Temporal dynamics of polysemous verb processing

Abstract

Semantic Settling Dynamics (SSD) model predicts that polysemy effect would change as a function of time spent in processing (Armstrong & Plaut, 2016). In early processing, the processing advantage would be large (compared to unambiguous words), whereas later, the effect would decrease. Research in Serbian described polysemy by quantitatively expressing the magnitude of the ambiguity of a particular word through measures such as number of senses (NoS). Such experiments revealed a facilitatory effect of polysemy in visual lexical decision task (vLDT; Filipović Đurđević, 2007). In SSD framework, this effect would be the equivalent of the early processing. Once the processing is prolonged, we expected reduction of this effect, similar to previous findings comparing polysemy to unambiguous words. This was demonstrated in experiments that employed the prolonging strategy of comparing vLDT (early processing) and auditory lexical decision task (aLDT; late processing). When the stimuli were nouns, we found the expected facilitatory polysemy effect in vLDT and null effect in aLDT (Mišić & Filipović Đurđević, 2019). Compared to the model predictions and the previous research, this loss of effect in later processing was unexpected. Therefore, the aim of this research was twofold – to extend the SSD testing on polysemous verbs and to replicate the results we found on nouns. We tested whether the NoS effect would change in the late processing by using the same prolonging strategy as in previous experiments. Two groups of participants (71 in vLDT, 75 in aLDT) were presented with 100 polysemous verbs in two versions of lexical decision task – visual and auditory. NoS was estimated based on the senses listed in the Matica Srpska dictionary (2007). Linear mixed modelling revealed that the aLDT was significantly slower than the vLDT. Marginal facilitatory NoS effect was found in vLDT (β = -.020, S.E. = .010, df = 100.07, t = -1.934, p = .056). The interaction with task modality was also marginally significant (β = .019, S.E. = .010, df = 98.84, t = -1.871, p = .064). Even though the NoS effect was marginally significant in both the baseline and the prolonged processing condition, the trends point to the same pattern of the results as the ones previously found. Early processing reveals a facilitatory effect, while in later processing, that effect is attenuated. This is a partial replication of the findings from the noun experiment and together they converge to the conclusion that the SSD model cannot give precise predictions regarding the effect change.