A text recommendation system helps language learners find suitable reading materials. Similar to graded readers, most systems assign difficulty levels or school grades to the documents in their database, and then identify the documents that best match the language proficiency of the learner. This graded approach has two main limitations. First, the common grade scale assumes that all learners find the same words difficult, ignoring different orders in vocabulary acquisition. Second, it cannot cater to the incremental vocabulary expansion typical of language learners, since discrete grades do not facilitate fine-grained adaptation in text recommendation. This article evaluates a personalized and adaptive method for text recommendation that aims to mitigate these limitations. This method recommends documents according to the new-word density preferred by the user. It estimates a document's new-word density with an open learner model that predicts the user's vocabulary knowledge. Users can dynamically update the learner model as they acquire new vocabulary. We conducted a user study involving ten learners of Chinese as a foreign language. Results shows that this method can estimate the learner model at 77.6% accuracy, outperforming the graded approach. Further, a user simulation demonstrates that it can recommend documents within 0.22% of the target new-word density, satisfying user preferences more accurately than the graded approach.