As speech technology continues to evolve, so do the heated debates among speech system developers over the best way to design natural language dialog systems. On one hand, there are speech recognition designers who champion strictly grammar-based speech recognition systems that find the best possible word match for the string of phonemes presented to the recognizer. On the other hand, there are those who reject context-free grammars altogether, claiming such methods of processing natural language input fail to reflect how humans comprehend speech.

Among those who reject the context-free grammar (CFG) approach to designing speech recognizers, some have actively championed a conceptually-based, Artificial Intelligence (AI)-driven speech recognizer that relies heavily on relevant knowledge domains when processing speech input. In such a system, recognition of the user’s speech input is more dependent on conceptual content derived from relevant knowledge domains (e.g., flight arrival and departure times) than on matching each of the user’s phonemes – regardless of the context – against the speech system’s application vocabulary.

Against this backdrop of contrasting approaches to Natural Language Processing, a new method that is neither entirely based on context-free-grammar, on the one hand, nor domain-driven, on the other hand, has emerged as an alternative approach to understanding natural language dialog. This new method is called Sequence Package Analysis, or SPA. [1]