Moses-support Digest, Vol 87, Issue 61

Send Moses-support mailing list submissions to
moses-support@mit.edu

To subscribe or unsubscribe via the World Wide Web, visit
http://mailman.mit.edu/mailman/listinfo/moses-support
or, via email, send a message with subject or body 'help' to
moses-support-request@mit.edu

You can reach the person managing the list at
moses-support-owner@mit.edu

When replying, please edit your Subject line so it is more specific
than "Re: Contents of Moses-support digest..."


Today's Topics:

1. translation probabilities in a string-to-tree system
(Marion Weller)
2. Re: Difference between lexical reordering and distortion?
(Philipp Koehn)
3. Re: translation probabilities in a string-to-tree system
(Rico Sennrich)


----------------------------------------------------------------------

Message: 1
Date: Mon, 27 Jan 2014 13:54:20 +0100
From: Marion Weller <wellermn@ims.uni-stuttgart.de>
Subject: [Moses-support] translation probabilities in a string-to-tree
system
To: moses-support@mit.edu
Message-ID:
<CAHfP6mCHgBNDMonG_iRSKCu_he-SZc2eJRbykvdQVszQ3dRSbg@mail.gmail.com>
Content-Type: text/plain; charset="iso-8859-1"

Hello,

I have a question about the p(e|f) probabilities in a string-to-tree system:

In the example below, the source-side strings "according to your [X][nn]"
and
"according to your [X][cnp]" are counted as different source-side phrases.
(The probabilities for the source-side string "according to your [X][nn]"
sum up to
1, and the source-side string "according to your [X][cnp]" has a
translation probability
of 1)

according to your [X][cnp] [X] ||| entsprechend ihrem [X][cnp]
[pp_entsprechend] ||| 1 0.000385979 1 0.00100755

according to your [X][nn] [X] ||| entsprechend ihren [X][nn]
[pp_entsprechend] ||| 0.05 0.000257047 0.166667 0.00130635
according to your [X][nn] [X] ||| laut auskunft ihrer [X][nn] [pp_laut] |||
1 0.00240725 0.166667 7.00232e-06
according to your [X][nn] [X] ||| nach ihrem [X][nn] [pp_nach] |||
0.0416667 0.000267865 0.5 0.008985
according to your [X][nn] [X] ||| nach ihren [X][nn] [pp_nach] |||
0.0217391 0.000178388 0.166667 0.0116496


To my understanding, the target-side non-terminals are copied to the
source-side string for technical reasons only in a string-to-tree system:
shouldn't then source-side strings as in the example
above be counted as one string ("according to your [X]") instead of
differentiating between
"according to your [X][cnp]" and "according to your [X][nn]" ?


Thank you in advance for helping with this,
Marion
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://mailman.mit.edu/mailman/private/moses-support/attachments/20140127/878ac9e3/attachment-0001.htm

------------------------------

Message: 2
Date: Mon, 27 Jan 2014 09:17:20 -0500
From: Philipp Koehn <pkoehn@inf.ed.ac.uk>
Subject: Re: [Moses-support] Difference between lexical reordering and
distortion?
To: Andrew <ravenyj@hotmail.com>
Cc: "moses-support@mit.edu" <moses-support@mit.edu>
Message-ID:
<CAAFADDBhyG-ydBkdEJO0Bsh4-ZG1_rNOVq_b5Pcss29og3k2Jg@mail.gmail.com>
Content-Type: text/plain; charset=ISO-8859-1

Hi,

both models deal with reordering.

The "distortion" model assigns a cost relative to the distance of jumps.
The lexicalized reordering model learns for each phrase pair if it is
likely to be reordered. Typically both are used.

-phi

On Sun, Jan 26, 2014 at 3:04 PM, Andrew <ravenyj@hotmail.com> wrote:
> Hi,
>
> in the tutorial (http://www.statmt.org/moses/?n=Moses.Tutorial),
>
> distortion model is said to be responsible for the reordering of the input,
>
> but in moses.ini file, there are separate weights for lexical reordering and
> distortion model.
>
> So I was wondering how they are different.
>
> Thank you in advance for your help.
>
> _______________________________________________
> Moses-support mailing list
> Moses-support@mit.edu
> http://mailman.mit.edu/mailman/listinfo/moses-support
>


------------------------------

Message: 3
Date: Mon, 27 Jan 2014 14:24:04 +0000 (UTC)
From: Rico Sennrich <rico.sennrich@gmx.ch>
Subject: Re: [Moses-support] translation probabilities in a
string-to-tree system
To: moses-support@mit.edu
Message-ID: <loom.20140127T151910-532@post.gmane.org>
Content-Type: text/plain; charset=utf-8

Marion Weller <wellermn@...> writes:

> To my understanding, the target-side non-terminals are copied to the
source-side string for technical reasons only in a string-to-tree system:
shouldn't then source-side strings as in the example
> above be counted as one string ("according to your [X]") instead of
differentiating between "according to your [X][cnp]" and? "according to your
[X][nn]" ?
>
> Thank you in advance for helping with this,
> Marion
>

Hi Marion,

the conditioning that you want is achieved with the option
"--alt-direct-rule-score-2". I agree that this is the more intuitive
definition, and I'm not sure if the 'weird' default behaviour (and this
flag) is documented anywhere.

best wishes,
Rico



------------------------------

_______________________________________________
Moses-support mailing list
Moses-support@mit.edu
http://mailman.mit.edu/mailman/listinfo/moses-support


End of Moses-support Digest, Vol 87, Issue 61
*********************************************

Tweet