Measuring perceptual and emotive dimensions of multi-word expressions: Can concreteness, emotional valence, and arousal be well estimated by some simple function of the constituent word ratings?

Abstract: Subjective ratings of dimensions of lexical meaning have long been used in experimental psychology and psycholinguistics―for example, in experimental studies of memory, lexical processing, and brain function. Three such dimensions of lexical meaning are concreteness (vs abstractness), emotional valence (degree of pleasantness), and arousal (degree of excitement). Ratings have typically been obtained by presenting lexical items to multiple respondents who rate each item on a Likert scale, after which the ratings for each item are averaged. For some dimensions of meaning and for some languages, ratings of many thousands of single words are freely available; but even for English there are as yet no remotely similar-size collections of ratings for multiword expressions (MWEs), such as collocations and spaced compound nouns. Researchers, including researchers of L2 vocabulary acquisition, may therefore wonder how well a MWE’s level of concreteness, valence or arousal can be estimated from the ratings of its constituent words. This article reports a study which addressed that question, concluding that MWE ratings derived from constituent word ratings must be used with caution. The study has doubled the e)xisting small stock of English MWEs rated for arousal and valence.

The data: The spreadsheet relates to seven correlational studies of which five concern concreteness and one each concern valence and arousal. The focal data in each substudy consist of a column of subjective ratings of (a) MWEs as wholes and (b) mean ratings, i.e., (rating of word 1 + rating of word 2) / 2.

The concreteness ratings mostly come from Brysbaert, Warriner, and Kuperman (2014) although for study two some of the word ratings come from the MRC Psycholinguistic Database (Coltheart, 1981; Wilson, 1988). The great majority of the MWE ratings of valence and arousal were collected by me through Amazon Mechanical Turk; some stem from Warriner, Kuperman, and Brysbaert (2013). All the word ratings stem from Warriner et al. An additional crucial variable in one study (Study 6, Valence) is the valence rating of the most-valenced constituent word--i.e., the word whole rating departs most in either direction from neutral, which is 5 on the 9-point Likert scale of the valence and arousal ratings. The concreteness ratings are on a 5-point scale. Among the data for valence and arousal are ratings for items for which WKB and AMT ratings are available. Correlations between these WKB and AMT items furnish some degree of validation. The data are described more fully in a soon-to-be-submitted article entitled, 'Measuring perceptual and emotive dimensions of multi-word expressions: Are constituent word ratings enough?'

Abbreviations: C-word = Constituent word; BWK = Brysbaert et al.; WKB = Warriner et al.; AMT = Amazon Mechanical Turk; MRC = The MRC Psycholinguistic Database

References

Brysbaert, M, Warriner, A, and Kuperman, V (2014) Concreteness ratings for 40,000 generally known English word lemmas. Behavior Research Methods 46: 904–11.

Coltheart, M (1981) The MRC Psycholinguistic Database, Quarterly Journal of Experimental Psychology 33A: 497–505.

Warriner A, Kuperman, V, and Brysbaert, M (2013) Norms of valence, arousal, and dominance for 13,915 English lemmas. Behavior Research Methods 45: 1191–207. List retrieved from: http://crr.ugent.be/archives/1003

Wilson, M. (1988). The MRC Psycholinguistic Database: Machine readable dictionary, Version 2, Behavioural Research Methods, Instruments and Computers 20: 6-11. Retrieved from: http://websites.psychology.uwa.edu.au/school/MRCDatabase/uwa_mrc.htm