Hi there - I'm working with complex documents feature densely tagged sentences (sometimes up to 4-40+ tags) and was wondering if the tags-projection method is 1. Language specific or 2. whether it can be tuned/trained somehow. Is there any information about this feature before I dive in and study the code? Many thanks!
nonsequitir
All 3 comments
Hi @nonsequitir
the tag projection feature is a 2-steps process:
- First the system computes the alignment between the source and the target sentence with FastAlign model.
- Then the tags are projected with some hard-coded heuristics using the computed alignment.
The first step uses the aligner model, so this step is a classic machine-learning task. The second part is hard-coded into this class: XMLTagProjector.java
davidecaroselli
on 20 Jun 2018
👍3
Awsome, I see that now. Thanks @davidecaroselli !
nonsequitir
on 27 Jun 2018
I'm closing this issue for now.
If you have any other question or something you want to discuss, please don't hesitate to re-open it!
davidecaroselli
on 25 Jul 2018
👍1
Was this page helpful?
0 / 5 - 0 ratings
Related issues
mzeidhassan
路
3Comments
LoekvanKooten
路
3Comments
mzeidhassan
路
5Comments
EtienneAb3d
路
4Comments
AndRossi
路
7Comments
Most helpful comment
Hi @nonsequitir
the tag projection feature is a 2-steps process:
The first step uses the aligner model, so this step is a classic machine-learning task. The second part is hard-coded into this class: XMLTagProjector.java