Correlation between memory recall and word similarity in sentence processing

Abstract

Regarding investigations on sentence processing, research is often done using complex models that use various parameters, which leads to opacity of models of human language. In addition, this causes the models to be extremely difficult to compare with other models from earlier experiments. With the intention of finding alternative features for computational models and to encourage further research, we aim to find a correlation between memory recall and word similarity in sentence processing. To test this theory, the outlines of a similar research of Smith & Vasishth (2020) were implemented, to investigate if reading times were affected by different word similarities regarding presupposition resolution. Reading times were subtracted from a memory recall experiment carried out by Jan Winkowski (currently unpublished) to compare reading data to cosine similarities between certain words in each sentence. Results indicate an unforeseen effect of word similarity on reading times, since reading times increased when word similarities became greater.