English. In this paper, we investigate the relation between negated adjectives and antonyms in English using Distributional Semantics methods. Results show that, on the basis of contexts of use, a negated adjective (e.g., not cold) is typically more similar to the adjective itself (cold) than to its antonym (hot); such effect is less strong for antonyms derived by affixation (e.g., happy - unhappy). Italiano. In questo lavoro, analizziamo la relazione fra aggettivi negati e antonimi in inglese utilizzando metodi di Semantica Distribuzionale. I risultati mostrano che, sulla base dei contesti di uso, la negazione di un aggettivo (ad es. “not cold”; it.: “non freddo”) e` tipicamente piu` simile all'aggettivo stesso (“cold”; it.: “freddo”) che al suo antonimo (“hot”; it.: “caldo”). Tale effetto e` meno accentuato per antonimi derivati tramite affissi (ad es. “happy”-“unhappy”; it.: “felice”“infelice”).
Negation has long represented a challenge for theoretical and computational linguists (see Horn (1989) and Morante and Sporleder (2012) for overviews): in spite of the relative simplicity of logical negation (:p is true $ p is false), complexity arises when negation interacts with morphosyntax, semantics and pragmatics.
In this work, we focus on the negation of adjectives in English, expressed by the particle not modifying an adjective, as in not cold. A na¨ıve
Part of the work presented in this paper was carried
out while the first author was at the University of Amsterdam.
account of these expressions would be to equate
them to antonyms, and hence take them to
convey the opposite of the adjective (e.g., not cold =
hot). In fact, this simplifying assumption is
sometimes made in computational approaches which
model negation as a mapping from an adjective to
its antonym (e.g., The Pham et al., (2015), Rimell
et al., (2017)). However, a range of studies
support what is known as mitigation hypothesis
In this work, we investigate negated
adjectives from the perspective of Distributional
Semantics
Our analyses show that, when considering distributional information, negated adjectives are more similar to the adjective itself than to the antonym (e.g., not cold is closer to cold than to hot), regardless of the type of antonym or of negation. However, we find that morphological antonymy is closer to negation than lexical one is. 2
We are interested in how negation acts with respect
to pairs of adjectives connected by the lexical
relation of antonymy
not adj.
# triples
Lexical antonyms
– contrary
– contradictory
Morphological antonyms
– simple negations
– double negations
difference might affect their relation with negated
adjectives: indeed, affixal negations have a
morphological structure that resembles negated
adjectives (e.g., un-happy vs. not happy). For this
reason, we keep triples derived from lexical and
morphological antonyms distinct, and compare them
in our analyses: in particular, we are interested
in testing whether in a distributional space
negation tends to be more similar to morphological
antonymy than to lexical one. Besides this
comparison, we apply other distinctions to the triples
obtained with lexical and morphological antonyms
respectively, in order to investigate further effects.
Contrary vs. contradictory Lexical antonyms
have been classified as either contradictory or
contrary
Since no data annotated with respect to this distinction is available, the three authors independently annotated the antonym pairs in the dataset as either contrary, contradictory or unclear, following the definition used by Fraenkel and Shul (2008).2 Not surprisingly, the interannotator agreement is only moderate (Fleiss’ k = 0:37): already Fraenkel and Shul (2008) noted that even for what they considered contradictory pairs it is possible to conceive a mid-value interpretation (e.g., not dead half-dead; Paradis and Willners (2006)). This suggests that the contrary 1In the dataset, the former are coded as regular antonyms and the latter as direct affixal negations.
2Annotation guidelines at https://lauraina. github.io/data/notadj.pdf vs. contradictory distinction involves a continuum rather than a dichotomy. We leave this aspect to be further clarified by future research and, for the purpose of our analysis, only consider pairs classified with full agreement.
Simple vs. double negation In the case of morphological antonyms, one of the two adjectives is an affixal negation, and hence already contains a negating prefix (such as un- in unhappy): adding not thus gives rise to a double negation (e.g., not unhappy). These expressions have been widely studied in the literature due to their difference with double negation in logic (e.g., Bolinger (1972), Krifka (2007) and recently Tessler and Franke (2018)). While in logic two negations cancel each other out (::p p), in natural language double negations are typically employed to weaken the meaning of the adjective that is negated twice (e.g., not unhappy 6= happy) . Our goal is to test whether evidence for this effect is found in a distributional space: in particular, if two negations were to cancel each other out then the negation of an affixal negation (e.g., not unhappy) should be particularly close to the antonym (e.g., happy). We then test whether simple (e.g., not happy) and double (e.g., not unhappy) negations exhibit similar trends in relation to an antonym pair (happy vs. unhappy). 3 3.1
Previous studies about negation of adjectives
described its effect as a meaning shift from the
adjective towards the antonym, that can be measured in
terms of semantic similarity
To this aim, we build a distributional semantic
model with standard techniques, but whose
vocabulary includes, besides word units, also negated
adjectives. In practice, each occurrence of a
negated adjective (adjacent occurrence of not and
an adjective without intervening words; e.g., we
exclude cases like not very cold) is treated as a
single and independent token (e.g., not cold ;
not cold). With this pre-processing, we train a
word2vec CBOW model
For our analyses, we consider triples as
those described in Section 2. Given a triple
hai; aj ; not aii (e.g., cold, hot, not cold), we
define the following score:
(3) Shift := Sim(not ai; aj )
Sim(not ai; ai)
where i6=j, and Sim(not ai; aj ) and Sim(not ai; ai)
are the cosine similarities of the negated adjective
with the antonym and the adjective, respectively.
This measures how much closer a negated
adjective is to the antonym than to the adjective (i.e.,
how much closer not cold is to hot than to cold),
and hence how much negation shifts the
meaning of an adjective towards that of the antonym.
Due to the well-known tendency of antonyms to be
close in a distributional space
From the data introduced in Section 2, we only consider triples where each of the three elements occurs at least 100 times in the training corpus of the distributional model. Table 1 shows the number of triples considered for each class and the average frequency of adjectives and negated adjectives.6 The number of contradictory triples is 3Gensim implementation.
4Vectors size: 400; window size: 5; minimum frequency: 20; sample: 0.005; negative samples: 1.
5Spearman’s of 0.75 on the MEN dataset
6Negated adjectives are overall less frequent than their non-negated counterparts, as shown in Table 1. small due to the choice of keeping only antonyms for which we had full agreement in the annotation; double negations triples are few due to the limited frequency of these expressions in the corpus.7 Lexical vs. morphological antonyms The average Shift scores of both classes are negative, showing that a negated adjective is typically closer to the adjective than to the antonym. Indeed, as shown in Table 3, the nearest neighbor of a negated adjective is often the related adjective. On one hand, this could be seen as supporting the idea that negated adjectives express an intermediate meaning between that of the adjective and the antonym (e.g., not small is close to normal-sized). More in general, it shows that negated adjectives have a profile of use that is more similar to that of the adjective than to the antonym.
The two classes of antonyms differ significantly in the extent of this effect: negated adjectives are closer to a morphological antonym than a lexical one (e.g., not perfect vs. imperfect, not wide vs. narrow). Such similarity in distribution can be explained by the similarity in structure, and hence possibly in meaning, of negated adjectives and affixal negations. Yet, in spite of the higher similarity in use, affixal negation still does not seem equivalent to negation by not, due to the negative average Shift value.
trast to the results from the linguistic literature (see Section 2), the behavior of contrary and contradictory antonym pairs is not significantly different in our analysis. When we look into a distributional space, even for contradictory antonyms, the negated adjectives tend to be more similar to the adjective itself than to the antonym.
This result points at the fact that distributional similarity is capturing a different type of similarity from that considered in the experiments of Fraenkel and Shul (2008). We cannot thus directly interpret our results as just a product of the mitigating aspect of negation. Distributional information may discriminate between the negation of 7Full list of triples at https://lauraina. github.io/data/notadj.pdf an adjective and the antonym, even when the two seem intuitively equivalent (e.g., not dead is closer to dead than to alive): indeed, the use of one or the other may serve different functions (e.g., contradicting an expectation, politeness, etc.), leading them to appear in different contexts. Moreover, we find that, since continuous representations are able to capture nuanced differences, the alleged dichotomy between contrary and contradictory antonyms may become a continuum in distributional space: for example, one of the closest adjectives to not dead is half-dead. This further underscores the difficulty in distinguishing between contrary and contradictory antonyms which we had already encountered in the annotation. Simple vs. double negations There is not a significant difference between negated adjectives that are instances of simple and double negations: crucially, it is not the case that double negations are very close to the antonym as a result of the two negations canceling each other out (e.g., not unhappy is closer to unhappy than to happy).
As before, the result cannot be interpreted only in terms of mitigation (though, e.g., not unhappy is close to unimpressed, hence a mid-value between happy and unhappy). In general, it suggests that the contexts of use of double negations are more similar to the ones of the adjective that is negated than to those of its antonym. Indeed, double negations typically appear in contexts where the use of the “logically” equivalent alternative (i.e., the antonym) is to be avoided for pragmatic reasons, as possibly too strong or direct (e.g., not unproblematic vs. problematic; Horn, (1984)). 4
We have investigated negated adjectives using the tools of Distributional Semantics, which allows us to quantify the similarities between expressions on the basis of how they are used. Our analyses show that, when considering contexts of occurrence, negating an adjective does not make it closer to the antonym than to the adjective itself. This can be seen as a result of the various functions of negation (e.g., mitigation, contradiction to an expectation, politeness) that may lead to different patterns of use for negated adjectives and antonyms. Further research may shed light on which type of contexts actually discriminate them, for example through a corpus study, and which other properties negated adjectives have in a distriLexical antonyms Contrary antonyms Simple negations butional space, such as their interaction with scalar dimensions (e.g., not hot vs. freezing, cold, lukewarm, hot etc.; Wilkinson and Tim (2016)). Finally, while for the purpose of this study we opted for a standard word2vec model, one could test for the same effects with differently obtained distributional vectors.
Despite its current limitations in covering truthrelated aspects of meaning, Distributional Semantics was shown by Kruszewski et al. (2017) to be apt to model at least some aspects of negation, especially if graded in nature, such as alternativehood. Our study provides supporting evidence for this line of research and in addition points at the utility of using Distributional Semantics to uncover nuanced differences in use between a negation and other expressions, even when logically equivalent. Moreover, we regard our results to be of general interest for the NLP community, since effects of negation like the ones we studied and how they are represented in a distributional space can be critical for tasks like sentiment analysis (e.g., what does it imply that a costumer is not happy or not unhappy with a product?; Wiegand et al, (2010)).
This project has received funding from the European Research Council (ERC) under the European Union’s Horizon 2020 research and innovation programme (grant agreement No 715154), and by the Catalan government (SGR 2017 1575). This paper reflects the authors’ view only, and the EU is not responsible for any use that may be made of the information it contains.