Procedural content generation (PCG)---the algorithmic creation of game levels, terrain, quests, and rules---has evolved from a memory-saving trick into one of game development's most active research areas. This article presents a narrative review of the field's canonical line: Perlin's 1985 image synthesizer, the search-based taxonomy of Togelius and colleagues, the ACM survey of Hendrikx and colleagues, the Springer volume of Shaker, Togelius, and Nelson, the AI-and-games synthesis of Yannakakis and Togelius, the machine-learning turn of Summerville and colleagues, and the reinforcement-learning frontier of Khalifa and colleagues. The synthesis is organized around three themes: foundations, in which noise functions, grammars, and search established the generative toolbox; design, in which PCG met authorship---level design as search space, evolution as game designer, mixed-initiative tools; and learning, in which generative models trained on human-authored corpora opened PCGML and its controllability problem. It is concluded that PCG's history is the progressive relocation of authorship---from the asset to the generator---and that controllability is the field's central open problem.
Zen Revista, 10 GAME· Zenodo (CERN European Organi...· 0 citations
Procedural content generation (PCG)---the algorithmic creation of game levels, terrain, quests, and rules---has evolved from a memory-saving trick into one of game development's most active research areas. This article presents a narrative review of the field's canonical line: Perlin's 1985 image synthesizer, the search-based taxonomy of Togelius and colleagues, the ACM survey of Hendrikx and colleagues, the Springer volume of Shaker, Togelius, and Nelson, the AI-and-games synthesis of Yannakakis and Togelius, the machine-learning turn of Summerville and colleagues, and the reinforcement-learning frontier of Khalifa and colleagues. The synthesis is organized around three themes: foundations, in which noise functions, grammars, and search established the generative toolbox; design, in which PCG met authorship---level design as search space, evolution as game designer, mixed-initiative tools; and learning, in which generative models trained on human-authored corpora opened PCGML and its controllability problem. It is concluded that PCG's history is the progressive relocation of authorship---from the asset to the generator---and that controllability is the field's central open problem.
Zen Revista, 10 GAME· Zenodo (CERN European Organi...· 0 citations
Dialogue systems---the machinery that lets games and agents converse---moved from Weizenbaum's 1966 ELIZA through the narrative engines of Si's Thespian, Mateas and Stern's Facade, and Cavazza's character-based storytelling, and the virtual humans of Swartout and Traum, to Rieser and Lemon's data-driven methodology, Serban's hierarchical networks, Vinyals and Le's neural conversational model, Zhang's personalized agents, Luger and Sellen's expectation gap, and Kiela's dodecathlon. This article presents a narrative review of that arc's canonical line: Weizenbaum's 1966 ELIZA, Cavazza, Charles, and Mead's 2002 storytelling, Si, Marsella, and Pynadath's 2005 Thespian, Mateas and Stern's 2005 Facade, Traum and colleagues's 2003 negotiation, Swartout and colleagues's 2006 virtual humans, Rieser and Lemon's 2011 methodology, Vinyals and Le's 2015 neural conversational model, Serban and colleagues's 2016 hierarchical networks, Luger and Sellen's 2016 expectation gap, Zhang and colleagues's 2018 personalization, and Kiela and colleagues's 2018 dodecathlon. The review is organized around three themes: the scripted origins, in which the pattern-matching's illusion, the drama managers's, and the virtual humans's architectures built the game's conversation; the statistical turn, in which the data-driven's methodology and the neural's hierarchies moved the dialogue from the rules to the corpora; and personalization and the language-model era, in which the personas's, the expectations's gap, and the benchmarks's carried the talk into the large models's future. It is concluded that dialogue systems are the game's social machinery---and that their arc is the conversation's engineering from the illusion's ELIZA to the persona's large models.
Zen Revista, 10 GAME· Zenodo (CERN European Organi...· 0 citations
Dialogue systems---the machinery that lets games and agents converse---moved from Weizenbaum's 1966 ELIZA through the narrative engines of Si's Thespian, Mateas and Stern's Facade, and Cavazza's character-based storytelling, and the virtual humans of Swartout and Traum, to Rieser and Lemon's data-driven methodology, Serban's hierarchical networks, Vinyals and Le's neural conversational model, Zhang's personalized agents, Luger and Sellen's expectation gap, and Kiela's dodecathlon. This article presents a narrative review of that arc's canonical line: Weizenbaum's 1966 ELIZA, Cavazza, Charles, and Mead's 2002 storytelling, Si, Marsella, and Pynadath's 2005 Thespian, Mateas and Stern's 2005 Facade, Traum and colleagues's 2003 negotiation, Swartout and colleagues's 2006 virtual humans, Rieser and Lemon's 2011 methodology, Vinyals and Le's 2015 neural conversational model, Serban and colleagues's 2016 hierarchical networks, Luger and Sellen's 2016 expectation gap, Zhang and colleagues's 2018 personalization, and Kiela and colleagues's 2018 dodecathlon. The review is organized around three themes: the scripted origins, in which the pattern-matching's illusion, the drama managers's, and the virtual humans's architectures built the game's conversation; the statistical turn, in which the data-driven's methodology and the neural's hierarchies moved the dialogue from the rules to the corpora; and personalization and the language-model era, in which the personas's, the expectations's gap, and the benchmarks's carried the talk into the large models's future. It is concluded that dialogue systems are the game's social machinery---and that their arc is the conversation's engineering from the illusion's ELIZA to the persona's large models.
Zen Revista, 10 GAME· Zenodo (CERN European Organi...· 0 citations