Polysemy and SSD model – Testing model predictions on sense uncertainty effects

Abstract

Large body of empirical work shown that words with multiple related senses (polysemy) are processed faster than unambiguous words, whereas ones with multiple related meanings (homonymy) are processed slower. At the same time, other empirical findings showed inconsistencies where the pattern of results changes, depending on the tasks used in the experiment. To account for this, Semantic Settling Dynamics (SSD) model was developed (Armstrong & Plaut, 2016) which relies on the hypothesis that different amounts of semantic demands between experimental tasks are the cause of varying effects. Model predicts that polysemous words produce a large effect in early phases that decrease with further processing. Therefore, in our two experiments, we tried to test this prediction by prolonging semantic processing of polysemous words described by entropy – a measure that combines number of senses with the balance of probabilities of those senses. In the first experiment, we presented a standard visual lexical decision task (LDT) as a baseline and a modified LDT with lower stimulus-background contrast as a slowed condition. This produced a standard entropy effect (direct correlation between entropy and RT) but did not produce a sufficient response delay in order for the effect to change. Second experiment employed a comparison of visual (baseline) and auditory (slowed) LDT and found the change the entropy effect, as predicted. However, the direction of the slope change was inverse to what the model predicted: instead of a smaller effect in slowed condition compared to baseline, we got no effect in baseline and a large polysemy effect in slowed condition. These results implicate that SSD model is able to predict general modulation of entropy effect incurred by the prolonged processing time. However, the predictions are not precise enough, and we believe that this limitation is a consequence of the absence of a reliable testing method.