Value expressed with alphabetic characters

[kermitt2]

For example "twenty kilos" - currently the recognition is very bad and there is no normalization into numerical values.

We should:

• add a matching feature for this very limited vocabulary in the quantity model (it will generalize the examples in the training data),
• add a dedicated normalization.

[kermitt2]

Basic words to number normalization - only English - with commit 18379a25a49b17bf61e4d9bf0cdcd3504ad90cb3

[kermitt2]

Add a number word matching feature in the quantity model with commit 5e0b1b72be33e0a25bdff769ec1d5f1f182e6cdc -> the detection works much better now

[lfoppiano]

FYI http://stackoverflow.com/questions/3911966/how-to-convert-number-to-words-in-java

[kermitt2]

We're doing the opposite from this stackoverflow, we convert words to numbers ;)

But anyway, it's almost finished, just need to cover a couple of unregular expressions (like "oh", "dozen", "half",...) as CRF feature and normalization.