When most commercial games are shipped, their gameplay has typically been prescripted. All game components are created during development, mostly as predetermined rigid artifacts with which a player will interact. This can lead to predictable and impersonal gameplay, while alienating unconventional players. Adaptivity in games has been recently proposed to overcome these shortcomings. Although adaptive games are a possible solution for this problem, in practice they are rare. The technology and development techniques to support such games are strongly ad-hoc and not easily accessible and controllable by game designers. Procedural content generation and semantic modeling can powerfully combine to tackle these issues. This thesis proposes the use of gameplay semantics, i.e. the knowledge on the gameplay meaning and value of a game world and its objects, to adaptively generate game worlds. Using gameplay semantics, we devised a generation framework aimed at creating personalized content for complex and immersive game worlds. This framework captures which content provided a given personal gameplay experience. This model is then used to generate content for the next predicted experience, through retrieval and recombination of semantic gameplay descriptions, i.e. case-specific mappings between content and gameplay semantics. This framework can be used to link the procedural generation of game worlds with gameplay, as measured by player modeling techniques. Gameplay semantics is created in a generic way and can be effectively used to steer the procedural generation of player-matching game worlds, both at design and at game stage. Gameplay semantics can steer the adaptive generation of game worlds by capturing the key features required for adaptivity. Both game world objects and properties can be synthesized in response to experience-driven features, e.g. the personalized difficulty of a game. Our player studies showed that this type of generation is successful in keeping such experience features in balance with different player types. Gameplay semantics can be used to actively include game designers in the creation loop, by allowing them to author adaptivity in a more expressive and specific fashion. Designers can: (i) interactively create gameplay semantics that describe players and content, and (ii) match one with the other. These matching rules are the underlying semantic gameplay descriptions that support our generation framework. User evaluation shows that gameplay semantics can provide game designers with a rich expressive range to convey specific adaptive gameplay experiences to its players. From the combined results of these contributions we can conclude that gameplay semantics is an effective, generic and controllable method to improve adaptation in games. We therefore hope that this work will encourage and facilitate the development of more and better adaptive games.