Number of senses and semantic settling dynamic model – speed/accuracy feedback as longer processing induction strategy

Abstract

Previous semantic ambiguity research typically found processing advantage for polysemous words (multiple related senses) and a disadvantage for homonymous words (multiple unrelated meanings), compared to unambiguous controls (Rodd, Gaskell, & Marslen-Wilson, 2002). However, any task change elicited different effects, so the traditional account needed revision. Semantic Settling Dynamics 74 model was developed in order to account for the observed effect variations (Armstrong & Plaut, 2016). Basic assumption of the model was that the apparent inconsistencies arose as a consequence of between-task differences in time spent in semantic processing. Shorter processing resulted in polysemy advantage and longer processing in weaker polysemy advantage and homonymy disadvantage. To test this prediction, authors applied a contrast reduction manipulation, where a low contrast condition was expected to prolong processing. In a factorial design (homonymy and polysemy compared at a group level) this manipulation was of a limited success. In an attempt to generalize model predictions to a finer measure of ambiguity, we have previously attempted employing the same manipulation in a correlational design where participants were presented with polysemous words (1-18 senses). Results showed number of senses effect, but the low contrast prolonging was marginal and had no influence on semantics. This indicated a need for an alternative manipulation which in this case was giving speed and accuracy feedback to participants compared to cases where such feedback was absent. This was expected to slow down average reaction time between groups, but its effect on the number of senses was previously unknown. We presented 160 words with 1-18 senses and 160 pseudowords. Visual lexical decision task was administered to 71 participants, divided into two groups – one that received speed/accuracy feedback after trials and the one that did not. Linear mixed effect regression revealed a significant number of senses effect (t(140.14)=-2.339, p=.021), with no significant differences between two feedback conditions. No interaction between feedback conditions and number of senses was found. Considering the manipulation failed to prolong overall reaction time and as a consequence, semantic processing, it is difficult to rely on these results in evaluation of the model. Further research is required to achieve significant slowdown and decisively assess SSD’s ability to predict effect variation on a finer measure of ambiguity.