Responding to nonwords in the lexical decision task: Insights from the English Lexicon Project

Abstract

Researchers have extensively documented how various statistical properties of words (e.g., word frequency) influence lexical processing. However, the impact of lexical variables on nonword decision-making performance is less clear. This gap is surprising, because a better specification of the mechanisms driving nonword responses may provide valuable insights into early lexical processes. In the present study, item-level and participant-level analyses were conducted on the trial-level lexical decision data for almost 37,000 nonwords in the English Lexicon Project in order to identify the influence of different psycholinguistic variables on nonword lexical decision performance and to explore individual differences in how participants respond to nonwords. Item-level regression analyses reveal that nonword response time was positively correlated with number of letters, number of orthographic neighbors, number of affixes, and base-word number of syllables, and negatively correlated with Levenshtein orthographic distance and base-word frequency. Participant-level analyses also point to within- and between-session stability in nonword responses across distinct sets of items, and intriguingly reveal that higher vocabulary knowledge is associated with less sensitivity to some dimensions (e.g., number of letters) but more sensitivity to others (e.g., base-word frequency). The present findings provide well-specified and interesting new constraints for informing models of word recognition and lexical decision.

Figure 1

Baseword frequency effects as a function of number of letters (left), Levenshtein orthographic distance (center), and response times (right). ***p< .001

Figure 2

Interactions between baseword frequency and number of letters (upper left), baseword number of syllables (lower left), orthographic neighborhood size (upper right), and number of affixes (lower right). The bars represent the standardized regression coefficient for each variable as a function of low-, medium-, and high-frequency words. Error bars denote standard errors.

Figure 3

Distributions of standardized regression coefficients across participants as a function of lexical variable.

Figure 4

Scatterplots (with 95% confidence intervals) between vocabulary knowledge and nonword response times (left) and accuracy (right).

Figure 5

Scatterplots (with 95% confidence intervals) between vocabulary knowledge and participant-level effects.

Figure 6

Scatterplots (with 95% confidence intervals) between boundary separation and participant-level effects.

Figure 7

Scatterplots (with 95% confidence intervals) between nondecision time and participant-level effects.

Figure 8

Scatterplots (with 95% confidence intervals) between nonword drift rate and participant-level effects.