Latent semantics of action verbs reflect phonetic parameters of intensity and emotional content

摘要

Conjuring up our thoughts, language reflects statistical patterns of wordco-occurrences which in turn come to describe how we perceive the world.Whether counting how frequently nouns and verbs combine in Google searchqueries, or extracting eigenvectors from term document matrices made up ofWikipedia lines and Shakespeare plots, the resulting latent semantics capturenot only the associative links which form concepts, but also spatial dimensionsembedded within the surface structure of language. As both the shape andmovements of objects have been found to be associated with phonetic contrastsalready in toddlers, this study explores whether articulatory and acousticparameters may likewise differentiate the latent semantics of action verbs.Selecting 3 x 20 emotion, face, and hand related verbs known to activatepremotor areas in the brain, their mutual cosine similarities were computedusing latent semantic analysis LSA, and the resulting adjacency matrices werecompared based on two different large scale text corpora; HAWIK and TASA.Applying hierarchical clustering to identify common structures across the twotext corpora, the verbs largely divide into combined mouth and hand movementsversus emotional expressions. Transforming the verbs into their constituentphonemes, the clustered small and large size movements appear differentiated byfront versus back vowels corresponding to increasing levels of arousal. Whereasthe clustered emotional verbs seem characterized by sequences of close versusopen jaw produced phonemes, generating up- or downwards shifts in formantfrequencies that may influence their perceived valence. Suggesting, that thelatent semantics of action verbs reflect parameters of intensity and emotionalpolarity that appear correlated with the articulatory contrasts and acousticcharacteristics of phonemes