The temporal dynamics of intonation planning during speech production remain poorly understood. Serial models of speech production propose that intonation is either integrated after word-form encoding or planned earlier via an independent prosodic template. Recent parallel activation accounts suggest that intonation may become available simultaneously with other linguistic information. We used EEG and multivariate pattern analysis (MVPA) to examine when intonation becomes neurally detectable during speech planning. Thirty native Dutch speakers performed a picture-naming task. The stimuli were disyllabic nouns varying orthogonally in semantic category (animate vs. inanimate), manner of articulation of the first phoneme (fricative vs. plosive), lexical stress (word-initial vs. word-final), and intonation (question vs. exclamation). Time-resolved decoding identified when each linguistic dimension became decodable from EEG. The results support serial accounts of speech production, with a progression from conceptual to phonetic encoding. Lexical stress had a weak independent representation, yet modulated the temporal profile of intonation decoding: above-chance decoding of intonation planning was earlier for words with word-final stress compared to word-initial stress. This interaction suggests a prosodic template in which intonation and stress are integrated and evolve together, and that is decodable from neural patterns, as demonstrated here for the first time.