The development of real-time spoken and word recognition derives from changes in ability, not maturation

Abstract

In typical adults, recognizing both spoken and written words is thought to be served by a process of competition between candidates in the lexicon. In recent years, work has used eye-tracking in the visual world paradigm to characterize this competition process over development. It has shown that both spoken and written word recognition continue to develop through adolescence (Rigler et al., 2015). It is still unclear what drives these changes in real-time word recognition over the school years, as there are dramatic changes in language, the onset of reading instruction, and gains in domain general function during this time. This study began to address these issues by asking whether changes in real-time word recognition derive from changes in overall language and reading ability or reflect more general age-related development. This cross-sectional study examined 278 school-age children (Grades 1-3) using the Visual World Paradigm to assess both spoken and written word recognition, along with multiple measures of language, reading and phonology. A structural equation model applied to these ability measures found three factors representing language, reading, and phonology. Multiple regression analyses were used to understand how these three factors relate to real-time spoken and written word recognition as well as a non-linguistic variant of the VWP intended to capture decision speed, eye-movement factors, and other non-language/reading differences. We found that for both spoken and written word recognition, the speed of activating target words in both domains was more closely tied to the relevant ability (e.g., reading for written word recognition) than was age. We also examined competition resolution (how fully competitors were suppressed late in processing). Here, spoken word recognition showed only small, developmental effects that were only related to phonological processing, suggesting links to developmental language disorder. However, in written word recognition, competitor resolution showed large impacts of development which were strongly linked to reading. This suggests the dimensionality of real-time lexical processing may differ across domains. Importantly, neither spoken nor written word recognition is fully described by changes in non-linguistic skills assessed with non-linguistic VWP, and the non-linguistic VWP was linked to differences in language and reading. These findings suggest that spoken and written word recognition continue past the first year of life and are mostly driven by ability and not only by overall maturation.

Keywords: Eye-tracking; Reading; SEM; School-age language development; Spoken word recognition; Written word recognition.

Figure 1.

A) Fixations to the target, cohort, rhyme and unrelated competitors over time for all children in this study in a spoken word VWP. B) Fixations to targets and competitors for all children in this study in a written word VWP task. C) Schematic looks to the target and cohort illustrating changes in activation rate (a key dimension associated with development in prior work; D) Schematic pattern of fixations illustrating changes in competition resolution (a key dimension associated with differences in language ability within an age.

Figure 2.

A) Schematic display of the non-linguistic VWP. Here, the central shape (the double triangle) is the stimulus and must be mapped to one of the four options (in this case, to the bottom-right one). The chevron in the upper left is a color competitor while the other two shapes are unrelated. B) Fixations to the target, color competitor and unrelated objects as a function of time averaged across all the children in this study.

Figure 3.

Distribution of participants by ability level. For each domain, we computed standard scores and averaged all the measures within that domain for a rough estimate of ability. A) Oral Language; B) Reading; C) Phonological Processing (excluding Rapid Automatized Naming).

Figure 4.

The final measurement models. Coefficients associated with each pathway are standardized. Coefficients for manifest/manifest covariance terms are not shown, but can be found on the OSF repository at: https://osf.io/ywfqa/?view_only=7eacaf0a41744aebb8347348c66b21fc

Fig. 5.

Sequence of events in the Spoken Word Recognition VWP.

Figure 6.

Sequence of events in the Written Word Recognition VWP.

Figure 7.

A/B) Functions used for fitting the fixation data. For each is shown the relevant free parameters, along with secondary functions that illustrate the effect of varying one or more of those parameters. A) A four-parameter logistic for target fixations; B) An asymmetric Gaussian for competitor fixations. C/D) Indices derived from those parameters. C) Activation rate is an average of the logged z-scores of the target slope and the target crossover. D) Competition Resolution reflects the degree to which the final asymptotes of the target (purple) separate from those of the cohort (green) and unrelated (gray): Target Maximum – (Cohort Offset Baseline + Unrelated Offset Baseline) / 2.

Figure 8.

Overview of fixations in the spoken word VWP. A) Fixations to Target as a function of time and grade. B) Cohort fixations. C) Cohort fixations after subtracting fixations to the unrelated item indexes unique activation for cohorts that is not just due to general looking. D) Rhyme fixations. E) Rhyme minus Unrelated fixations.

Figure 9.

Overview of fixations in the written word VWP. A) Fixations to Target as a function of time and grade. B) Cohort fixations. C) Cohort fixations after subtracting fixations to the unrelated item indexes unique activation for cohorts that is not just due to general looking. D) Rhyme fixations. E) Rhyme minus Unrelated fixations.

Figure 10.

A-D shows the relationship between Activation Rate in the Spoken Word VWP with Age, Language, Reading, and Phonology. Note that two outliers are not shown (they were out of range of the axies). Panels E-H represent the same predictors with Activation Rate in the Written Word VWP.(one outlier not shown).

Figure 11.

Results of commonality analyses on the effects of age and ability on activation rate for a) spoken word recognition and b) written word recognition. Shown is the proportion of variance accounted for by each factor. In each figure, the univariate effect of age (shared + unique variance) is shown first. Stacked bars parcel the unique and shared variance for age and ability. *: p<.05; +: p<.1. No significance test is possible for shared variance.

Figure 12.

A-D shows the relationship between Competition Resolution in the Spoken Word VWP with Age, Language, Reading, and Phonology. Panels E-H represent the same predictors with Competition Resolution in the Written Word VWP (one outlier not shown).

Figure 13.

Results of commonality analyses on the effects of age and ability on competitor resolution for a) spoken word recognition and b) written word recognition. Shown is the proportion of variance accounted for by each factor. In each figure, the univariate effect of age (shared + unique variance) is shown first. Stacked bars parcel the unique and shared variance for age and ability. *: p<.05; +: p<.1. No significance test is possible for shared variance.

Figure 14.

Performance in the nlVWP. A) Fixations to the target as a function of time and grade. B) Fixations to the color competitor as a function of time and grade. Note that in both, the x axis is truncated to 1500 msec to show the smaller effects; the analysis window was 0 to 2000 msec.

Figure 15.

Results of commonality analyses on the effects of performance in the nlVWP and ability on a) activation rate for spoken word recognition; b) activation rate for written word recognition; c) competition resolution for spoken word recognition; and d) competition resolution for written word recognition. Shown is the proportion of variance accounted for by each factor. In each panel, the univariate effect of age (shared + unique variance) is shown first. Stacked bars parcel the unique and shared variance for age and ability. *: p<.05; +: p<.1. No significance test is possible for shared variance.

Figure 16.

Results of commonality analyses on the effects age and latent abilities on two dimensions of performance in the nlVWP and ability on a) activation rate; b) resolution. Shown is the proportion of variance accounted for by each factor. In each figure, the univariate effect of age (shared + unique variance) is shown first. Stacked bars parcel the unique and shared variance for age and ability. *: p<.05; +: p<.1. No significance test is possible for shared variance.