In some domains, certain expressions (phrases, compound words) are very common and meaningful. Having the simple tokenizer recognize such words would be very useful, and it could be done simply by passing all tokens through a transformation that recognizes those expressions and replaces sequences of tokens with said expressions. Reference:
In some domains, certain expressions (phrases, compound words) are very common and meaningful. Having the simple tokenizer recognize such words would be very useful, and it could be done simply by passing all tokens through a transformation that recognizes those expressions and replaces sequences of tokens with said expressions. Reference:
http://www.mimno.org/articles/phrases/
That would improve the performance of models using
tokenizers.simple, especially in certain domains.