OpenNMT Forum

Compared patterns

Some organizations realize the value of ruled based translation while acknowledging the great possibilities offered by Neural MT and they’d like the NMT engine to provide them with the compared patterns generated by the millions of translated sentences they have inserted in their databases.

Is this possible? Can this be done?

Which compared patterns are generated by the sentences?
Have a look at the illustrated transfomer to get a hint how the common NMT-architecture works.
Checkout modernNMT to reuse tmx-files.

I am talking about patterns indicated below the sentence usually.

Example of a compared pattern:

Ex: The white dog
ART COl-ADJ NS

=

Le chien blanc
RT NMS ADJ-COUL

Probably you mean the part of speech tagging executed on the sentence. They can be used in NMT as additional input feature. In many cases the parser limitation prevails the small quality improvements in NMT. Other approaches based on the grammar have the same problem with this limitation.

Tx Bacnsteize for your excellent info. Where do we find this transformer? .

http://letmegooglethat.com/?q=transformer+nmt or read the manual

Bachteike, please stop giving me links to Google searches.

I am not 5 years old and I know the Google concept very well. I need to have info to understand the NMT concept and not links to all sorts of places. Links I can find myself thank you.

Guillaume, what is needed is examples. Thanks! Theory without examples isn’t worth much.