The inhibitory effect of word neighborhood size when reading with central field loss is modulated by word predictability and reading proficiency

Individuals with central field loss (CFL) from maculopathy experience severely impaired functional vision and major reading deficits. Given that reading speed strongly affects quality of life, improving reading performance is a key rehabilitation goal. The role of psycholinguistic factors in low-vision reading has been underexplored; prior work showed low-frequency words disrupt fixation patterns in CFL, suggesting lexical factors matter. In CFL, central scotomas and reliance on eccentric vision degrade letter identification through occlusion, blur, and crowding, increasing misidentifications (e.g., the word 'father' may be misread as similar neighbors like 'farmer', 'feather', or 'halter'). Consequently, CFL readers depend more on top-down linguistic inference. In normally sighted readers, orthographic neighborhood size (Coltheart's N) often facilitates recognition, though effects are task- and language-dependent. With degraded input in CFL, the authors hypothesize a reverse effect: larger N should hinder reading (Hypothesis 1). Beyond frequency and orthographic similarity, word predictability from context speeds recognition in normal reading. Predictability may mitigate confusion among neighbors; thus, the authors hypothesize that higher predictability reduces the inhibitory effect of neighborhood size (Hypothesis 2). They also examine modulation by reading proficiency (daily reading practice), expecting larger inhibitory effects among less proficient readers.

Prior literature shows that in normal vision, word frequency reliably speeds recognition and reading, while orthographic neighborhood size (Coltheart's N) typically facilitates performance in tasks like naming and lexical decision, though effects depend on task and the frequency of neighbors (higher-frequency neighbors can inhibit). Eye-movement studies in normal reading indicate that predictability from context increases skipping rates and shortens fixations. For low vision, earlier work found strong frequency effects on CFL reading times, sometimes larger than in normal vision, and inconsistent across synonym pairs, implying additional factors beyond frequency. The present study builds on these findings to test orthographic similarity (N) and predictability during sentence reading under CFL and explores how daily reading practice (proficiency) may modulate these effects.

Participants: N=19 patients (13 female), ages 32–89 (mean 75±15), with bilateral central scotoma and monocular acuity ≥0.4 logMAR (worse than or equal to 20/50) in the better eye. Exclusion: other ophthalmologic diseases (e.g., glaucoma), cognitive/reading disorders pre-impairment. Data collected: demographics, etiology, lens status, disease onset, field loss type, rehabilitation history, daily reading time, profession, and pre-impairment heavy reading status. Mean best-corrected acuity: 0.81±0.28 logMAR. Ethics approvals obtained; informed consent provided. Apparatus: LCD HP LE2201W, 60 Hz; PsychoPy stimuli; viewing distance 40 cm; sentences in Courier font, black on white; individualized print size at each participant's critical print size (MNREAD). Reading monocularly (better eye) with near correction. Stimuli and reading material: 32 pairs of French synonyms selected to control length (3–8 chars; equal within pair), neighborhood size difference (5–10), and variable frequency ratios. For each pair, two matching sentences were authored so either synonym fits the same sense; target positions were n-1, n-2, or n-3 to avoid wrap-up effects at sentence end. Two experimental conditions counterbalanced sentence-word pairing, yielding 128 sentences total (64 per condition). This design also provided two predictability measures per target word. Procedure: Self-paced reading with non-cumulative masking: sentences appeared fully masked with 'x'; participants revealed one word at a time via keyboard (forward/backward unmasking allowed), reading aloud for speed and accuracy. A trial ended when participants indicated completion. Training with proverbs preceded testing. Participants completed 2–4 blocks of 16 trials (depending on fatigue/speed). Measures: target-word accuracy (correct/incorrect) and total reading time (sum over multiple unmaskings if any). Predictability: For each target, computed trigram occurrence (percentage of occurrence of the three-word sequence ending with the target) using Google Books Ngram 'French 2012' corpus (792,118 books, 1800–2008). This served as the predictability metric. Reading proficiency: Self-reported daily reading time was dichotomized due to skew (many zeros): Daily reading (yes) vs Daily reading (no). Former heavy reader (yes/no) captured pre-impairment literacy level. Statistical analysis: Mixed-effects modeling in R. Accuracy analyzed via GLME (binomial); reading time via LME on log-transformed times. Fixed effects included neighborhood size (N), word frequency (ln), word length, trigram occurrence (ordered quantile normalized), and reading proficiency; interactions explored among N, predictability, and proficiency. Random intercepts for participant and item (target word). Transformations: reading time and word frequency in natural log; trigram occurrence via ordered quantile normalization; context amount square-root (if used); frequency and length mean-centered. Model selection by AIC and likelihood-ratio tests; significance defined by |z| or |t| > 2 (approx. 5%). 95% CIs reported.

- Accuracy (Analysis 1): Mean target-word accuracy was 94% (range 62–100%). In a GLME including N, frequency (ln), length, and trigram occurrence, none significantly affected accuracy. Daily reading (no) slightly reduced accuracy compared to daily reading (yes): estimate -1.126 on logit scale (z=-2.064, p=0.039), corresponding to 99.1% vs 97.3% accuracy (≈1.8% absolute difference). No effect of neighborhood size on accuracy. - Reading time: Simple LME (Analysis 2, no interactions) showed a modest but significant inhibitory effect of N on log RT: estimate 0.013 (t=2.507, p=0.013). This corresponds to ≈8% longer time from N=0 to N=6 and ≈14% longer from N=0 to N=10. Predictability reduced reading time (trigram occurrence estimate -0.115; t=-4.129, p<0.001). Frequency reduced time (estimate -0.047; t=-3.419, p<0.001). Word length and daily reading were not significant. - Interaction model (Analysis 3): A 3-way interaction revealed that the inhibitory effect of N depended on predictability and reading proficiency. For daily readers, N had no significant effect across predictability levels. For non-daily readers, N strongly increased reading time at low predictability: slope 0.07 per neighbor (t=5.22, p<0.001), implying ≈+52% time from N=0 to N=6 and ≈+101% from N=0 to N=10. As predictability increased, the N effect weakened significantly in non-daily readers (N:trigram occurrence estimate -0.017; t=-2.03, p=0.043): at average predictability, ≈+65% from N=0 to 10; at high predictability, ≈+35% from N=0 to 10. Frequency remained a robust facilitator (estimate -0.046; t=-3.39, p<0.001); word length not significant. No significant interactions for frequency with proficiency or N.

Findings support the primary hypothesis that, under CFL, orthographic neighborhood size has an inhibitory effect on word processing time during sentence reading, contrasting with the facilitative or mixed effects reported in normal vision. Degraded and incomplete visual input increases confusability among orthographic neighbors, requiring additional processing to reconcile visual evidence with sentence meaning, which prolongs reading time without substantially affecting accuracy. Critically, the inhibitory effect of neighborhood size is modulated by contextual predictability and by reading proficiency: greater predictability mitigates neighbor-induced uncertainty, but this benefit is most evident among less proficient readers who have ceased daily reading. In proficient (daily) readers, practice appears to offset ambiguity costs, yielding negligible neighborhood effects across predictability levels. These results underscore the interplay between bottom-up visual constraints and top-down linguistic inference in CFL, and they highlight the importance of maintaining reading practice. Clinically, they point to text simplification—favoring lower-neighborhood, higher-frequency synonyms—as a promising strategy to reduce processing costs and promote more fluent reading in severely impaired, less proficient CFL readers.

This study demonstrates that, in readers with central field loss, larger orthographic neighborhood size slows word processing during sentence reading, with the strongest inhibition occurring when contextual predictability is low and among individuals who no longer read daily. Predictability attenuates this cost, and regular reading practice appears to neutralize it. These findings extend psycholinguistic models to degraded vision and suggest practical interventions: lexical text simplification (choosing lower-neighborhood, higher-frequency synonyms) to enhance accessibility and support rehabilitation. Future research should broaden participant demographics, incorporate precise retinal function measures (e.g., microperimetry), explicitly control neighborhood frequency and alternative neighbor definitions (including transpositions, additions, deletions), and evaluate these effects in paradigms closer to natural binocular reading with permissible word skipping.

- Sample skewed toward older adults (16 of 19 participants aged 70–89); broader age and disease-onset ranges are needed. - Lack of detailed retinal characterization (e.g., microperimetry, scotoma size/shape, fixation locus) limits generalizability. - Neighborhood frequency not fully controlled in stimulus design; future work should balance neighbor frequencies. - Orthographic neighbor definition was position-specific and length-dependent (Coltheart’s N); CFL reading may require extended definitions including transpositions, additions, and deletions. - Self-paced reading paradigm disallowed word skipping and testing was monocular, which may not fully reflect natural reading conditions.