``When it comes to natural language, you can't do anything until you've done everything.''
L.R. Harris [8]The field of natural language processing (NLP) has for a long time been kept separate from that of speech recognition (SR), the differing requirements of the two research areas sending them down different paths. NLP required large storage requirements and generally took long periods of time to produce results, whilst SR was aimed at getting results as quickly as possible. This made them incompatible as far as most research projects were concerned.
However, with the increase in computing power and the knowledge that a degree of NLP, such as a knowledge of the language of acceptable phrases, can greatly aid speech recognition, the fields are combining more and more.
Despite his quote shown above, Harris argues that natural language can be used successfully, because much ambiguity in language can be resolved when looking at the specific domain the application resides in.
There is also the matter of understanding the recognised text the speech engine has produced. NLP and natural language understanding (NLU) are both names for this incredibly hard task, and the following sections outline the main areas in a natural language system, and methods with which to compute them.