Abstract
The question whether the representation of numbers is abstract and notation-independent still sparks a fervent debate in the numerical cognition field. Here, we employed distributional semantic models to quantify the distributional history of Arabic digits and number words in natural language (i.e., from large-scale written corpora). In a first computational experiment, we demonstrated that the distributional history of numbers in language reflected distance effect-like relationships for both numerical notations. Next, in a second behavioral experiment, we used these specific distributional histories as notation-dependent experiential priors to predict participants' performance in comparison tasks employing both Arabic digits and number words. We observed that the linguistic experiential priors are not only constituting better models for behavioral performance than the real numerical distance, but we also found evidence for (partial) notation-dependent behavior. That is, our results showed that notation-dependent linguistic priors accounted for specific notation-dependent behaviors. These observations suggest that the specific distributional pattern of Arabic digits and number words in natural language reflect the way humans represent numbers, thus supporting an experiential-driven representation.