Improvements to the normalization module

[yosi-dediashvili]

We need to add another normalization method that concatenates words from the title. The new method will come right after the 2nd normalization, and for will:

• For each pair of words (as a sliding windows) splitted by underscore _, concatenate them, and define it as normalized string
  • For each normalized string, we'll apply the 3rd and 4th normalization method

This step will actually return a list of normalized strings and not a single ont.

[yosi-dediashvili]

Another normalization method: Replace any number with its string representation:

• 1 becomes one
• 10 becomes ten
• 20 becomes twenty

We should limit ourselves to normalization of up to the number 20. Past that, the numbers are converted to two words, so it get too complicated.