Why is Semantic Change Asymmetric? The Role of Concreteness and Word Frequency and Metaphor and Metonymy

ABSTRACT

Metaphors and other tropes are commonly thought to reflect asymmetries in concreteness, with concrete sources being used to talk about relatively more abstract targets. Similarly, originating senses in diachronic semantic change have often been argued to be more concrete than extended senses. In this paper, we use a dataset of cross-linguistically attested semantic changes to empirically test the idea that asymmetries in figurative language are predicted by asymmetries in concreteness. We find only weak evidence for the role of concreteness and argue that concreteness is not a helpful notion when it comes to describing changes where both originating and extended senses are highly concrete (e.g., skin > bark, liver > lungs). Moreover, we find that word frequency data from English and other languages is a stronger predictor of these typologically common semantic changes. We discuss the implications of our findings for metaphor theory and theories of semantic change.

Acknowledgments

We thank Matthias Urban and Graham Thompson for help with earlier versions of this project (Winter, Thompson, & Urban, 2013). We thank Márton Sóskuthy for helping us out with the Google Translate API. We thank the Evolang 2013 conference organizers for awarding the James Hurford prize to this work. MS is supported by awards from the National Science Foundation (SBE-16302040) and the James S. McDonnell Foundation.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Notes

¹ We use “source” and “target” as synonymous with “originating sense” and “extended sense” throughout this paper, but we do not use these terms to imply that semantic changes necessarily reflects intentional, novel extensions of words on the part of speakers.

² At a reviewer’s suggestion, we assessed how well the cross-linguistic asymmetry was predicted by a measure of word prevalence (Brysbaert, Mandera, McCormick, & Keuleers, 2018), since prevalence of a word within a community may be a good measure of how much that meaning is in the common ground (and thus how intersubjectively accessible it is; Keuleers et al., 2015). We also assessed the role of other variables that could potentially be relevant for psycholinguistic processing, including comprehension speed from the English Lexicon Project and British Lexicon Project (Balota et al., 2007; Keuleers, Lacey, Rastle, & Brysbaert, 2012), sensory experience ratings (Juhasz & Yap, 2013), sensorimotor ratings (Lynott, Connell, Brysbaert, Brand, & Carney, 2019), emotional valence ratings (Warriner, Kuperman, & Brysbaert, 2013), age of acquisition ratings (Kuperman, Stadthagen-Gonzalez, & Brysbaert, 2012), and contextual diversity from the COCA (Davies, 2010) and BNC (McEnery et al., 2017). Computing effect size for each of these variables with Cohen’s d revealed that word frequency differences (log10 differences) most strongly separated cross-linguistic sources from targets (see online supplementary materials: https://osf.io/qukgh/).

³ The model formula was: brm(SourceVsTarget ~ Frequency + (1 + Frequency|ConceptPair) + (1 + Frequency|Language)); MCMC sampling was performed with 8,000 iterations for 4 chains (4,000 warmup samples). We set a weakly informative prior on the beta coefficient (a normal distribution centered at 0 with SD = 1) (cf. Lemoine, 2019). Convergence was successful (all Rhat values = 1.0).

⁴ Just as the demands of communication may constrain semantic change, so may the demands of language acquisition. The process of learning a new word for an abstract or subjective concept could be simpler if children have already learned a prior, meaning for that word that is more inter-subjective (Srinivasan & Carey, 2010; Starr, Cirolia, Tillman & Srinivasan, 2021). For example, if a child has already learned the spatial sense of high – which denotes a relatively inter-subjective concept – this could help them understand what long might mean when it is metaphorically used to refer to time. This predicts that metaphors that employ intersubjective concepts as source domains will be preferred because they scaffold the acquisition of otherwise difficult-to-learn word meanings.

Additional information

Funding

This work was supported by the National Science Foundation [SBE-16302040]; Bodo Winter was supported by the UKRI Future Leaders Fellowship [MR/T040505/1].