Semantic Settling Dynamics model and number of senses

Abstract

Large body of empirical investigation revealed that lexical ambiguity effects have been unstable. Compared to unambiguous words, polysemy (multiple related senses) typically elicited shorter processing latencies, whereas homonymy (multiple unrelated meanings) prolonged processing. However, depending on the task used in the experiment, the effects have varied. Semantic Settling Dynamics model was recently developed to account for these variations by assuming that they originated from differences in the amount of semantic processing, closely linked to processing time that is evoked by the specific task. According to this model, tasks that elicit shorter processing time favour the polysemy advantage, whereas longer processing is needed for homonymy effect to become observable. To test this prediction, the authors of the model applied manipulation that was aimed to prolong processing by presenting stimuli in high and low contrast, low contrast leading to 100ms longer processing and more pronounced homonymy disadvantage, followed by less pronounced polysemy advantage. However, this was demonstrated only for comparison of polysemous words and unambiguous words at the group level – in a factorial design. In this research we aimed to test whether the predictions could be generalized to number of senses effect (the finding that an increase in number of senses is followed by a decrease in processing time), in a correlational design. Considering that the model predicts that prolonged processing would lead to attenuation of polysemy effect, we predicted that prolonged processing would also attenuate the number of senses effect. We presented 160 words with 1-18 senses and 160 pseudowords. Visual lexical decision task was administered to 84 participants, divided into two groups. One group was presented with white stimuli on a black background (high contrast), whereas the same items were presented to the other group as dark-grey on a black background (low contrast). Linear mixed effect regression revealed marginally slower processing in a low contrast condition (t(72) = 1.756, p = .083) and significant effect of number of senses (t(138) = -2.099, p = .037). We did not observe interaction between number of senses and contrast, but there was a marginally significant word length by contrast interaction (t(10, 80) = -1.857, p = .063). This interaction revealed that low contrast tends to attenuate the effect of longer words taking longer to process. Overall, our results contradict predictions derived from the SSD model and suggest further research. However, one should bear in mind that our contrast manipulation prolonged the processing time by just 40ms, less than the authors of the model reported. Therefore, the same predictions should be tested by using alternative strategies of inducing longer processing.