Estimating affective word covariates using word association data

Abstract

Word ratings on affective dimensions are an important tool in psycholinguistic research. Traditionally, they are obtained by asking participants to rate words on each dimension, a time-consuming procedure. As such, there has been some interest in computationally generating norms, by extrapolating words' affective ratings using their semantic similarity to words for which these values are already known. So far, most attempts have derived similarity from word cooccurrence in text corpora. In the current paper, we obtain similarity from word association data. We use these similarity ratings to predict the valence, arousal, and dominance of 14, 000 Dutch words with the help of two extrapolation methods: Orientation towards Paradigm Words and k-Nearest Neighbors. The resulting estimates show very high correlations with human ratings when using Orientation towards Paradigm Words, and even higher correlations when using k-Nearest Neighbors. We discuss possible theoretical accounts of our results and compare our findings with previous attempts at computationally generating affective norms.

The “Small World of Words” English word association norms for over 12,000 cue words

“…Likewise, estimates of similarity are also the key component in predicting other aspects of word meaning such as connotation based on valence, arousal and potency, concreteness or even ageof-acquisition. In these cases as well, our findings suggest that word associations often out-perform predictions based on the most recent text models Van Rensbergen et al, 2016;Vankrunkelsven et al, 2018) using a very sparse representation. More generally, we expect that these findings will be useful across a range of studies about psychological meaning, including priming studies and patient studies where semantic effects might be small and go undetected when the relatedness reflects distributional properties in external language.…”

The Small World of Words: English word association norms for over 12,000 cue words

“…Likewise, estimates of similarity are also the key component in predicting other aspects of word meaning such as connotation ENGLISH WORD ASSOCIATIONS 31 based on valence, arousal and potency, concreteness or even age-of-acquisition. In these cases as well, our findings suggest that word associations often out-perform predictions based on the most recent text models Vankrunkelsven, Verheyen, Storms, & De Deyne, 2018;Van Rensbergen, De Deyne, & Storms, 2016) using a very sparse representation. More generally, we expect that these findings will be useful across a range of studies about psychological meaning, including priming studies and patient studies where semantic effects might be small and go undetected when the relatedness reflects distributional properties in external language.…”

How well do similarity measures predict priming in abstract and concrete concepts?

“…This is particularly likely when a continuous word-association task is employed (i.e., when participants are asked to produce three associates for each cue stimulus) vs. a discrete word-association task (i.e., when participants are asked to produce just one associate), as in our case. Compared to lexical co-occurrence, association strength also is better at capturing words' affective features (Van Rensbergen, De Deyne, & Storms, 2016), which especially characterize abstract concepts (Kousta et al, 2011). A further asset of word association is that every association reflects a meaningful relation with a high signal-to-noise ratio, in contrast to co-occurrence in text corpora characterized by a low signal-to-noise ratio (Van Rensbergen et al, 2016).…”

“…Compared to lexical co-occurrence, association strength also is better at capturing words' affective features (Van Rensbergen, De Deyne, & Storms, 2016), which especially characterize abstract concepts (Kousta et al, 2011). A further asset of word association is that every association reflects a meaningful relation with a high signal-to-noise ratio, in contrast to co-occurrence in text corpora characterized by a low signal-to-noise ratio (Van Rensbergen et al, 2016). Associations are also clearly linguistically based, as they are provided in response to a word (in contrast to the standard feature-listing task which requires participants to explicitly consider the concept that the word refers to rather than the word itself), and reflect patterns of co-occurrence in semantic representation of both abstract and concrete concepts (Hoffman, McClelland, & Lambon Ralph, 2018).…”