Orthographic Neighbourhood Size Effects in Chinese Character Recognition: Small, Inconsistent, and Theoretically Ambiguous

Sub-character units. In the literature the term radical is used flexibly to refer to various parts of a character: the indexing component, the residual component, a general stroke pattern, among others; we adopt the first definition. The term 部件 is translated as ‘Chinese character component’ in the GF 0014-2009 standard (National Language Commission, 2009). We instead use stroke pattern to differentiate it from the broader use of component elsewhere in the paper, and to emphasise its close link to strokes. Stroke patterns vary in stroke complexity, ranging from single-stroke components (e.g., 一 in character 丛) to compound components composed of multiple stroke patterns (e.g., 相 in character 想).

Definition of the independent variables used in Study 1. Consistency and regularity were included to control for potential phonological influence. Nstrokes and character frequency were included as they are strong predictors of word recognition performance.

The criterion for removal at each step of data cleaning. Incorrect responses, IQR outliers, and RT outliers were removed at the trial level; Mean accuracy was assessed at the character level. The number of trials and characters removed is provided.

Descriptive statistics of the continuous variables used in Study 1. For the variable regularity, there are 1,410 regular characters and 1,350 irregular characters.

Spearman’s rank correlations between response times and independent variables for characters (N = 2767) used in the replication of neighbourhood size effects based on semantic components and phonetic components.

Fixed effects results from the LMM in Study 1 (fitted to 2,767 characters and 69,370 trials). The variance inflation factors, marginal R², and 95% confidence intervals with lower and upper limits are provided.

Comparison of absolute and normalised edit distance measures. The steps of WLD calculation for the English word examples are illustrated in Figure 5. Chinese characters: 鬰 yu4 ‘lush and growing abundantly’, 礬 fan2 ‘alum’, and 匕 bi3 ‘spoon’. The IDS of 鬰 is ⿳⿲木⿱㐅⿻丿乀 木冖⿰⿱⿶凵⿻ 㐅⿳丶⿰丶 丶丶⿺乚丿彡, of 礬 is ⿱⿱⿲ 木⿱㐅⿻丿乀木⿻一人⿸⿱一 丿口, and of 匕 is ⿺乚丿.

Descriptive statistics for variables used in Study 2.

Correlations between variables used in Study 2 (Spearman’s r).

Results for fixed effects factors across the LMMs in Study 2. For model architectures, the three models only differed in the neighbourhood measure used (initial N = 4,384 characters, 111,518 observations). After removing residual outliers, the number of characters remained unchanged for all models and the final observation counts were similar across models (2a: 109,632; 2b: 109,631; 2c: 109,630).

Ten nearest characters for two example characters 憬 (left, lower WLD-10) and 器 (right, higher WLD-10). Both have an N-ch of 4, pointing to the first four characters listed in both columns that were formed by one substitution of one stroke pattern from each character. Character 憬 (NStrokes = 15, WND-10 = .103) occupies a denser orthographic neighbourhood beyond its N-ch neighbours, while character 器 (NStrokes = 16, WND-10 = .242) occupies a sparser neighbourhood. Our results show that characters like 憬 with denser neighbourhood are easier to recognise, after controlling for character frequency and number of strokes.

Loadings from principal component analysis for predictors used in Study 1.

Results of the fixed effects factors from the LMM with rotated components from PCA in Study 1 (N = 2,767). For each rotated component, the primary representative variable is shown. Variables with moderate contributions to the components are listed in descending order of absolute loading values with + showing positive loading and - negative loading.

Results for fixed effects factors across the LMMs combining variables from Study 1 and Study 2 (initial N = 2,767 characters, 70,534 trials). After identical outlier removal, final trial counts were 69,374 for the N-ch model, 69,370 for the WLD-10 model, and 69,372 for the WND-10 model. The unrounded estimate for N-ch is 0.0004 (SE = 0.0001).