Emojis and conditionals: exploring the super linguistic interplay of pictorial modi ﬁ ers and conditional meaning

: In recent years, formal linguistic analysis has expanded its scope to include objects of study beyond natural language, under the umbrella of Super Linguistics (where the intended meaning of super is its Latinate meaning ‘ beyond ’ ); see (Patel-Grosz et al. 2022. Super linguistics: An introduction. Unpublished manuscript, April 2022 version. at: https://ling.auf.net/lingbuzz/005242). One super linguistic object of study is emojis, which can be analyzed as digital counterparts of gestures and facial expressions, but which also share properties with natural language expressions such as alas and unfortunately (Grosz, Patrick Georg, Gabriel Greenberg, Christian De Leon &Elsi Kaiser. 2021b. A semantics offace emoji in discourse . Manuscript, December 2021 version. https://ling.auf.net/lingbuzz/005981 (Accepted with minor revisions for publication in Linguistics and Philosophy)). In this paper, I use conditionals as a case study to argue that natural language semantics can bene ﬁ t from investigating the semantics of emojis. I start by arguing that face emojis (disk-shaped pictographs with stylized facial expressions) operate on contextually salient propositions. I show that they can comment on the presuppositions of wh -questions and de ﬁ nite descriptions, but not on conversational implicatures. I then show that face emojis can also comment on the counterfactual inferences of subjunctive conditionals (or, more broadly, subjunctive if -clauses). This suggests that these counterfactual inferences may be presupposition-like and not, as widely assumed, an instance of implicature (see Zakkou, Julia. 2019. Presupposing counterfactuality. Semantics and Pragmatics 12(21). 1 – 20, for recent discussion). The study of emojis, a nonstandard object for linguistic inquiry, can thus directly inform more traditional linguistic exploration.


Introduction
Emojis ( , , ) are pictographs that have become common place in written digital communication since they were first introduced to an international market in 2011. 1 The increasing popularity of emojis can be connected to the drastic increase of written online text in everyday communicative exchanges, in the form of blog posts, social media posts, emails, and live chats, or via messaging apps. Written text lacks the multimodal features of face-to-face communication, such as facial expressions (see, e.g., Russell and Fernández-Dols 1997), gestures (see, e.g., Abner et al. 2015), and intonation (see, e.g., Pierrehumbert and Hirschberg 1990); emojis are plausibly a means of reintroducing this multimodality into the written text. 2 Correspondingly, in theoretical linguistics, a view has emerged where emojis are the digital counterparts of gestures and facial expressions, as advocated, for instance, by Cohn et al. (2019), Gawne and McCulloch (2019), Grosz et al. (2021a), Pierini (2021), and Pasternak and Tieu (2022). This paper builds on a recent formal semantic analysis of emojis, and highlights ways in which the study of emojis can directly inform the study of natural language expressions. The focus is on face emojis ( , ), defined as disk-shaped pictographs with stylized facial expressions, and how they are interpreted in connection with natural language conditionals. An emoji-conditional combination is illustrated in (1). 3 Here, the emoji seems to communicate that the author is unhappy about the rain and how it affected plans for the beach. (1) If it hadn't been raining, we would have gone to the beach Section 2 presents the general theoretical framework, and Section 3 shows how face emojis interact with conditionals. 4 The empirical exploration in Section 3 argues that face emojis that follow conditionals can comment on the counterfactual inference of subjunctive conditionals, which is argued to suggest that these are presupposition-like inferences rather than conversational implicatures.
2 Outline of a formal semantics of face emojis

Face emojis as emotive markers of non-at-issue evaluation
In recent formal semantic approaches to the meaning of emojis, Grosz et al. (2021a) and Kaiser and Grosz (2021) argue that face emojis should be analyzed as expressive modifiers; see, for example, Potts (2007) and Gutzmann (2013). In related work, Grosz et al. (2021b) propose a formal semantic analysis (summarized in Section 2.2) that compares face emojis to the emotive markers of Rett (2021). Rett (2021: 307) defines an emotive marker "as a morpheme, syntax, or prosody that encodes the speaker's emotive attitude towards some proposition made salient by the utterance in which it occurs, and does so in backgrounded, not-at-issue content". She focuses on the English words alas and unfortunately, but also includes exclamation intonation and mirative markers in various languages. The parallels between face emojis and Rett's unfortunately are illustrated in (2), where A1, B, and C are cited from Rett (2021: 309), and A2 has been added as a modification of A1. 5 A speaker who utters A1 conveys both that Jane lost and that this is saddening. As shown in B, the first of these two inferences can be directly denied. By contrast, C shows that the second inference cannot be directly denied. The judgments for B and C remain exactly the same if we change A1 to A2. 6 (2) A1: Unfortunately, Jane lost the race. A2: Jane lost the race B: That's not true, she won! C: #That's not true, you're glad she did! Importantly, we also find differences between face emojis and alas/unfortunately, suggesting that the emojis are related to Rett's emotive markers, but not subsumed by them. As shown in (3a), alas/unfortunately cannot combine with an imperative, but emojis do not exhibit such a restriction. Example (3b) is entirely well-formed and may convey a certain level of resignation, for example, concerning the author's expectation of whether the addressee will actually carry out the requested action; alternatively, the emoji may comment on the presupposition that the room has not been cleaned (see Section 2.3.1 on face emojis that comment on presuppositions).  (2) is discussed in Grosz et al. (2021b: 41). 6 Prior work shows that emoji users have reliable introspective intuitions on emoji use, which can be confirmed in controlled experiments, as demonstrated by Kaiser and Grosz (2021) for the intuitions in Grosz et al. (2021a). Similarly, intuitions from Grosz et al. (2021b) have been corroborated experimentally in Bjertnes (2022). (3) a. #Alas/#Unfortunately, clean up your room! (Rett 2021: 320) b. Clean up your room As an additional data point, (4) shows that face emojis are also acceptable in connection with wh-questions, which further sets them apart from Rett's (2021) emotive marker unfortunately. Rett (2021: 318) argues that such emotive markers only occur with wh-questions that make a single proposition salient, which explains why they are unacceptable in (5a). Interestingly, though, alas differs from unfortunately in being quite acceptable in (5b). This indicates that face emojis are more similar to alas than unfortunately (as also suggested by Grosz et al. 2021b).

(4)
What's that noise? 2.2 Face emojis as propositional operators Grosz et al. (2021b) present an analysis of face emojis as a type of expressive modifier (see also Grosz et al. 2021a;Maier 2021), summarized in (6b) and (7b). Here, (6a) and (7a) are two different direct messaging exchanges that could occur in exactly the same context, and which only differ in the information available to B. The analysis of Grosz et al. (2021b) has two core components: face emojis comment on a salient proposition p, which is anaphorically retrieved from the immediate linguistic context (typically from the preceding sentence), (6d) and (7d). The emojis comment on p in relation to a contextually provided discourse value V, a proposition that represents the author's desires, aspirations, wishes or hopes, (6c) and (7c). The interplay between p and V accounts for the observation that face emojis with contradictory valence ( vs. ) can be used with reference to the same proposition, as in (6a) and (7a), in light of additional contextual information.
(6) a. A: you must be starving B: I've already eaten b. ⟦ ⟧ = λxλpλV . {w | x is happy about how p bears on V at w} c. V = the author is satiated d. p = the author has already eaten (p contextually entails V) For the purposes of the present paper, the connection to discourse values is less central than the fact that face emojis comment on a proposition p which they access through an anaphoric relation. In Section 2.3, I show how we can repurpose this property of face emojis as a tool to diagnose the status of salient propositions; for example, whether they are presuppositions or implicatures of the preceding text.

Targets of face emojis: presuppositions versus implicatures 2.3.1 Face emojis can target presuppositions
Face emojis often comment on the proposition expressed by the accompanying text, as in (6) and (7), but this is not always the case. Grosz et al. (2021b: 10) document that face emojis can also comment on presuppositions of the accompanying text. This is particularly visible with wh-questions: in (8a) (from Grosz et al. 2021b: 10), the emoji appears to comment on the existential presupposition, (8b), rather than, say, an expected answer, (8c), which may be a possible reading in its own right. 7 (Here, "✓" marks the prominent reading, "(✓)" a less prominent one.) (8) a. Who drank my coffee?
b. Presupposition-targeting reading: ✓ I'm unhappy that p = somebody drank my coffee c. Answer-targeting reading: (✓) I'm unhappy that p = I expect that the answer will be unpleasant Another type of presupposition that face emojis can target is plausibly the existential presupposition of definite descriptions. This is shown in (9a), where the emoji seems to comment on the existential presupposition of the evidence, (9b), 8 rather than on the asserted content of the preceding text, (9c).
(9) a. I long thought that there was no evidence in this case, and I was about to give up. The evidence was in the trash can b. Presupposition-targeting reading: ✓ I'm happy that p = there is evidence c. Assertion-targeting reading: (✓) I'm happy that p = the evidence was in the trash can

Face emojis fail to target conversational implicatures
Importantly, there are limits when it comes to the potential targets of face emojis, that is, they cannot simply target any proposition that is inferable in the context. Specifically, face emojis do not standardly target conversational implicatures. We see this for scalar implicatures (see, e.g., Chierchia et al. 2012) in (10A 1 ), where the sad face ( ) seems to be deviant. 9 If the emoji could freely target the some-but-not-all implicature in A 1 , 7 In biased yes/no-questions (see, e.g., Ladd 1981; Büring and Gunlogson 2000; Romero and Han 2004;Romero 2006;Krifka 2017: 360), face emojis can comment on the expected answer, but not on its opposite, which explains the acceptability of the unhappy face emoji in (ii), and the happy face emoji in (iii): (i) Context: The author only eats at vegetarian restaurants; the author is currently looking for a place to eat.
(ii) Is there no vegetarian restaurant around here? / # ⤳ I'm unhappy that p = I expect that there is no vegetarian restaurant Isn't there some vegetarian restaurant around here? / ?# ⤳ I'm happy that p = I expect that there is a vegetarian restaurant The text in (ii) and (iii), to which the face emojis have been added, is taken from Büring and Gunlogson (2000: 5), whose versions are modifications of an example from Ladd (1981: 164).
8 Examples of this type raise interesting questions about the nuances associated with different face emojis. An anonymous reviewer asks if the choice of emoji in (9a) (the "grinning face" instead of the "smiling face with smiling eyes" ) indicates that the author is not commenting on the presupposition there is evidence, but rather expressing amusement along the lines of it is funny that the evidence was in the trash can. In this paper, I generally use the emojis that seem most natural in a given example; that being said, as far as I can tell, the variant with , given in (i), is as acceptable as (9a) and less compatible with an amusement-expressing reading, thus strongly favoring the reading in (9b).
(i) I long thought that there was no evidence in this case, and I was about to give up. The evidence was in the trash can 9 The intended reading of (10A 1 ) is one that does not involve irony or sarcasm, which can improve otherwise deviant examples such as (i) (see also Giustolisi and Panzeri 2021): A parallel point can be made for ignorance implicatures; example (11A 1 ) triggers an ignorance implicature very much like what is openly asserted in (11A 2 ). However, while the emoji in (11A 2 ) can convey "I'm sad that I don't know exactly", this is not possible in (11A 1 ), so the ignorance implicature of (11A 1 ) is not accessible to the emoji either.
(11) Q: Where does C live? A 1 : Somewhere in the South of France # A 2 : I don't know exactly (modeled after Grice 1975: 51) Prior to discussing the interaction of face emojis with conditionals, we can thus conclude that face emojis can be analyzed as expressive modifiers, which comment on a salient proposition p; while p is often expressed by the accompanying text, it may also be a presupposition of the accompanying text, but not a conversational implicature.
Turning to our case study, conditionals, a central question in their semantic analysis concerns the inferences that are drawn from subjunctive mood marking in the antecedent clause. To set the scene, take (12a) as a representative past subjunctive conditional. 11 Simplifying significantly for the purpose of this paper, the entailment of (12a) can be stated as p > q, which is informally paraphrased in (12b) (ignoring tense). Crucially, (12a) gives rise to a counterfactual inference, (12c).
(12) a. If he had been late, he would have called us.
(past subjunctive conditional) b. p > q ≈ in situations in which he is late, he calls us c. ¬p ≈ he was not late (strong counterfactual inference) There is an ongoing debate on how to model the counterfactual inference (12c). At the very least, it could be a counterfactual presupposition or a counterfactual implicature (see, e.g., Portner 1992;von Fintel 1998von Fintel , 1999Ippolito 2003Ippolito , 2007Schlenker 2004;Schulz 2014;Arregui and Biezma 2016;Leahy 2011Leahy , 2018Wimmer 2020; von Fintel and Iatridou 2020; among many others). Building on Anderson (1951) and Stalnaker (1975), the dominant view has long been that the counterfactual inference is an implicature as opposed to a presupposition. However, presuppositional analyses have occasionally been pursued; for example, by Portner (1992) and Karawani (2014). Most recently, Zakkou (2019) argues that the arguments against the presupposition view 10 An anonymous reviewer points out that the sad face in (10A 1 ) seems to improve if prosodic cues are added, such as stress on some. While I share this intuition, its interpretation is complicated by the fact that emojis only occur in written text, whereas prosody and stress are spoken language phenomena. Moreover, stress on some may involve the insertion of a covert only (see, e.g., Chierchia et al. 2012), which would make the example equivalent to (10A 3 ). 11 The labels subjunctive and counterfactual have often been questioned as problematic in the literature on conditionals, most recently in von Fintel and Iatridou (2020), who propose the term X-marked conditional for constructions of the type in (12a). I use the term subjunctive in a loose informal sense, adopting the label (non)past subjunctive conditional from Ippolito (2003).

Emojis and conditionals
are inconclusive, and that the presuppositional viewwith a falsity presupposition as stated in (12c)is a viable contender. Zakkou (2019: 3) points out that her discussion only applies to past subjunctive conditionals, (12a), and that the facts may be different for nonpast subjunctive conditionals, like (13a), for which she considers a presupposition view to be less plausible. However, she focuses on falsity inferences, and the "counterfactual inference" may well be weaker in (13c) (where it merely amounts to the unlikelihood of p) than in (12c) (where it amounts to the falsity of p). Such an asymmetry has been explicitly proposed, for example, by Karawani (2014: 167-172). In (13a-c), we may then also ask if unlikelihood is presupposed or implicated.
(13) a. If he were late, he would call us.
(nonpast subjunctive conditional) b. p > q ≈ in situations in which he is late, he calls us c. unlikely(p) ≈ he is not expected/likely to be late (weak counterfactual inference) In an attempt to shed new light on this question, we can now ask what happens if we add a face emoji to a conditional. Specifically, given that face emojis seem to target presuppositions, but not implicatures, we can ask if a face emoji can target the counterfactual inference. In what follows, I argue that face emojis can in fact target the counterfactual inference of subjunctive if-clauses (including conditionals); this suggests that such inferences are presupposition-like, and do not pattern like conversational implicatures.

Face emojis and negated subjunctive if-clauses
To see that face emojis can target the counterfactual inferences of subjunctive if-clauses (including conditionals), let us start by looking at negated counterfactuals. This is a reasonable strategy to pursue since negated counterfactuals can embed strong evaluative adverbs such as unfortunately, as in (14), in connection with a particularly strong falsity inference; see, for example, Liu (2012) and Axel-Tober and Grosz (2013), pace Ernst (2009). Our study of face emojis can be informed by insights on such emotive markers due to their similar nature (see Section 2.1): since unfortunately in (14) can comment on the counterfactual inference ¬p (= I ordered the tofu), it is a reasonable strategy to ask if emojis can also access this inference.

(14)
If I had not unfortunately ordered the tofu, I would have loved this restaurant.
To begin with, consider example (15b), modeled after an example from Axel-Tober and Grosz (2013: 6), and supplied with the context (15a). The context entails that loving the restaurant (the consequent proposition) would correspond to the author's desired outcome.
(15) a. Context: I ate at the new vegan restaurant in my neighborhood, hoping to love it. I ordered dumplings and summer rolls, which I liked, but I also ordered tofu, which I did not like. b. If I had not ordered the tofu, I would have loved this place.
If we add a face emoji to (15b), there are at least four propositions that the emoji could in principle target. First of all, the entire conditional p > q is informally paraphrased in (16a) (ignoring tense), and the counterfactual inference φ is given in (16b). Since (15b) is a past subjunctive conditional, the counterfactual inference amounts to a falsity inference, that is, ¬p.
Due to conditional strengthening ("perfection") from p > q to ¬p > ¬q (see, e.g., Geis and Zwicky 1971), we can also include ¬p > ¬q, in (16c). From a combination of (16b) and (16c), most readers of (15b) will also infer that ¬q is true, as spelled out in (16d), since ¬q follows from [¬p > ¬q]∧¬p. This means that ¬q is another conceivable target for an emoji. derives from a similarity-based analysis of counterfactual conditionals (Lewis 1973): if the counterfactual nontofu-worlds that are most similar to the actual world are worlds in which the author loved the restaurant, then this implies that the author already liked the restaurant to a relatively high degree (i.e., the author almost loved it). The proposition in (16a) thus pragmatically conveys that the restaurant was quite good, which promotes the goal of loving the place (i.e., a discourse value V = the author loves the place). By contrast, all of (16b-d) are plausibly negatively evaluated, and thus compatible with , as they demote the goal of loving the place. A reader may wonder if the prejacent propositions in the antecedent (p = I did not order the tofu) or the consequent (q = I loved this place) could be further potential targets for a face emoji. This is highly unlikely, since these two propositions are inferred to be false, given that (16b) and (16d) are implied to be true, and positively/negatively evaluative face emojis ( / ) exhibit a factivity requirement. To see this, consider (17), which is non-factive, as shown by the two possible continuations in parentheses. If we add a happy face emoji that targets the embedded proposition, 12 the resulting reading is one where the author implies that Mel is coming to the party, illustrated in (18). In other words, the face emoji forces a factive inference; this would collide with the falsity of p and q in (15b).

(17)
Sam didn't say that Mel was coming to the party (… but she is in fact coming / … and, indeed, she isn't coming) (18) Sam didn't say that Mel was coming to the party ⤳ I am happy about the fact that Mel is (or will be) at the party Let us now explore the patterns that we find when adding face emojis to (15b) Since face emojis can generally target the entire proposition asserted by a preceding declarative (see Grosz et al. 2021b: 10), we expect that a happy face emoji can be added to (15b) and target the p > q proposition (16a). This expectation is confirmed in (19a), for which the most natural interpretation is given in (19b). 13 (19) a. if I had not ordered the tofu, I would have loved this place b. ⤳ I am happy that p > q (= in non-tofu situations, I love this place) More importantly, the addition of an unhappy face emoji is equally acceptable, (20a), though the effect is different. Now, the face emoji appears to comment on an inference that is distinct from the conditional p > q. The reading that is intuitively available is compatible with three competing analyses: the sad face emoji may comment on φ (20b), φ > ψ (20c), or ψ (20d).
(20) a. if I had not ordered the tofu, I would have loved this place b. Analysis 1:⤳I am unhappy that φ (= I ordered the tofu) c. Analysis 2:⤳I am unhappy that φ > ψ (= in tofu situations, I do not love this place) d. Analysis 3:⤳I am unhappy that ψ (= I did not love this place) Prior to looking at more data, we observe the following: if the emoji can comment on φ, in line with (20b), then this lends credence to the presuppositional analysis of counterfactuality, since we have established that face emojis can comment on presuppositions (Section 2.3.1). 14 However, we first need to rule out the option that the emoji can only comment on φ > ψ or on ψ, in line with (20c-d). It may initially seem difficult to decide between (20b), (20c), and (20d), as their pragmatic effects are 12 There is also an irrelevant reading where the emoji targets the main clause proposition, i.e., that Sam didn't say it. 13 Face emojis often have a reading where they comment on an individual referent rather than a proposition (see Grosz et al. [2021b: 3] for discussion). This gives rise to an additional reading of (19a), where the emoji comments on this place and expresses that the author simply has a positive stance towards the restaurant. I am grateful to an anonymous reviewer for pointing out this reading of (19a). This paper is only concerned with face emojis that comment on propositions. 14 An attentive reader may find the terminology jarring, since (18) demonstrates that face emojis have a factivity requirement, whereas the face emoji in (20a) is argued to comment on the counterfactual inference of a past subjunctive conditional, (20b). This is by design: the addition of the face emoji communicates that ¬p (the falsity of the conditional's antecedent proposition) is a fact; similar intuitions hold for (14). roughly equivalent in context (15a). However, I argue that an analysis where the emoji comments on φ > ψ or ψ is less plausible to begin with than an analysis where the emoji comments on φ, for the following reason. While the φ inference is triggered by subjunctive marking, the proposition φ > ψ arises via conditional strengthening, which has been argued quite uncontroversially to be a conversational implicature (see, e.g., von Fintel 2001); based on our findings in Section 2.3.2, φ > ψ should thus not be a suitable target for an emoji. Crucially, ψ is a pragmatic inference that is derived from [φ > ψ]∧φ, and thus parasitic on conditional strengthening, which makes it an even less likely target. This favors (20b) over (20c-d), and we can tentatively conclude that face emojis that follow past subjunctive conditionals can target the counterfactual inference φ (= ¬p) of the antecedent if-clause as their propositional argument.
Naturalistic Twitter examples provide direct evidence that face emojis can comment on the counterfactual inferences (φ) of subjunctive if-clauses, though the clearest examples to date involve subjunctive optatives, illustrated in (21). We can include such evidence, since the mechanism for modeling counterfactual inferences in optatives, (21), and conditionals, (20), has been argued to be the same (see, e.g., Grosz 2012). In line with (12c) and (13c), we can differentiate between a falsity inference in (21a) and an unlikelihood inference in (21b). 15 (21) a.
if only I hadn't moved [twitter] (past subjunctive optative) ⤳ I am unhappy [that it is false that I didn't move] b. if only they weren't weirdos [twitter] (nonpast subjunctive optative) ⤳ I am unhappy [that it is unlikely that they aren't weirdos] Before we move on to non-negated subjunctive if-clauses in Section 3.3, it is worth showing that examples of the type in (19) and (20) can also be found in a naturalistic data source. The near-minimal pair in (22a) and (23a) is reproduced from Twitter. Here, too, the happy face emoji in (22a) can comment on the entire conditional, (22b). 16 By contrast, (23a) has a plausible interpretation where the emoji comments on the counterfactual inference, (23b), modeled as unlikelihood since we are dealing with a nonpast subjunctive conditional.
(22) a. I would buy it if I wasn't broke [twitter] b. ⤳ I am happy that p > q (= in worlds in which I am not broke, I buy it) (23) a. I would totally buy this if I wasn't broke [twitter] b. ⤳ I am unhappy that unlikely(p) (= it is unlikely that I am not broke)

Face emojis and non-negated subjunctive conditionals
Consider now a scenario where, in the restaurant context of (15a), your friend sends the message in (24). Once again, we can list the candidate propositions in (25). The judgments with regards to (24) seem to generally reproduce those in Section 3.2 in that the sad face emoji can target the counterfactual inference (25b). (The reasoning for (20c-d) carries over to (25c-d).) 15 Examples that are cited from Twitter are marked with a [twitter] superscript. When citing social media text, questions relating to ethics and privacy arise (see, e.g., Ayers et al. 2018;Tatman 2018). To provide anonymity, this paper omits user names and URLs, but keeps examples unmodified in order to preserve their linguistic integrity. The study obtained approval from the Norwegian Centre for Research Data (NSD), reference number 414881. 16 Interestingly, the intuitions with regards to the meaning of (22a) stood out in that they were controversial among emoji users who were consulted, and they were also questioned by an anonymous reviewer. Several consultants report the intuition that the author of (22a) is actually laughing at the current situation, i.e., expressing gallows humor. While this is an interesting observation, it is not critical for the discussion in this paper, as the reading in (22b), implausible as it may be, is predicted to be possible under any approach.

(24)
If you had ordered the tempeh, you would have loved this place! (25) a. p > q: in tempeh situations, you love this place b. φ (= ¬p): you did not order the tempeh (= the counterfactual inference) c. φ > ψ (= ¬p > ¬q): in non-tempeh situations, you do not love this place d. ψ (= ¬q): you did not love this place Once again, naturalistic examples can be provided, as in (26), which show that face emojis can comment on the counterfactual inferences of non-negated subjunctive optatives.
(26) a. If only we had stopped [twitter] (past subjunctive optative) ⤳ I am unhappy [that it is false that we stopped] b. if only you knew. [twitter] (nonpast subjunctive optative) ⤳ I am unhappy [that it is unlikely that you know] To wrap up this discussion, (20)-(26) indicate that face emojis can target the counterfactual inferences of subjunctive if-clauses (including subjunctive conditionals). This corroborates a view where such counterfactual inferences are presupposition-like in nature.

Conclusion: lessons to be learned
Face emojis can shed new light on the semantic inferences that arise from natural language expressions such as conditionals (and if-clauses more generally), since the emojis target propositions that are salient in the context. The discussion in this paper is based on introspective intuitions of emoji users (see footnote 6) with the aim of establishing testable hypotheses. These hypotheses of course need to be validated by further empirical and experimental testing in follow-up research. For the time being, we can cautiously summarize and conclude as follows. In Section 2, I argued that face emojis can comment on presuppositions of the preceding text, but not on conversational implicatures. In Section 3, face emojis were combined with if-clauses, and it was observed that face emojis can comment on the counterfactual inference of subjunctive if-clauses. This indicates that counterfactuality is an inference that is presupposition-like and does not behave like a conversational implicature. Having arrived at this conclusion, we thus see that emoji semantics can provide new insights into natural language semantics.