d-ataman / lmm Goto Github PK

Hi there,

I recently started going through the code in this repository after having read your paper, which I found very fascinating.

I would be very interested in trying to reproduce the results of the paper using the code from the repository. With that in mind, I had the following questions:

0. Where can I download a version of the TED corpus used in this paper? What data format do I need the TED corpus to be in? Is the default XML fine?
1. What version of the IWSLT evaluation data does the paper use? For instance, IWSLT 2019 seems to have used English -> Czech but I'm not sure where to obtain the data for other language pairs. (found the answer to this on p.13 of the paper)

2. Is there a need to incorporate your other repo, Char-NMT in order to make this work? I noticed it is mentioned in the shell scripts under ./examples.
3. I noticed that the onmt folder is very similar to OpenNMT-py version 0.1. Do I need to separately install OpenNMT-py to use this repository or is this a fork + extension of OpenNMT-py? If I do, is v1.0 compatible?