This paper presents a feature-space theory of the production effect in recognition memory. The production effect refers to the phenomenon where words read aloud are remembered better than words read silently. The authors propose that this effect arises from the dimensionality of feature spaces, where phonological features are drawn from a compact feature space, and deeper features are sparsely subselected from a larger subspace. They argue that production enhances memory by increasing the effective strength of encoded items through the addition of features, which also increases the distinctiveness of encoded items. This mechanism explains both the strength theory and the distinctiveness heuristic, and has been able to explain a wide range of empirical findings related to the production effect. The authors extend attentional subsetting theory, which posits that when studying an item, one attends to a small subset of all known features of an item. They propose that production leads to more phonological features being attended and encoded, which enhances memory. This is in contrast to other manipulations of strength, such as spaced repetition and study time, which are thought to influence strength in a different way. The authors argue that the differences between production and other strength manipulations can be explained by the characteristics of the feature subspaces they operate upon. The authors also propose that the production effect can be influenced by semantic aspects of words without needing to assume production acts directly on semantic features. They show that the production effect is not limited to vocalization but can also occur with typing, writing, and even whispering, albeit with smaller effects. The authors also show that the production effect can be influenced by the number of features stored, the size and properties of the production feature-space, and how this relates to other relevant feature subspaces. The authors develop a mathematical model of the production effect based on the matched-filter model, which assumes that memory is a vector that is a sum of the vectors representing list-items. They show that the production effect can be explained by the addition of features, which increases the effective strength of encoded items and also increases the distinctiveness of encoded items. The model also accounts for the list-strength effect, where the advantage of strong items is greater in mixed lists than in pure lists. The authors also show that the production effect can be influenced by the number of features stored, the size and properties of the production feature-space, and how this relates to other relevant feature subspaces. The authors also show that the production effect can be influenced by the number of features stored, the size and properties of the production feature-space, and how this relates to other relevant feature subspaces. They argue that the production effect is not limited to vocalization but can also occur with typing, writing, and even whispering, albeit with smaller effects. The authors also show that the production effect can be influenced by the number of features stored, the size and properties of the production feature-space, and how this relates to other relevant feature subspaces.This paper presents a feature-space theory of the production effect in recognition memory. The production effect refers to the phenomenon where words read aloud are remembered better than words read silently. The authors propose that this effect arises from the dimensionality of feature spaces, where phonological features are drawn from a compact feature space, and deeper features are sparsely subselected from a larger subspace. They argue that production enhances memory by increasing the effective strength of encoded items through the addition of features, which also increases the distinctiveness of encoded items. This mechanism explains both the strength theory and the distinctiveness heuristic, and has been able to explain a wide range of empirical findings related to the production effect. The authors extend attentional subsetting theory, which posits that when studying an item, one attends to a small subset of all known features of an item. They propose that production leads to more phonological features being attended and encoded, which enhances memory. This is in contrast to other manipulations of strength, such as spaced repetition and study time, which are thought to influence strength in a different way. The authors argue that the differences between production and other strength manipulations can be explained by the characteristics of the feature subspaces they operate upon. The authors also propose that the production effect can be influenced by semantic aspects of words without needing to assume production acts directly on semantic features. They show that the production effect is not limited to vocalization but can also occur with typing, writing, and even whispering, albeit with smaller effects. The authors also show that the production effect can be influenced by the number of features stored, the size and properties of the production feature-space, and how this relates to other relevant feature subspaces. The authors develop a mathematical model of the production effect based on the matched-filter model, which assumes that memory is a vector that is a sum of the vectors representing list-items. They show that the production effect can be explained by the addition of features, which increases the effective strength of encoded items and also increases the distinctiveness of encoded items. The model also accounts for the list-strength effect, where the advantage of strong items is greater in mixed lists than in pure lists. The authors also show that the production effect can be influenced by the number of features stored, the size and properties of the production feature-space, and how this relates to other relevant feature subspaces. The authors also show that the production effect can be influenced by the number of features stored, the size and properties of the production feature-space, and how this relates to other relevant feature subspaces. They argue that the production effect is not limited to vocalization but can also occur with typing, writing, and even whispering, albeit with smaller effects. The authors also show that the production effect can be influenced by the number of features stored, the size and properties of the production feature-space, and how this relates to other relevant feature subspaces.