25 Stochastic language models 2

Outline of development methodology:

Transcription

Normalisation of transcription (filtering)

Count of words, calculation of a priori probabilities

Efficient organisation of language model as tree structure

Classification in terms of perplexity

Test of speech recogniser with and without language model

Extensions:

Trigrams, ... (Problem: decreasing data)

Class n-grams

Morphological n-grams (e.g. stem-based)

Integration with multiple knowledge bases, including linguistic parsers