| Statistical language modelling |
|
|
|
|
|
|
|
|
|
|
|
|
|
| • | Goal: Model the
distribution of the next word in |
|||||||||||
| a sentence. |
||||||||||||
| • | N-grams are the
most widely used statistical |
|||||||||||
| language models. |
||||||||||||
| – | They
are simply conditional probability tables |
|||||||||||
| estimated
by counting n-tuples of words. |
||||||||||||
| – | Curse
of dimensionality: lots of data is needed if n |
|||||||||||
| is
large. |
||||||||||||