proposal, but one cannot take back the hearer’s excitement at hearing the words “Will you marry me?” (nor one can make the locution itself disappear, for once uttered, it will sit in the world as a mere physical fact). Similarly, I can apologize for making you cry by saying something offensive and indeed retract my statement, but I clearly cannot cancel the tears you have shed. Once produced, non-conventional, material effects cannot be recanted, nor can they just turn out to be null and void. Conventional states of affairs are instead liable to be annulled (should some fatal flaws being disclosed), as well as to be retracted or adjusted on the fly. Just as one can, under appropriate circumstances, illocutionarily engender local obligations or confer somebody certain rights, so too one can, under adequate conditions, wipe those obligations and rights away by means of further (second-order) illocutions. (More on the second-order character of such illocutions below.)

A terminological note: even though Austin calls both the illocutionary and the perlocutionary ‘effects’, only the perlocutionary are real, consequential effects.⁶ Illocutionary effects are to be understood as normative outputs which obtain by default, provided that certain felicity conditions hold. The illocutionary effects of a speech act are determined by its force and intrinsic to its felicity: once the act is felicitously performed, those effects will automatically enter the context. This is not the case for perlocutionary effects, which are virtually unpredictable and largely independent from the felicity of the act they are effects of.⁷

3 Undoing things

It is to be noted that nothing that is past is an object of choice, e.g. no one chooses to have sacked Troy; for no one deliberates about the past, but about what is future and capable of being otherwise, while what is past is not capable of not having taken place; hence Agathon is right in saying: ‘For this alone is lacking even to God, To make undone things that have once been done’ (Aristotle, Nicomachean Ethics, VI, 2, Eng. transl. by D. Ross, Oxford: OUP, 1925).⁸

Within certain limits, our future seems open and pliable. The same cannot be said for our past. Suppose that it is noon and that you have a (presumably) very boring lunch at your great-aunt Betty’s at 12:30. Jimmy, a good friend of yours, asks you to have lunch together at a lovely Japanese restaurant. It is in your power to accept or reject Jimmy’s proposal. However, if you opt for having lunch at your great-aunt’s and, at 12:45, bored already, you regret it, there is hardly anything you can do about it. You can obviously get up from your seat and try to catch Jimmy at the Japanese restaurant, but you cannot bring it about that you never went to your great-aunt’s place at 12:30. Since the past is not open, nobody can ‘make undone things that have once been done’ (as Agathon reminds us). And yet, as Austin suggested, illocutionary acts appear to invalidate, at least to some extent, the logic of the ‘closed past’, for they may turn out to be null if certain conditions are discovered not to be satisfied. A speaker makes a certain utterance, its conventional effects are supposed to enter into force, and indeed they are acted upon up to a certain point; but once that utterance is found to be fatally flawed, the involved parties cannot but realize that it was a mere purported act and agree about its nullity. The past, according to Austin, is ‘undone’ only in the sense that «it is made different to how it would otherwise have been» (Langton 2018).⁹ Annulments of speech acts change the past from the way it was mistakenly taken to be (and would have been, were certain conditions fulfilled) to the way it actually was.

I will argue that, even though Austin only mentions the possibility of nullifying a certain act, many illocutions are also cancellable—that is, they can, once performed, be recanted, rectified, or suspended by means of illocutionary glosses or further acts by the utterer. In this section, after saying something more on the undoing strategy already envisaged by Austin (i.e., annulment), I delve into an analysis of two further mechanisms to backpedal on one’s speech acts (i.e., retraction and amendment).

One more remark before we proceed. When one calls off an illocution, what one does is an exercitive speech act whose internal purpose is to cancel or alter the conventional effects of that illocution.¹⁰ Exercitives are acts that enact permissibility facts [cf. McGowan (2004)]. A speaker who voids or takes a given act back is removing the deontic update associated with that act from the conversational context. In doing so, she shifts the boundaries of what is permissible in that context and hence performs an exercitive (for example, if I successfully retract a promise to do $\phi$, then blaming me for not doing $\phi$ will become impermissible). Undoing is, thus, a form of doing. If so, annulling, retracting, and amending, like every speech act, must comply with certain felicity conditions. However, an issue occurs when we try to spell them out. Since undoing acts are second-order ones, their felicity conditions cannot be defined once and for all, for they will vary with the type of illocution the act operates on. We can nevertheless pinpoint the general felicity conditions of the strategies at issue—i.e., those conditions which are common to all speech acts with the same undoing force.

3.1 Annulling

The Annulment Strategy applies to fatally infelicitous acts that were mistakenly taken as felicitous and whose deontic effects were deemed to be binding for a while due to the ignorance of the involved parties. To get a concrete idea of what such a strategy amounts to, consider the example that follows.

The Suffolk County couple’s case: In 1986, a Suffolk County couple got married by a friend who had been mail-ordained by the Universal Life Church. Having heard of other couples turning to friends or relatives to officiate their weddings, they assumed it to be legal. However, a few years later, the Appellate Division of the Supreme Court ruled in their divorce case that their marriage and prenuptial agreements were void, because a minister whose «authority rests solely on his having obtained in the mail a card entitled ‘Credential of Ministry’ [...] does not qualify for licensing to marry» [Ranieri v. Ranieri, 146 A.D.2d 34 (1989)].

Since the act of marrying can be felicitously performed only by an authorized minister, and ULC ministers’ qualifications did not meet the required standards under the marriage laws in force in 1980s New York State, the Suffolk County couple got only purportedly married. The ceremony was nonetheless taken as a valid marriage (and thus as engendering an update of the couple’s marital statuses) up to the Supreme Court’s decision about its nullity. The case illustrates that, while several attempted acts are immediately recognized as not complying with their success conditions, there are also illocutions whose infelicity goes unnoticed. I would not speak of them as misfires but as annullable (or voidable) acts. An annullable act is a flawed act that gets mistaken for a valid one and whose conventional effects are alleged to enter into force. This gives them causal efficacy for as long as the deficiencies of the act stay buried. A misfire is instead a mere attempt that has no efficacy whatsoever, for its irregularity is manifest from the very moment the utterance is spoken. (Imagine that Janet’s uncle Jerry, drunk, takes the floor in the middle of her thesis defense to confer the title of Doctor upon her. That would be an undeniable case of misfire.) In spite of being infelicitous, the Suffolk County couple’s marriage was at first registered as a lawful wedlock, and the illocutionary effects it allegedly engendered passed as binding. The couple’s rights and duties got prima facie updated, and this allowed them, say, to open a joint bank account or to receive the ‘family rate’ on health insurance. Such possibilities would have never opened up for them had the act merely misfired. Clearly, not all annullable acts are in fact annulled. There are plenty of cases where the infelicity of a certain act is never detected or is detected too late. Imagine that an individual is found guilty of murder and sentenced to the death penalty. However, after the execution, the sentence is annulled because of discovered errors in the trial procedure. Albeit it may serve some purposes (e.g., restore dignity to the defendant’s name or flesh out anti-death penalty arguments), such an annulment seems of little significance since the (alleged) illocutionary effects of the act remained in force for as long as they had an impact on the defendant’s life.

It is no coincidence that the Suffolk County couple’s case concerns an act of marrying and my last example has to do with an act of condemning. The Annulment Strategy is indeed typically employed to undo institutional or ritual acts (e.g., marrying, christening, condemning), whose annulment goes through a formal ratification of nullity. Notice, however, that all unsuccessful acts are, in principle, annullable. If you invite me to spend the summer at your house in Miami Beach, but afterwards it turns out that you do not have a house there, then your invitation will be voided; and if I offer you a slice of cake but then find out that my kids have already finished it, then my offer will be dismissed as null. It is worth underlining that no speech act of annulling is required for a communicative act to be annulled. After realizing that there is no cake left, I can annul my offer by saying something like “Forget about what I said. No cake”, but it may also be the case that, when I open the fridge, both of us see that the cake plate has just a few scattered crumbs on it. In such circumstances, my offer seems to simply default. There is no need for any of us to illocute; in the light of new shared evidence, a tacit recognition of nullity is enough. On the contrary, annulments of institutional acts require some arbitrator to ascertain the facts and perform an official annulment act.

The above examples point out that annulment cases—be they institutionally bound or not—are very often related to pragmatic and/or propositional presupposition failure. In officiating a marriage, for instance, the speaker pragmatically presupposes that she occupies a certain position within a religious or civil institution that empowers her to conduct wedding ceremonies. When this presupposition is not met (as in the Suffolk County couple’s case), the act is a good candidate for annulment. Similar considerations can be drawn for those acts whose propositional presuppositions do not obtain. When you invite me to your house in Miami Beach, you are therewith presupposing that you do have a house in Miami Beach. If this is not so, then your invitation is voidable.

What we have said thus far provides us with enough basis for a presentation of the felicity conditions of the illocutionary act of annulling. As already mentioned, I will only deal with those conditions that are shared by all instances of annulment (no matter the illocutionary type of the act to be annulled). Following Searle and Vanderveken’s terminology, I call them general conditions of annulment.¹¹

General felicity conditions of annulment: a speech act A performed by a speaker S at a time t may be annulled at a later time $t_{{n}}$ only if

• i. A was infelicitously performed at t;

• ii. A was not already annulled at any time between t and $t_{{n}}$.¹²

As one can see, i. and ii. are shaped as necessary but not sufficient success conditions. One cannot perform an act with the force of annulling without i. and ii. being fulfilled, but for the success of a specific annulment they must be supplemented with some special conditions. Such special conditions are dependent on the type of illocution to be annulled. For example, for an institutional act being annulled, the Annulment Strategy must be deployed by a speaker endowed with the appropriate sort of authority (e.g., a Catholic marriage can be annulled only by the Roman Rota, and a federal law can be declared unconstitutional and therewith voided only by the Supreme Court).

3.2 Retracting

Nigel Farage’s decision to ‘unresign’ in May 2015 offers a good example of how the Retraction Strategy works. As Steve Crowther, chairman of the UK Independence Party, declared:

As promised Nigel Farage tendered his official resignation as leader of UKIP to the NEC. This offer was unanimously rejected by the NEC members who produced overwhelming evidence that the UKIP membership did not want Nigel to go [...]. On that basis Mr Farage withdrew his resignation and will remain leader of UKIP. In addition the NEC recognised that the referendum campaign has already begun this week and we need our best team to fight that campaign led by Nigel. He has therefore been persuaded by the NEC to withdraw his resignation and remains leader of UKIP.¹³

To provide a bit of background: as promised in his book (The Purple Revolution), Farage stepped down as UKIP leader on 8 May 2015, following his electoral defeat in South Thanet. Three days later, however, he withdrew his resignation—as, in fact, he had not promised. The first thing to note is that, if the Retraction Strategy is a means for canceling the conventional effects of one’s acts (as I claim), then a successful retraction should restore the normative state of affairs that preceded the performance of the retracted illocution.¹⁴ This means that, by retracting his resignation, Farage broke his promise to step down—even if he did quit after losing the election.

The second thing to underline is that, in fact, the party’s NEC was not in a position to reject Farage’s resignation. To put it differently, no official acceptance from NEC members was needed for Farage’s resignation to take effect.¹⁵ Instead, NEC members were in a position to recommend that Farage retraced his steps—and this is what they have supposedly done. This remark is significant for our discussion, since Farage’s unresignation constitutes an act of retracting only if the resignation was successful in the first place. (It would have been a misfire, had its performance been conditional on NEC members’ acceptance.) While we annul infelicitous acts whose effects were deemed to be binding because of the ignorance of the conversational participants, we retract felicitously performed acts whose conventional effects truly entered into force.¹⁶ The very difference between annulling and retracting is that, though both strategies end up making the act’s (real or alleged) normative effects ineffective, in retraction cases the act is made null, whereas in annulment cases it is recognized as null.

Retractions’ internal goal is to cancel the deontic updates generated by some past illocution, or in MacFarlane’s words, «to ‘undo’ the normative changes effected by the original speech act» (MacFarlane 2014, p. 108). Notice that, while we often retract our acts by means of utterances like “I take that back” or “Scratch that”, retractions do not scratch one’s illocutions from the conversational record. The original speech act was successfully performed at the time of utterance, and that performance cannot be retroactively removed from the record. Nor can we magically change the fact that the act’s conventional effects were valid up to the time of retraction. What we can do is instead render those effects no longer valid (and, thus, causally ineffective) from then on.¹⁷

Though the literature on retraction focuses almost exclusively on assertives,¹⁸ the target of a retraction is not restricted to assertions and the like. One may retract promises, offers, orders, advice, and so forth. It is very often the case, for example, that speakers retract promises once they realize that they will not be able to keep them (“My car broke down, so I can’t pick you up, I’m sorry”). Similarly, advice is likely to be taken back if the utterer finds out that the state of affairs that P is actually bad for the hearer (“Don’t quit your job right now! I didn’t know you are flat broke”). But one can also retract more institutional acts (as Farage’s case shows)—and even fully-fledged ritual acts, as when two spouses obtain the dissolution of their ratum sed non consummatum marriage.¹⁹

One of the very few explicit accounts of retraction in the literature is offered by Bach and Harnish (1979). On their view, (claim) retractions are a subtype of constatives—i.e., acts that express both a belief and an intention that the hearer form a like belief.

In uttering e, S retracts the claim that P if S expresses:

• i. that he no longer believes that P, contrary to what he previously indicated he believed, and

• ii. the intention that H not believe that P (Bach and Harnish 1979, p. 43).

This account does not seem right, though. The above conditions are shaped as (necessary) sincerity conditions for the successful performance of an act of retraction.²⁰ However, as MacFarlane (2011, p. 83ff) suggested, retracting an assertion whose content one still believes in poses no insincerity problems. The utterer might do so because she realizes that she does not have enough ground to assert that P, or because she does not want the hearer to rely on her words.²¹ Furthermore, retractions do not necessarily express an intention that the hearer not believe what was asserted—the speaker may have nothing against the audience believing it, provided that they rely on different grounds than the retracted assertion. Notice that Bach and Harnish’s proposal appears to be equally flawed once we extend it to speech acts other than assertions. Consider, for example, the act of requesting (directive), which they conceive of as expressing a desire and an intention that the hearer fulfil that desire (Bach and Harnish 1979, p. 47). Following the above model for claim retraction, one should conclude that request retractions are subject to the following rules.

In uttering e, S retracts the request that H do A if S expresses:

• i. that she no longer desires that H do A, contrary to what she previously indicated she desired, and

• ii. the intention that H no longer take S’s desire as a reason to do A.

It is not hard to see why such a proposal would not work. First of all, a speaker might retract a request because she realizes that it was not nice of her to issue it, or that the request was too demanding for the hearer. Thus, it seems wrong that request retractions express lack of desire. Nor do they necessarily express an intention that the hearer no longer take the speaker’s words as a reason to act—after all, a speaker who desires that the hearer do something might be quite happy with her doing it, whatever her reasons may be.

In the light of the above considerations, let us leave behind Bach and Harnish’s proposal (which, anyhow, was not meant to be as general as to account for the retraction of non-assertive speech) and try to move a step forward by sketching the general felicity conditions of retraction.

General felicity conditions of retraction: a speech act A performed by a speaker S at a time t may be retracted at a later time $t_{{n}}$²² only if

• i. A was felicitously performed at t²³;

• ii.

  • (a) A was performed by the retractor, or

  • (b) the retractor is a third party whose position grants her the right to take back A.

Conditions ii.(a)–(b) point at a further difference between annulling and retracting. In brief, while there are no general felicity conditions on who can perform an act of annulment, there are some conditions on who can perform a retraction. The Annulment Strategy, as we saw in the previous section, is typically deployed by a third party (as when the Roman Rota annuls a Catholic marriage). However, it can also be the case that a certain act is annulled by its speaker (reconsider the recruitment case: it might well have been that the recruiter herself, after finding out that the position had already been filled, recognized her original act as null) or by its audience (as shown by me realizing that the invitation to your house in Miami Beach was in fact void). On the contrary, in paradigmatic cases, the Retraction Strategy is deployed by the original performer (as in Farage’s case). It might also be mobilized by an authoritative third party (as when the Pope dissolves an unconsummated marriage). But it does not make much sense to claim that the audience of the original illocution can also retract it. If I order you to do something, for example, you can unearth the flaws of my order (as when I have no authority to order, or I ordered you to teleport to Mars), and therewith annul it. But you can in no way undo an order that was validly issued.²⁴

3.3 Amending

It is rather obvious that speakers sometimes alter or change the content of their own utterances. We are pretty used to people who change their stories for all sorts of reasons. They may do so because they lied in the first place and want to come clean or, vice versa, because they told the truth and now try to cover it up. They may do so because their first (false) account of what happened was very unlikely and want to provide a more convincing (false) story. They often do so because of being under intimidation or threat. In this section, I will argue that, in the same way as one can amend what one has said (‘propositional amendment’), one can, under certain circumstances, amend what one has done (‘illocutionary amendment’). Even though it is less obvious, it is still pretty common that speakers alter the force of their own acts. The most easily recognizable instances of amendment are utterances of the form

• (2) P...at least, I guess that P.

• (3) I promise to do $\phi$...well, I’ll try to.

• (4) Do it!....Sorry, I mean, can you do it?

As one can see, (2)–(4) involve no propositional content adjustment. Rather, in each of those cases, the speaker seeks to reshape the illocutionary force of her utterance: (2) is an assertion amended into a conjecture, (3) is a promise amended into the expression of an intention, and (4) is an order amended into a request.

An important point brought out by these (very simple) examples is that a speech act cannot be amended into any other act. Indeed, when we amend an illocution, the original one leaves traces in the normative context, as its force imposes some constraints on the act we can replace it with. One can adjust, for instance, an assertion to a conjecture (or the other way round, as in

• (5) I guess that P...Actually, I’m sure that P),

but one cannot sensibly adjust it to a command, a refusal, or a promise.²⁵ To account for this feature, we may construe the Amendment Strategy as a mechanism to tamper with the degree of strength of one’s performed acts, i.e., to weaken or strengthen their normative burden. For example, in amending an order into a request (say, for reasons of politeness), one is reducing the strength of the normative bond imputed on the hearer, who will now be petitioned to grant the request but will have no duty to do so. Similarly, if I express my intention to help you with your math test and then I turn that into a promise, I will have committed myself to helping you out in a more serious way. This characterization of amendment implies that speakers can amend only those illocutionary types whose conventional effects come in degrees. Communicative acts usually are like that, while institutional acts are not. Consider the assertive type: in general, assertives commit the speaker to the truth of the proposition expressed, but such commitment varies in degree of strength. ‘Assert’, for example, marks an illocution that commits the speaker to truth to a higher degree than ‘suggest’ or ‘conjecture’ and to a weaker one than ‘insist’ or ‘swear’. Unlike assertives and communicative acts more broadly, institutional acts do not seem to be amendable, for we cannot perform any alternative acts whose conventional effects are the same, just weaker or stronger, than, say, marrying, excommunicating, or vetoing [cf. Searle and Vanderveken (1985, p. 54ff)].

To make the strategy at issue more vivid, let us consider an example with a bit more context.

Paul::

Madame Archer, I have an impossible request of you. Is there any way that you could return to the old woman and bring the girls back here to us?

Pat::

Paul, that side of town has been destroyed in the fighting. They’re probably dead.

Paul::

Madame Archer, I’m begging you. Those girls have no one now.

Pat::

I’ll try.

This is an extract from the script of Hotel Rwanda (2004) by Terry George. The movie tells the story of Paul Rusesabagina, a Hutu hotel manager who housed over a thousand Tutsi at the Hôtel des Mille Collines in Kigali during the Rwandan genocide. Paul here is talking to Red Cross worker Pat Archer. At first, he asks Pat for help in reuniting with his little nieces:

• (6) I have an impossible request of you. Is there any way that you could return to the old woman and bring the girls back here to us?

However, facing her reluctance (“Paul, that side of town has been destroyed...”), he subsequently adjusts that request to an outright plea,

• (7) Madame Archer, I’m begging you,

made while putting forth some evidence supporting the urgency of the requested course of action (“Those girls have no one now”). Requests and pleas are instances of what Mark Lance and Rebecca Kukla refer to as ‘calls’—that is, second-person illocutions that call for a specific response on the addressee’s part (Lance and Kukla 2013). Simplifying a little, we can divide second-person calls into two categories, namely, imperative (or closed) calls and interrogative (or open) calls. Imperative calls, such as orders or commands, result in obligations on the part of the hearer; interrogative calls, by contrast, give the recipient a reason to do what was requested, but such a reason does not take the form of an obligation. To put it another way, any authentic interrogative call leaves the decision to grant or reject it up to the hearer. Requests and pleas both belong to the category of interrogative calls, but they differ in that pleas have a higher degree of strength than requests.²⁶ Unlike imperatives, which derive their greater strength from the position of power or authority occupied by the speaker, pleas derive it from the greater intensity of the desire expressed. A felicitous plea puts a lot more pressure on the recipient to grant it than a simple request, because the beggar, unlike the requester, presupposes her being in dire need of what she is begging for. This is why, when Paul turns his request into a plea, he manages to overcome Pat’s reluctance. As he explains, his little nieces have no one to rely on, and Pat’s action can make the difference between their life and death.

Before concluding this section, let us try to identify the general felicity conditions of amendment.

General felicity conditions of amendment: a speech act A performed by a speaker S at a time t may be amended at a later time $t_{{n}}$²⁷ only if

• i. A was felicitously performed at t²⁸;

• ii. At $t_{{n}}$, S makes a supplementary utterance that weakens/strengthens the normative burden of A.

Unlike the other undoing mechanisms, the Amendment Strategy can be deployed only by the original speaker (condition ii.). If I assert that it is going to rain tomorrow, you can challenge my assertion and eventually make me amend it, but you cannot amend it on my behalf (nor can any third party).

4 Concluding remarks

By now it should be clear that undoing things with words is a rather widespread phenomenon. There are cases where one’s performed act was somehow fatally flawed and, once recognized as such, it has to be undone. Or cases where one has spoken insincerely, inappropriately or falsely, and recanting stands as the better option on the table. Speakers have a number of reasons for calling off their illocutions—and a number of strategies at their disposal in order to do so. Amongst them are what I have called the Annulment Strategy, the Retraction Strategy, and the Amendment Strategy. Annulment amounts to recognizing a certain act as invalidly performed in the first place. It typically applies to institutional acts—such as marrying or condemning—and is usually mobilized by a third party endowed with some sort of authority (e.g., the Roman Rota, the Supreme Court). Retraction and amendment are instead ways for making valid acts no longer such. Both of them typically apply to communicative acts—such as promising or asserting—and are usually deployed by the speaker of the act to be undone. The difference lies in their normative function: retraction is a way to cancel the illocutionary effects of one’s words, whereas amendment is a way to alter their degree of strength. As I suggested throughout the paper, when a speaker successfully performs a speech act, the conversational context adjusts in such a way as to include new normative facts (e.g., obligations, commitments, rights) illocutionarily engendered by the act. In the course of a conversation, the deontic statuses of the participants evolve as new speech acts are accomplished. Annulling, retracting, and amending are mechanisms to rub out a certain deontic update—to disavow or undo the normative constraints generated by speakers’ illocutions.