Fig 1.
Non-linear pattern of simulated data.
Left panel: original simulated data. Right panel: re-arranged simulated data.
Fig 2.
The results of three ‘linear’ classification algorithms applied to the simulated data, using 1-hot encoding for column and row membership.
Left panel: standard logistic regression, middle panel: logistic regression with lasso regularization, right panel: standard Rescorla-Wagner learning. Red pixels indicate predicted class A responses, blue pixels indicate true class A responses.
Table 1.
Example cues and outcomes for synonyms and homographs.
Fig 3.
Layout of a Rescorla-Wagner network.
Network layout (top) and weight matrix (bottom), obtained with the equilibrium equations of [68] for the Rescorla-Wagner equations, for the lexomes paid, pail, qaid, said, sail presented with frequencies 550, 50, 1, 9900, and 50, using letter bigrams as orthographic cues.
Table 2.
Semantic categories, difference of mean within group and outside-group similarity, p-value, and number of words in a category, using the similarity matrix derived from the L2L weight matrix estimated from the British National Corpus.
Fig 4.
Example probability density of afferrent weights.
Estimated probability density for the weights on the incoming links from orthographic trigrams to the lexome corner. Red lines present boundaries for the highest 50 absolute weights, blue lines indicate the boundaries for the 150 highest absolute weights.
Table 3.
Generalized additive model fitted to simplex words from the English Lexicon Project using classical lexical-distributional predictors.
Fig 5.
Regression spline smooths for classical predictors.
Smooths in the generalized additive model fitted to the visual lexical decision latencies (English Lexicon Project) using classic lexical-distributional predictors.
Table 4.
Generalized additive model fitted to simplex words from the English Lexicon Project using discrimination-based predictors.
Fig 6.
Regression spline smooths for discrimination-based predictors.
Bivariate smooths (upper panels) and univariate smooths (lower panels) in the generalized additive model fitted to the lexical decision data from the English Lexicon Project, using discrimination-based predictors.
Table 5.
Main correlations between classic and discriminative variables.
Table 6.
Means and standard deviations for item properties.
Table 7.
Loadings of predictors on the principal components.
Table 8.
Generalized additive mixed model fitted to the lexical decision latencies of experiment 1a–b using lexical-distributional predictors.
Fig 7.
Nonlinear effects in experiments 1a and 1b.
Left panel: tensor product smooth for the nonlinear interaction of the principal component for neighborhood similarity structure (pc1) and frequency of occurrence (pc2) in Experiment 1. Darker colors indicate shorter response latencies. Right panel: Partial effect of the by-participant random smooths for Trial. Each curve represents a different participant.
Table 9.
Generalized additive model for the response latencies in experiment 1, using discrimination-based predictors.
Fig 8.
Tensor product regression smooths for experiment 1, discrimination-based predictors.
Upper panels: the three-way interaction of G2L prior, G2L a-diversity, and Spelling. Lower panels: the two-way interaction of L2L prior by L2L I-diversity.
Table 10.
Generalized additive mixed model fitted to the lexical decision latencies of experiment 2 using classical lexical-distributional predictors.
Fig 9.
Tensor product regression smooths for experiments 2, classical lexical-distributional predictors.
Left panel: tensor product smooth for the nonlinear interaction of the principal component for neighborhood similarity structure (pc1) and frequency of occurrence (pc2). Darker colors indicate shorter response latencies. Right panel: tensor product smooth for the nonlinear interaction of individual differences in spelling proficiency and vocabulary.
Table 11.
Generalized additive model for the response latencies in experiment 2 using discrimination-based predictors.
Fig 10.
Tensor product regression smooth for the three-way interaction of G2L prior, G2L a-diversity and the spelling-vocabulary product score in Experiment 2.