How can I do Train And Test step in Giza++?

Question

In artificial intelligence methods we have two stages of training. These stages are data and testing.

In the training stage we give a huge amount of data to a system and we normally test it with smaller volume of data. Then we evaluate the output.

Now the question is can this training be done through the built in functionality embedded in GIZA++ or we should write a separate application for that?

If we should write a separate application can anybody help me by suggesting an already written application? Or a manual? Note: I want to have an alignment program not a statistical machine translation

I would prefer to train in Giza++ so I can test with unobserved data.

Thanks in advance.

Answer 1

This should be a good starting point to train a baseline mt system using Moses.
Normally GIZA++ is used for word-aligning your parallel corpus.
You'll need some other tools such as IRSTLM, SRILM, or KenLM for language model estimation.