Obtaining Reliable Human Ratings of Valence, Arousal, and Dominance for 20,000 English Words

Mohammad, Saif M.

doi:10.18653/v1/p18-1017

Cited by 459 publications

(358 citation statements)

References 53 publications

(48 reference statements)

Supporting

Mentioning

303

Contrasting

Unclassified

Order By: Relevance

“…We focus on an objective corpus‐based approach, to avoid such potential criticisms. Second, in a similar vein, we decided to use an emotional model that learns affective information indirectly, by predicting the co‐occurrence of emojis and text in a corpus, rather than using emotional representations derived directly from valence, arousal, and dominance norms (Mohammad, 2018; Warriner et al, 2013). This also increases the coverage of our model.…”

Section: Discussionmentioning

confidence: 99%

“…The model is very different from the one by De Deyne et al (2018), which was constructed by concatenating valence, arousal, and potency ratings, for men and women separately (i.e., six dimensions), from the study by Warriner, Kuperman, and Brysbaert (2013), with valence, arousal, and dominance ratings, from the study by Mohammad (2018). DeepMoji provides better representations for our purposes than ratings because first, a model trained over a corpus of tweets, rather than subjective ratings, makes the emotion model more comparable to the linguistic and visual models, both trained over corpora.…”

Section: Methodsmentioning

confidence: 97%

“…To obtain a more detailed understanding of the emotional information captured by the DeepMoji model, we used PCA and extracted the first 10 principal components from the model representations. Then, we computed the (absolute) Spearman correlations between each component and the affective norms collected by Mohammad (2018), covering subjective ratings of word valence, arousal, and dominance. Our analysis included 13,678 words common to both the model and the norms.…”

Section: Methodsmentioning

confidence: 99%

See 2 more Smart Citations

Constructing Semantic Models From Words, Images, and Emojis

Rotaru

Vigliocco

2020

Cognitive Science

View full text Add to dashboard Cite

A number of recent models of semantics combine linguistic information, derived from text corpora, and visual information, derived from image collections, demonstrating that the resulting multimodal models are better than either of their unimodal counterparts, in accounting for behavioral data. Empirical work on semantic processing has shown that emotion also plays an important role especially in abstract concepts; however, models integrating emotion along with linguistic and visual information are lacking. Here, we first improve on visual and affective representations, derived from state-of-the-art existing models, by choosing models that best fit available human semantic data and extending the number of concepts they cover. Crucially then, we assess whether adding affective representations (obtained from a neural network model designed to predict emojis from co-occurring text) improves the model's ability to fit semantic similarity/relatedness judgments from a purely linguistic and linguistic-visual model. We find that, given specific weights assigned to the models, adding both visual and affective representations improves performance, with visual representations providing an improvement especially for more concrete words, and affective representations improving especially the fit for more abstract words.

show abstract

Section: Discussionmentioning

confidence: 99%

Section: Methodsmentioning

confidence: 97%

Section: Methodsmentioning

confidence: 99%

See 1 more Smart Citation

Constructing Semantic Models From Words, Images, and Emojis

Rotaru

Vigliocco

2020

Cognitive Science

View full text Add to dashboard Cite

show abstract

“…We focus on an objective corpus-based approach, to avoid such potential criticisms. Secondly, in a similar vein, we decided to use an emotional model that learns affective information indirectly, by predicting the co-occurrence of emojis and text in a corpus, rather than using emotional representations derived directly from valence, arousal and dominance norms (Mohammad, 2018;Warriner, Kuperman, & Brysbaert, 2013). This also increases the coverage of our model.…”

Section: Discussionmentioning

confidence: 99%

“…The model is very different from the one by De Deyne et al (2018), which was constructed by concatenating valence, arousal, and potency ratings, for men and women separately (i.e., 6 dimensions), from the study by Warriner, Kuperman, and Brysbaert (2013), with valence, arousal, and dominance ratings, from the study by Mohammad (2018). DeepMoji provides better representations for our purposes than ratings because firstly, a model trained over a corpus of tweets, rather than subjective ratings, makes the emotion model more comparable to the linguistic and visual models, both trained over corpora.…”

Section: Emotion Modelmentioning

confidence: 94%

Constructing Semantic Models From Words, Images, and Emojis

Rotaru¹,

Vigliocco²

2020

Preprint

View full text Add to dashboard Cite

A number of recent models of semantics combine linguistic information, derived from text corpora, and visual information, derived from image collections, demonstrating that the resulting multimodal models are better than either of their unimodal counterparts, in accounting for behavioural data. Empirical work on semantic processing has shown that emotion also plays an important role especially for abstract concepts, however, models integrating emotion along with linguistic and visual information are lacking. Here, we first improve on visual and affective representations, derived from state-of-the-art existing models, by choosing models that best fit available human semantic data and extending the number of concepts they cover. Crucially then, we assess whether adding affective representations (obtained from a neural network model designed to predict emojis from co-occurring text) improves the model’s ability to fit semantic similarity/relatedness judgements from a purely linguistic and linguistic-visual model. We find that, given specific weights assigned to the models, adding both visual and affective representations improve performance, with visual representations providing an improvement especially for more concrete words, and affective representations improving especially the fit for more abstract words.

show abstract

Representational similarity analysis of self‐ versus other‐processing: Effect of trait aggressiveness

Knyazev,

Savostyanov,

Bocharov

et al. 2024

Aggressive Behavior

View full text Add to dashboard Cite

In this study, using the self/other adjective judgment task, we aimed to explore how people perceive themselves in comparison to various other people, including friends, strangers, and those they dislike. Next, using representational similarity analysis, we sought to elucidate how these perceptual similarities and differences are represented in brain activity and how aggressiveness is related to these representations. Behavioral ratings show that, on average, people tend to consider themselves more like their friends than neutral strangers, and least like people they dislike. This pattern of similarity is positively correlated with neural representation in social and cognitive circuits of the brain and negatively correlated with neural representation in emotional centers that may represent emotional arousal associated with various social objects. Aggressiveness seems to predispose a person to a pattern of behavior that is the opposite of the average pattern, that is, a tendency to think of oneself as less like one's friends and more like one's enemies. This corresponds to an increase in the similarity of the behavioral representation with the representation in the emotional centers and a decrease in its similarity with the representation in the social and cognitive centers. This can be seen as evidence that in individuals prone to aggression, behavior in the social environment may depend to a greater extent on the representation of social objects in the emotional rather than social and cognitive brain circuits.

show abstract

Obtaining Reliable Human Ratings of Valence, Arousal, and Dominance for 20,000 English Words

Cited by 459 publications

References 53 publications

Constructing Semantic Models From Words, Images, and Emojis

Constructing Semantic Models From Words, Images, and Emojis

Constructing Semantic Models From Words, Images, and Emojis

Representational similarity analysis of self‐ versus other‐processing: Effect of trait aggressiveness

Contact Info

Product

Resources

About