The interpretation of negated antonyms is characterised by a polarity asymmetry: the negation of a positive polarity antonym (
Opposition is a fundamental relation expressed by all natural languages between propositions or predicates that are incompatible with one another (for seminal work on antonymy in philosophy and linguistics see
Many scholars have investigated the interplay between sentential negation, on the one hand, and contrary antonyms, on the other (
Semantic interpretation and negative strengthening of ‘not happy’. The range of the pragmatically strengthened interpretation of negation is highlighted with colour.
Horn (
Negative strengthening is a particularly interesting phenomenon at the interface between semantics and pragmatics since it appears to be modulated by adjectival polarity. Each antonymic pair consists of two opposites: a positive polarity predicate and a negative polarity one. Polarity is typically defined based on three distinct – often, but not always, converging – criteria (see
Polarity appears to play a crucial role in the interpretation of negated adjectives. Linguists since Ducrot (
In the last twenty years, the asymmetric interpretation of positive and negative adjectives has been investigated in a few experimental studies. Colston (
Recently, the polarity asymmetry of negative strengthening has received strong empirical support from a study by Ruytenbeek et al. (
Across both studies, participants were significantly more likely to strengthen the negation of a positive antonym than the negation of a negative antonym, thus confirming the main effect of polarity on the degree of negative strengthening. Furthermore, Ruytenbeek et al.’s (
Before turning to this, though, it is worth specifying that the linguistic observations and empirical findings reported above concern
(1) | a. | He’s not mean. | (≠ He’s nice) |
b. | She’s not sad. | (≠ She’s happy) | |
c. | She’s not ecstatic. | (≠ She’s miserable) |
Horn (
Many scholars explain the polarity asymmetry observed in the interpretation of negated adjectives as originating from the fact that negation can pragmatically function as a tool for face-management.
To begin with, it is thus worth defining some of the key concepts related to face-management that will play a role in our discussion of negative strengthening. The core notion is that of
In their discussion on the interpretation of negated antonyms, Brown and Levinson (
This face-management based explanation of the polarity asymmetry of negative strengthening has been embraced and developed by many scholars. For instance, Horn (
In sum, according to these authors, the asymmetry in the interpretation of negated positive and negative antonyms is ultimately due to their relative tendency to be employed in the performance of a face-threatening act (for an overview of the relevant considerations, see
In Gotzner and Mazzarella (
Brown and Levinson (
In what follows, we elaborate on Brown and Levinson’s discussion, which we take as the starting point of our experimental investigation. Based on Brown and Levinson’s examples, we define a context as
(2) | Ordinary context for |
You want to join a prestigious company and you are competing with one of your current colleagues. After looking at your CV, your officemate tells you: |
(3) | Non-ordinary context for |
You want to join a prestigious company and you are competing with one of your current colleagues. After looking at your competitor’s CV, your officemate tells you: |
In the ordinary context (2), the utterance
The face-management explanation of the polarity asymmetry of negative strengthening predicts that this asymmetry should be reversed in non-ordinary contexts. This is because, according to Brown and Levinson (
We recruited 60 participants with US IP addresses on Mechanical Turk (across two experimental lists). Participants were screened for native language and only included in the analysis if their self-reported native language was English. 24 women and 36 men participated in the study. Their mean age was 35.56, with a standard deviation of 10.61 (age range 23 to 72). The experiment lasted about 10 minutes and participants were paid $0.80 in compensation.
We selected 20 antonymic pairs from Ruytenbeek et al. (
Example item for the adjective
Context: You have decided to quit your current job and change career path. You are not popular in your team. A colleague from another department, tells you: |
According to your colleague, the team is: |
sad 1 2 3 4 5 6 7 happy |
The task of the participants was to indicate what the speaker wanted to communicate on a scale ranging from the negated adjective to its antonym.
The main manipulation was adjectival polarity (positive vs. negative), which was administered in a within-subject but between-item design. That is, one participant would only see either the positive or negative adjective of a given antonymic pair. Hence, each participant saw 10 statements with positive and 10 statements with negative adjectives, resulting in 20 critical trials and an overall number of 1200 critical observations. In addition to the critical items, participants were presented with 10 filler statements not involving negation, such as ‘John is gorgeous’ (where the response scale was anchored at the adjectives ‘gorgeous’ and ‘ugly’). The filler sentences also served as attention checks.
The experiment was programmed in HTML and run via MTurk’s built-in environment. The experimental procedures and predictions were pre-registered with the as.predicted.org template. The pre-registration is available on the Open Science Framework at the following link:
Participants read an instruction explaining the task with an example. The running example was an adjective not used in the stimulus set (i.e., You ask your friend John:
The prediction based on the face-management explanation of the polarity asymmetry of negative strengthening is that the negation of negative polarity antonyms will be more likely to be strengthened than the negation of positive polarity antonyms given the ‘non-ordinary’ nature of the contexts. That is, the prediction is a reversal of the traditional polarity asymmetry (as pre-registered at
The data were analysed using R (version 3.6.2). We excluded the data of four participants based on inconsistent responses in the filler trials (more than 50% responses not in line with the adjective used in the filler statements).
Mean degree of negative strengthening by polarity (Experiment 1). Error bars represent +/– 1 SEM.
All results were analysed with cumulative link mixed-effects models using the function clmm() in the ordinal package (
Summary of cumulative link mixed-effects model including the sum-coded fixed effect polarity (Experiment 1).
Estimate | SE | z-value | p-value | |
---|---|---|---|---|
Polarity | –0.26985 | 0.06164 | –4.378 | 0.000012 |
Following Ruytenbeek et al. (
Finally, as previous research suggests the existence of gender differences in negative strengthening (
The results of Experiment 1 disconfirmed the prediction, based on the face management explanation of the polarity asymmetry of negative strengthening, that the negation of negative antonyms would be more likely than the negation of positive antonyms to be strengthened in non-ordinary contexts. Despite the nature of the context, the results revealed a robust polarity asymmetry in the usual direction, with negated positive antonyms being strengthened to a higher degree than negated negative antonyms. This finding leads us to conclude that, when adjectival polarity and face-threatening potential compete with each other, adjectival polarity wins over face-threatening potential. The asymmetric interpretation of antonyms does not seem to be driven by their respective face-threatening potential, but adjectival polarity
Further, as Experiment 1 only included non-ordinary contexts (that is, context type was not experimentally manipulated), it is not possible to directly compare the strength of the polarity asymmetry across ordinary and non-ordinary contexts. In light of the results of Experiment 1, though, this comparison would be useful to determine whether face-management contributes to – even if it does not explain – the polarity asymmetry of negative strengthening. This question is addressed in Experiment 2.
We recruited 90 participants with US IP addresses on Mechanical Turk (45 participants across two experimental lists).
We used the same list of adjectives from Experiment 1 (see Appendix A). The experimental stimuli were constructed as in Experiment 1, for a total of 40 critical items. Target utterances, containing the negated antonym, were embedded in ordinary contexts, in which the bare negative poses a potential face-threat, or non-ordinary contexts, in which the bare positive poses a potential face-threat. For each antonymic pair, we constructed a minimal pair of contexts in which the manipulation of minimal linguistic material would turn the context from an ordinary to a non-ordinary one.
Example item for the adjective
Context: After thinking of a career change for a long time, you have decided to |
According to your colleague, the team is: |
happy 1 2 3 4 5 6 7 sad |
The task of the participants was the same as in Experiment 1. Our two factors, Polarity and Context, were all within-subject but spread across two different item lists in a Latin square design. Each participant saw 20 statements with positive and 20 statements with negative adjectives, rotated over context conditions. That is, each participant completed 40 critical trials. The resulting overall number of critical observations was 3560, hence the sample size was three times as large as that of Experiment 1. In addition to the critical items, participants were presented with 10 filler statements not involving negation, which also served as attention checks.
The experiment was programmed in HTML and run via MTurk’s built-in environment. The pre-registration form of the second experiment is available at the following link:
The procedure was the same as in Experiment 1.
Experiment 2 tests the contribution of face-management considerations to the asymmetric interpretation of negated antonyms. In what follows, we outline three alternative hypotheses on the relative contribution of face-threatening potential and adjectival polarity and their respective predictions (pre-registered at
These three alternative hypotheses were put forth to establish
The data were analysed using R (version 3.6.2). We excluded four participants based on inconsistent responses in the filler trials (more than 50% responses not in line with the adjective used in the filler statements).
Mean degree of negative strengthening by Polarity and Context (Experiment 2). Error bars represent +/– 1 SEM.
All results were analysed with cumulative link mixed-effects models using the function clmm() in the ordinal package. We included the sum-coded factors polarity, context, their interactions as well as random intercepts for items and participants. The results of the model showed a main effect of polarity with positive adjectives involving a higher degree of negative strengthening than negative ones (B = –.47, SE = .035, z = –13.32, p < .001). This finding replicates the polarity asymmetry discussed in previous work (e.g.,
Summary of cumulative link mixed-effects model including the sum-coded fixed effects polarity and context (Experiment 2).
Estimate | SE | z-value | p-value | |
---|---|---|---|---|
Polarity | –0.469497 | 0.035259 | –13.316 | 0.000000001 |
Context | –0.006586 | 0.034092 | –0.193 | 0.847 |
Polarity:Context | 0.014234 | 0.052891 | 0.269 | 0.788 |
Following Ruytenbeek et al. (
Finally, we computed a model with polarity and participant gender, as we had found such an interaction in Gotzner and Mazzarella (
Experiment 2 aimed at establishing the relative contribution of face-management considerations and adjectival polarity to the asymmetric interpretation of negated antonyms. The results confirmed the prediction of Hypothesis 3, that is, the hypothesis that this asymmetry is due to the polarity of the adjective, independently of its face-threatening potential. Not only does the face-management explanation of the asymmetric interpretation of negated antonyms not explain this asymmetry, but it appears not to contribute to it.
First, the results of Experiment 2 replicated those of Experiment 1. In the non-ordinary context condition, the traditional polarity asymmetry stands: negated positive antonyms were more likely to be strengthened than negated negative antonyms. Second, they confirm the robustness of Ruytenbeek et al.’s (
Taken together, the results of Experiment 1 and Experiment 2 provide strong evidence of an asymmetric interpretation of negated positive and negative antonyms. Crucially, this asymmetry holds across both ordinary and non-ordinary contexts. Hence, our results disconfirm the prediction of the face-management explanation of the polarity asymmetry that this asymmetry should be reversed in non-ordinary contexts (
Shall we conclude that face-management considerations are irrelevant when interpreting negated antonyms? This conclusion appears to be too strong. In Gotzner and Mazzarella (
In light of the results of the present study, proponents of the face-management explanation of the polarity asymmetry of negative strengthening could appeal to a
In the following, we consider two alternative explanations of this polarity asymmetry that we believe could be fruitfully explored in future research. A first alternative explanation, outlined by Ruytenbeek et al. (
A second alternative explanation focuses on the valence of the adjectival lexical meaning. This explanation, whose development is inspired by Terkourafi, Weissman and Roy (
The present study investigated the face-management explanation of the asymmetric interpretation of negated antonyms by testing one of its direct predictions (
We have discussed three further explanations of the polarity asymmetry of negative strengthening. The first explanation appeals to the existence of a pragmatic convention, originally based on face-management considerations, which would apply to the interpretation of negated adjectives independently of their actual face-threatening potential in context. The second explanation is based on the complexity of alternative expressions (
The additional file for this article can be found as follows:
Appendices. DOI:
For a characterisation of markedness that does not rely on morphology, see Vendler (
It is worth noting that in Giora et al.’s study participants were presented with pairs of target sentences (e.g.
These observations are in line with the results of an experiment by Gotzner and Kiziltan (
Note that the avoidance of overcommitment is also a face-saving strategy, targeted at preserving the speaker’s reputation as a trustworthy source of information (see, e.g.,
In contrast to Ruytenbeek et al.’s (
The function clmm() is the more recent variant of clmm2(), allowing for the implementation of multiple random effects. However, since the status of random slopes for ordinal models is debated, we only included random intercepts.
We note that this experiment contained half the number of critical observations compared to the Gotzner and Mazzarella’s (
In Experiment 2 we recruited 90 rather than 60 participants in order to maximize the possibility of observing differences across contexts and polarity. The choice of this bigger sample size was pre-registered.
To further verify our experimental manipulation, we asked an annotator (trained linguist, native English speaker and naive to the purpose of our study) to decide which contexts he considered ordinary vs. non-ordinary for the experimental material of Experiment 2. The interrater agreement between his annotation and ours was ϰ = 0.5. We then did a further analysis excluding the 6 contexts for which there was a divergence between our coding and the annotator’s coding. The analysis with the resulting subset of the data showed the same results as the main analysis, that is, a main effect of polarity (p < .0001) but no effect of context (p = .81), and no interaction between polarity and context (p = 0.71).
In contrast to this, Experiment 1 did not reveal any effects related to gender. Note that the sample size in Experiment in 1 was smaller compared to Experiment 2 (both in terms of the number of items and participants), which might explain the lack of significant gender effects.
We thank an anonymous reviewer for drawing our attention to this possibility.
An anonymous reviewer rightly pointed out that the Pollyanna hypothesis alone cannot explain why weak positive adjectives are more likely to be strengthened than strong positive adjectives. To do this, it would need to combine with considerations about adjectival markedness and/or frequency.
This work was supported by the German Research Foundation (DFG) as part of the
We thank the audience of the DegPol workshop at the Leibniz-ZAS (Berlin) for useful discussion of this work. We are grateful to Rick Nouwen for providing us with the idea underlying Experiment 2. We also thank Stanley Donahoo for annotating our stimuli and the three anonymous reviewers of
The authors have no competing interests to declare.