This paper presents a formal semantics of verbal irony in assertions. In particular, it makes precise what is meant by the common intuition that an ironic utterance expresses the opposite of its literal meaning. We start by considering cases of verbal irony that are marked by a particular prosodic tune in English. We then demonstrate that an extant model for intonational meaning can be extended to capture ironic prosody. Afterwards, we discuss how to expand this semantics to cases of irony that are not marked by prosody.
The goal of this paper is to formally model some intuitions about verbal irony. We first approach the topic as a problem in the semantics of intonation and assign a semantics to one particular prosodic tune that appears to mark an utterance as ironic. This semantics in particular specifies how to compute the opposite of the literal content of an ironic utterance. We then investigate how this semantics could generalise.
Formalising intonational meaning faces many
difficulties. One central problem is that tunes fall
on a spectrum that resists comprehensive sorting
into discrete categories. This presents a problem
for symbolic approaches in general, as then
intonational meaning also resists discretisation
Empirical data suggests that irony is linked to
prosodic contrasts
; inflection is very uninteresting.
The significance of prosody with respect to irony can be appreciated by considering a minimal pair where the tune of an utterance makes the difference between acceptance and rejection. (2) is one such case (% marks a high pitched accent). (2) a. A: Are you going to Mike’s show tonight? b. B: I’ll %definitely go to that. ; will go b.0 B: I’ll #de-efinitely go to that. ; won’t go Tune is the only variable that distinguishes (2b) from (2b0). The goal of this paper is to isolate a semantics (in the sense of Ladd’s (1980) intonational lexicon) for the tune of (2b0) that makes the right predictions. Note that we are not claiming that the tune in (1) and (2b0) is the only tune that marks irony—or that irony requires any particular intonation. We return to this in section 4.
There are many competing explanations of verbal irony, but they—by and large—revolve around a 1The scene containing (1) can be listened to at https://www.youtube.com/watch?v=zIavvxoqxvs common intuition: that the speaker of an ironic utterance means the opposite or the inverse of the literal content of their utterance.
On a Gricean account, verbal irony is the
flouting of the Maxim of Quality, i.e. the speaker
asserts something recognisably false and therefore
means the opposite; on the echoic account an
ironic utterance mentions a sentence and indicates
dissent from it
Schlo¨der and Lascarides (2015) present a
formal model of intonation that already goes some
way towards a semantics of irony; they assign a
semantic term to ironic tunes that expresses
dissent from the literal content of an utterance and
allow this dissent to be strengthened to assent to
the negated utterance. The predictions of their
account are too weak, however. It is not sufficient to
just add a negation to the ironic utterance’s
propositional content. Recall (2b0).
(2) b.0 B: Yeah, I’ll #de-efinitely go to that.
The negation of the literal content of (2b0) is it is
not the case that B will definitely go to the show
which resolves to B might not go to the show. The
correct reading is however that B will (definitely)
not go. Giora (1995) can potentially account for
this: she makes the implicit negations stemming
from irony be subject to a preference for contrary
negation over contradictory negation
This preference, however, must be formalised to be predictive. One at least has to indicate how to select from the multiple different possibilities to form a contrary negation. In the case of (2b0), at least two such readings are available. (3) a. B will definitely not go to Mike’s show. b. B will definitely go somewhere that is not
Mike’s show.
Both (3a) and (3b) are contrary to the literal content of (2b0). However, (3b) overstates what B expresses in (2b0) because it entails that B will go somewhere—but this does not seem to be part of the meaning of (2b0). It is unclear what privileges (3a) over (3b) (or other options) if all we have is a general preference for contrariness.
While there might be ways to spell this out, there is another option. The position of the fall in the ironic tune we are considering seems to indicate the placement of the negation. For instance, (3b) would be the appropriate interpretation for a third possible answer to (2a). (4) a. A: Are you going to Mike’s show tonight? b.00 B: Yeah, I’ll definitely go to #tha-at. (4b00) is appropriate in a context where there is a salient alternative activity that B could attend and the context moreover suggests that B would prefer this one (e.g. because there is a much better show overlapping with Mike’s).
The pattern that the contrary negation is scoped on the word immediately following the fall seems to be robust. Consider some variants of (1). (5) a. Inflection is #so-o interesting.
a.0 Inflection is so #i-interesting.
a.00 #Infle-ection is so interesting.
One can equally well place the fall before the intensifier or before the predicate to obtain the contrary negation intensely uninteresting; (5a00) is felicitous, but gets an additional implicature like in (4b00). However, in (6), placement on ‘movie’ instead of ‘amazing’ sounds odd. (6) a. A: Showgirls is a nice movie.
b. B: It’s an #ama-azing movie. ; terrible #b.0 B: It’s an amazing #mo-ovie.
We will now model this as follows. We consider the word following the fall to be the prosodic focus of the ironic utterance and amend an existing theory for focus to include an implicit negation.
Geurts and van der Sandt (2004) provide us
with a minimal theory of prosodic focus. Their
background–presupposition rule (BPR) states that
whenever prosodic focus gives rise to a
background '(x), there is a presupposition that 9x:'.
In a dynamic model for discourse update, we can
specify this as follows (the intended notion of
presupposition is van der Sandt’s presupposition as
anaphora model
The focus placement separates an utterance into a foreground f and a background '(x). The variable x occurs freely in the formula ', and the constituent f is of the type required by x. Write hf; '(x)i for a foreground–background pair. Updating a discourse with hf; '(x)i is to update with the proffered content ( x:')(f ) and the presupposed content 9x:'.
This rule is minimal in the sense that other
models for focus make at least the predictions of
the BPR. Two prominent models are the QUD
model
presupposes: Rachel likes someone. b. B: Rachel likes Michael.
presupposes: Rachel likes someone.
In this case, the presupposition of (7b) is bound to the presupposition of A’s question (7a). In contrast, there are cases where the presupposition effected by the BPR is accommodated. (8) a. A Does Rachel like anyone? b. B: Rachel likes Michael.
presupposes: Rachel likes someone.
In (8) the presupposition of (8b) cannot be bound and must be accommodated; the contribution of B’s utterance can be paraphrased as Rachel does like someone—specifically, she likes Michael.
However, while the BPR models (prosodic)
focus, it is not sensitive to the overall tune of an
utterance. The tune can potentially affect the content
of both the foreground and the background; see
Thus, it is not surprising that we need to make amendments to the BPR when attempting to model ironic intonation. Schlo¨der and Lascarides (ms) argue that fall-rise tunes work by placing an implicit negation in the background. Here, we adapt the BPR to include an implicit negation in the foreground. This negation is placed to result in a contrary reading.
2Some (e.g. Dryer (1996)) have challenged the idea that focus is directly related to presupposition; Geurts and van der Sandt offer responses that we cannot repeat or evaluate here.
If an utterance is intonated with the ironic tune,
and the fall is immediately preceding the
constituent f then the foreground–background pair
of the utterance is h f; '(x)i where: (i) '(x) is
the background resulting from considering f the
foreground of the utterance and (ii) is a
metaoperator3 that specifies contrary negation:
– if f is a modal or quantifier, f is f :.
– if f is on a scale, f is an item from the
opposite end of the scale;
– if f is a bivalent predicate, then f is :f ;
– if f is an entity, then f is a meta-variable such
that for any predicate P , P ( f ) = P (f ).
That is, updating a discourse with this utterance
is to update (by usual methods) with the
presupposition 9x:' and the proffer ( x:')( f ).
This rule can be regimented in SDRT
The Irony Rule models some of the examples we have seen so far as follows. Consider first (6b): (6) a. A: Showgirls is a nice movie.
b. B: Yeah, it’s an #ama-azing movie.
presupp: 9xpredicate:x(s) ^ movie(s) proffers: amazing(s) ^ movie(s)
terrible(s) ^ movie(s) The presupposition of (6b) indicates that B’s utterance matches the current topic of the discussion, i.e. the properties of Showgirls. Roberts (2012) provides an account of what the current topic is and what it means to match it; this too can be regimented in the present model (Schlo¨der and Lascarides, ms) and we will not go into the details here. Then, the Irony Rule modifies what B is taken to proffer by adding a contrary negation.
By specifying the relative scope of the contrary negation we avoid ambiguities. We show this for (2b0) and (4b00): (In the logical forms we simplify or ignore a number of ancillary details, including tense, possessive case, and the presuppositions associated with proper names.) (2) a. A: Are you going to Mike’s show tonight? presupp: 9s:of(m; s) ^ show(s) proffers: ?go(b; s) b.0 B: Yeah, I’ll #de-efinitely go to that.
presupp: 9xaux:x(go(b; s)) proffers: 2go(b; s) 2:go(b; s) 3That is, it is an operator on logical forms; its application is computed when logical form is constructed. (4) b.00 B: Yeah, I’ll definitely go to #tha-at. presupp: 9xentity:2go(b; x) proffers: 2go(b; s) 2:go(b; s) Thus, the proffered contents of (2b0) and (4b00) are the same: both are—by way of irony—negative answers to A’s question in (2a). But while in (2b0) the presupposition effected by the Irony Rule is a tautology (because for any p there is a modality r such that rp), the presupposition of (4b00) entails that B is going somewhere—just not to Mike’s show. This is precisely the difference between the two competing contrary negations in (3).
Overall, the placement of the steep fall in the tune we are considering here appears to be quite flexible, and the Irony Rule predicts this. Similarly to (2) and (4), the utterances in (5) are all assigned the same proffered content, but the presupposition varies in (5a00). The anomalous (6b0) on the other hand is assigned the absurd interpretation that Showgirls is not a movie. (6) #b.0 B: It’s an amazing #mo-ovie.
presupp: 9xpredicate:amazing(s) ^ x(s) proffers: amazing(s)^ movie(s)
amazing(s) ^ :movie(s) 4
As said, we do not claim that every ironic utterance must carry a particular tune. The following is an example by Cutler (1977) for irony that is not marked by prosody. (9) Upon entering a restaurant devoid of custom.
a. A: Looks like a really popular place.
In cases like (9) it is the salient contrast between what is said and what is actually the case that leads us to an ironic interpretation. Cutler does not provide a tune to go with the utterance, but it seems to us that (10a) would be natural, and moreover that one can use ironic intonation as in (10b). (10) a. A: Looks like a %really popular place.
b. A: Looks like a #rea-ally popular place. Now note that the Irony Rule makes the correct prediction for (10b). With this in mind, there does not seem to be anything that would stop us from saying that we use the Irony Rule instead of the BPR in any situation where an utterance is ironic. That is, we generalise the Irony Rule to also capture utterances like (10a), and take the foreground f to be the focus of the utterance.4
4There seems to be a tacit consensus that every assertion has a focus; McNally (1998) spells this out.
Similar extensions could be made to cases of written verbal irony, e.g. as marked by scare quotes. (11) is cited from Predelli (2003). (11) a. this remarkable piece of ‘art’ consists of a large canvas covered with mud (...) Again, applying the Irony Rule to (11a) under f = art makes the correct predictions here. Similar things can be said about written irony marked by some form of irony punctuation.
However, one needs to spell out such extensions
of the Irony Rule with great care. Not every false
utterance is ironic, and neither is every instance
of scare-quoting
Then there are still cases of irony where it takes the form of playful mockery instead of the implicit assertion of a negative. Wilson (2006) demonstrates this with (12a,b), said to a very careful driver that always makes sure the tank is filled; the utterances mock this behaviour ironically.5 (12) a. A: Do you think we should stop for petrol? b. A: I really appreciate cautious drivers. (12a) is no assertion, and it does not seem to be the case that (12b) means that A dislikes (contrary to appreciate) cautious drivers. Thus, the cases in (12) go beyond what our Irony Rule captures. 5
We have given a fully formal model for verbal irony, insofar as irony is understood as meaning the opposite of what one has asserted literally. The contribution of the model is in particular to make formally precise what we mean by ‘the opposite’. The model starts out as a model of ironic intonation and embeds seamlessly into an extant model of intonation in discourse, but it stands to reason that it may extend to ironic assertions that are not specifically marked by intonation.
5A reviewer points out that there is also pragmatic irony: Thanks for holding the door after a door has not been held. The Irony Rule can potentially explain this; it yields thanks for not holding the door. This is a proffer that fulfills preparatory conditions for thanking, so we can continue with standard pragmatic reasoning. Note that one can explicitly utter thanks for not holding the door to make, by and large, the same speech act as an ironic thanks for holding the door.
I am grateful to Raquel Ferna´ndez, Alex Lascarides, and three anonymous reviewers for their helpful comments.
The research leading to these results has received funding from the People Programme (Marie Curie Actions) of the European Union’s Seventh Framework Programme FP7/2007-2013/ under REA grant agreement no. 607062.