Spoken Language Understanding

Spoken Language Understanding studies how systems interpret what people say, from identifying intents and filling slots to tracking dialogue state, answering questions, and taking actions. Approaches include ASR pipelines paired with language models, end-to-end speech and audio language models, and structured methods that combine transcript, acoustic, and conversational evidence. Work also uses fine-tuning, synthetic data, and multilingual benchmarks to address low-resource languages, accents, noise, and varied speaking styles. A central challenge is making models reason reliably from speech itself: transcription errors and difficult acoustic conditions can undermine semantic decisions even when text-based performance is strong.