Abstract Philosophy of language and computer science, despite being very distinct fields, share a great interest in natural language. However, while philosophy has traditionally opted for a formalist approach, computer science has been increasingly favoring probabilistic models. After presenting these two approaches in more detail, we discuss some of their main virtues and limitations. On the one hand, formalist models have trouble in acquiring semantic information from corpora and learning from large amounts of data. Probabilistic approaches, on the other hand, have difficulty in operating with compositionality, in dealing with contrast sets and hierarchical relations, and in distinguishing normative and descriptive views of meaning. We argue that a more fruitful dialogue between philosophers and computer scientists may help to produce a better approach to natural language and stimulate the integration of logical and probabilistic methods.