1980s

• In the early 1980s some work on the automatic induction of lexical and syntactic information from corpora (Brown Corpus, Lancaster-Oslo-Bergen Corpus).
• In the sphere of speech recognition, research originating at IBM Yorktown resulted in statistical methods based on HMMs (hidden Markov models) that outperformed previous knowledge-based approaches.
• These methods use a probabilistic finite state machine to model the pronunciation of words and make use of a hill-climbing training algorithm to fit the model parameters to the actual speech data. Most existing commercial speech recognition systems are based on HMMs.
• Starting in the late 1980s the success of statistical methods in speech spread to other areas of language processing, notably POS tagging, spelling correction, and parsing.