I'm considering a new Moses branch that would allow users to create
proprietary binarized tables that only work in their deployments. The
purpose is to protect data models from use with unauthorized systems. The
project:

   1. after creating binarized models with the processPhraseTable and
   processLexicalTable steps, would encode the binarized data tables to
   a proprietary format
   2. would modify the Moses PhraseDictionaryTreeAdaptor
   (PhraseDictionaryTree & other as required) to a) detect open or proprietary
   format b) if open format, execute to existing code, if proprietary execute
   branched code to decode proprietary format for every transaction and lookup
   3. publish static mods to Moses as open source. Static mods could,
   however, could point to dynamic linked libraries that could be proprietary.


Questions:
A) Does a branch like this exist or has anyone considered it?
B) What problems, if any, would this create in the smart pointer's ability
to search and find what it's looking for?
C) Encoding the .xxxtree and .xxx data files would require decoding with
every lookup... potentially degrading performance. Instead, would it be
possible to encode/decode only the .xxxvoc and .idx files to achieve the
same level of protection without the performance hit of decoding the data
and tree for each lookup?

Thanks,
Tom


---------- Forwarded message ---------- From: [email protected] <
[email protected]>
Date: Fri, Feb 20, 2009 at 6:53 PM
Subject: Re: [Moses-support] zlib compression and binarized tables
To: Adam Lopez <[email protected]>



Excellent. Exactly what I was looking for.

Thank you.
Tom

On Fri, Feb 20, 2009 at 6:42 PM, Adam Lopez <[email protected]> wrote:
Yes, though not directly.  PhraseDictionaryTreeAdaptor forwards its calls to
PDAImp, which is a wrapper on PhraseDictionaryTree, which uses a PrefixTreeF
template class.  PrefixTreeF reads binary files using a smart pointer
defined in FilePtr.h.
Adam

On Fri, Feb 20, 2009 at 11:25 AM, [email protected] <[email protected]> wrote:
> Thanks Adam. >
> Following up, it looks like PhraseDictionaryTreeAdaptor.cpp is responsbile
> for reading the binarized phrase tables and language model. Is this
correct?
>
>
> Tom
_______________________________________________
Moses-support mailing list
[email protected]
http://mailman.mit.edu/mailman/listinfo/moses-support

Reply via email to