CJK language packs

[Tommaso Teofili <to...@gmail.com>]

Hi all,

since CJK LPs are missing, I was thinking to pursue a PoC leveraging a
different approach for providing LPs.
My plan is to use more recent methods that build on pretrained embeddings
which are still simple enough for us not to require to incorporate a deep
learning library (e.g. we may just use commons-math3 for working with
matrices and vectors).
A few examples might be found at [1].

WDYT ?
Regards,
Tommaso

[1] : http://ruder.io/cross-lingual-embeddings/index.html

Re: CJK language packs

[Tommaso Teofili <to...@gmail.com>]

thanks Michael, having another pair of eyeballs to look what I'll be doing
wrong is already quite a lot ;)
I plan to work in parallel on a LP trained using out of the box Joshua
pipeline and another one based on one of the embeddings approach outlined
in the previous link I've shared.

Regards,
Tommaso

On Sat, 28 Sep 2019 at 20:53, Michael Wall <mj...@apache.org> wrote:

> I am interested but I don't how much help I will be.
>
> On Wed, Sep 25, 2019 at 3:04 AM Tommaso Teofili <tommaso.teofili@gmail.com
> >
> wrote:
>
> > if anyone is interested in helping out, I will be starting working on the
> > mentioned stuff in the next few weeks.
> > Feel free to reach out, share ideas, etc.
> >
> > Regards,
> > Tommaso
> >
> > On Wed, 17 Jul 2019 at 10:59, Tommaso Teofili <tommaso.teofili@gmail.com
> >
> > wrote:
> >
> > > Hi all,
> > >
> > > since CJK LPs are missing, I was thinking to pursue a PoC leveraging a
> > > different approach for providing LPs.
> > > My plan is to use more recent methods that build on pretrained
> embeddings
> > > which are still simple enough for us not to require to incorporate a
> deep
> > > learning library (e.g. we may just use commons-math3 for working with
> > > matrices and vectors).
> > > A few examples might be found at [1].
> > >
> > > WDYT ?
> > > Regards,
> > > Tommaso
> > >
> > > [1] : http://ruder.io/cross-lingual-embeddings/index.html
> > >
> >
>

Re: CJK language packs

[Michael Wall <mj...@apache.org>]

I am interested but I don't how much help I will be.

On Wed, Sep 25, 2019 at 3:04 AM Tommaso Teofili <to...@gmail.com>
wrote:

> if anyone is interested in helping out, I will be starting working on the
> mentioned stuff in the next few weeks.
> Feel free to reach out, share ideas, etc.
>
> Regards,
> Tommaso
>
> On Wed, 17 Jul 2019 at 10:59, Tommaso Teofili <to...@gmail.com>
> wrote:
>
> > Hi all,
> >
> > since CJK LPs are missing, I was thinking to pursue a PoC leveraging a
> > different approach for providing LPs.
> > My plan is to use more recent methods that build on pretrained embeddings
> > which are still simple enough for us not to require to incorporate a deep
> > learning library (e.g. we may just use commons-math3 for working with
> > matrices and vectors).
> > A few examples might be found at [1].
> >
> > WDYT ?
> > Regards,
> > Tommaso
> >
> > [1] : http://ruder.io/cross-lingual-embeddings/index.html
> >
>

Re: CJK language packs

[Tommaso Teofili <to...@gmail.com>]

if anyone is interested in helping out, I will be starting working on the
mentioned stuff in the next few weeks.
Feel free to reach out, share ideas, etc.

Regards,
Tommaso

On Wed, 17 Jul 2019 at 10:59, Tommaso Teofili <to...@gmail.com>
wrote:

> Hi all,
>
> since CJK LPs are missing, I was thinking to pursue a PoC leveraging a
> different approach for providing LPs.
> My plan is to use more recent methods that build on pretrained embeddings
> which are still simple enough for us not to require to incorporate a deep
> learning library (e.g. we may just use commons-math3 for working with
> matrices and vectors).
> A few examples might be found at [1].
>
> WDYT ?
> Regards,
> Tommaso
>
> [1] : http://ruder.io/cross-lingual-embeddings/index.html
>