Moses-support Digest, Vol 124, Issue 29

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. Re: ReportingOptions and mosesserver XML (Ulrich Germann)
2. Re: SMT decoding complexity (Philipp Koehn)
3. Re: SMT decoding complexity (Barry Haddow)


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

Message: 1
Date: Mon, 27 Feb 2017 15:39:13 +0100
From: Ulrich Germann <ugermann@inf.ed.ac.uk>
Subject: Re: [Moses-support] ReportingOptions and mosesserver XML
To: Lane Schwartz <dowobeha@gmail.com>
Cc: "moses-support@mit.edu" <moses-support@mit.edu>
Message-ID:
<CAHQSRUp8BinPVzxQnM6RYnQBY2p8huCcAtRGw_zZYGtuduc6sA@mail.gmail.com>
Content-Type: text/plain; charset="utf-8"

If I understand you correctly you are worrying about changing global
options with XML translation requests sent to the server. This is not the
case. Each request gets its own local copy of parameters at line
https://github.com/moses-smt/mosesdecoder/blob/master/moses/server/TranslationRequest.cpp:276,
so all parameter changes are local to the XML request being processed.

- Uli

On Mon, Feb 20, 2017 at 6:39 PM, Lane Schwartz <dowobeha@gmail.com> wrote:

> Hi,
>
> In moses/parameters/ReportingOptions.cpp, the following code was added by
> Uli in commit 69e5a1cc152de40 about a year ago:
>
>
> m = param.find("align");
> if (m != param.end() && Scan<bool>(xmlrpc_c::value_string(m->second)))
> ReportSegmentation = 1;
>
>
> This is relevant when using XML-RPC requests via mosesserver, and it turns
> on ReportSegmentation if the XML response includes segmentation. Most of
> the time this would be superfluous, since the reason that the XML response
> contains segmentation is typically because ReportSegmentation was already
> set to 1 in the moses server config file.
>
> I'm interested in re-enabling the moses server code that allows this to
> alternatively be set via XML-RPC request, but I don't want
> ReportSegmentation turned on in this case. The reason is that I want the
> segmentation information, but I don't want the returned hypothesis to
> include the segmentation in the reported hypothesis string. If
> ReportSegmentation is set to 1, then that will happen.
>
> My question is, is it OK to delete or comment out this block of code?
>
> If someone has a good reason why it's there, that's fine. But I can't seem
> to find one.
>
> Thanks,
> Lane
>



--
Ulrich Germann
Senior Researcher
School of Informatics
University of Edinburgh
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://mailman.mit.edu/mailman/private/moses-support/attachments/20170227/a410fe70/attachment-0001.html

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

Message: 2
Date: Mon, 27 Feb 2017 10:54:16 -0500
From: Philipp Koehn <phi@jhu.edu>
Subject: Re: [Moses-support] SMT decoding complexity
To: amir haghighi <amir.haghighi.tehrani@gmail.com>
Cc: Moses-support <moses-support@mit.edu>
Message-ID:
<CAAFADDA7wD1mrEXUfPMvt46zkhPqEeg0L+y2EHasb-7AabJpvg@mail.gmail.com>
Content-Type: text/plain; charset=UTF-8

Hi,

I am not sure if you follow your question - in the formula you cite,
there are exponential terms: 2^n and T^n.

The Knight paper is worth trying to understand (it's on IBM Models,
but applies similarly to phrase-based models).

Also keep in mind that limited reordering windows and beam search
makes actual decoding algorithm implementations linear.

-phi

On Sun, Feb 26, 2017 at 1:16 PM, amir haghighi
<amir.haghighi.tehrani@gmail.com> wrote:
> Hi all,
>
> In the Moses manual and also in SMT textbooks it is mentioned that the
> decoding complexity for PB-SMT is exponential in the source sentence length.
> If we have a source sentence with length n, in decoding by hypothesis
> expansion, we have power(2,n) state, each of them can be reordered in n!
> orders, and each state can be translated in power(T,n), where T is the
> number of translation options, right?
> so the decoder complexity is power(2,n)*n!*power(T,n), so why its mentioned
> that the complexity is exponential?
>
> Could someone please explain for me how the decoder complexity is
> calculated?
> I've read the Knight(1999) paper, but I couldn't understand it. Could you
> please introduce another reference?
>
> Thanks
>
>
> _______________________________________________
> Moses-support mailing list
> Moses-support@mit.edu
> http://mailman.mit.edu/mailman/listinfo/moses-support
>


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

Message: 3
Date: Mon, 27 Feb 2017 16:04:48 +0000
From: Barry Haddow <bhaddow@staffmail.ed.ac.uk>
Subject: Re: [Moses-support] SMT decoding complexity
To: Philipp Koehn <phi@jhu.edu>, amir haghighi
<amir.haghighi.tehrani@gmail.com>
Cc: Moses-support <moses-support@mit.edu>
Message-ID: <00a2dfb9-818c-d259-b1e5-b443f450afc0@staffmail.ed.ac.uk>
Content-Type: text/plain; charset=windows-1252; format=flowed

Hi Amir

You could also try this paper for a derivation of the complexity of PBMT
decoding
https://www.aclweb.org/anthology/E/E09/E09-1061v2.pdf

cheers - Barry

On 27/02/17 15:54, Philipp Koehn wrote:
> Hi,
>
> I am not sure if you follow your question - in the formula you cite,
> there are exponential terms: 2^n and T^n.
>
> The Knight paper is worth trying to understand (it's on IBM Models,
> but applies similarly to phrase-based models).
>
> Also keep in mind that limited reordering windows and beam search
> makes actual decoding algorithm implementations linear.
>
> -phi
>
> On Sun, Feb 26, 2017 at 1:16 PM, amir haghighi
> <amir.haghighi.tehrani@gmail.com> wrote:
>> Hi all,
>>
>> In the Moses manual and also in SMT textbooks it is mentioned that the
>> decoding complexity for PB-SMT is exponential in the source sentence length.
>> If we have a source sentence with length n, in decoding by hypothesis
>> expansion, we have power(2,n) state, each of them can be reordered in n!
>> orders, and each state can be translated in power(T,n), where T is the
>> number of translation options, right?
>> so the decoder complexity is power(2,n)*n!*power(T,n), so why its mentioned
>> that the complexity is exponential?
>>
>> Could someone please explain for me how the decoder complexity is
>> calculated?
>> I've read the Knight(1999) paper, but I couldn't understand it. Could you
>> please introduce another reference?
>>
>> Thanks
>>
>>
>> _______________________________________________
>> Moses-support mailing list
>> Moses-support@mit.edu
>> http://mailman.mit.edu/mailman/listinfo/moses-support
>>
> _______________________________________________
> Moses-support mailing list
> Moses-support@mit.edu
> http://mailman.mit.edu/mailman/listinfo/moses-support
>


--
The University of Edinburgh is a charitable body, registered in
Scotland, with registration number SC005336.



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

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


End of Moses-support Digest, Vol 124, Issue 29
**********************************************

Tweet