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: Moseserver "report-all-factors" marks all words as UNK
(Etienne Monneret (LM))
2. Metrics Task -- signup for manual evaluation (Ondrej Bojar)
3. Re: Moseserver "report-all-factors" marks all words as UNK
(Hieu Hoang)
----------------------------------------------------------------------
Message: 1
Date: Fri, 16 Jun 2017 08:55:17 +0200
From: "Etienne Monneret (LM)" <em@lingua-et-machina.com>
Subject: Re: [Moses-support] Moseserver "report-all-factors" marks all
words as UNK
To: moses-support <moses-support@mit.edu>
Message-ID:
<c467b49f-7802-8e49-1908-bd67a4cc1189@lingua-et-machina.com>
Content-Type: text/plain; charset="utf-8"
The idea of this code was certainly to produce an output of each factor,
for each token.
With the functioning of outputFactorOrder[i], it's not a good idea to
remove this line : it produces empty factor output for many tokens, and
it could produce a bad factors order, due to the lack of one of them for
a given token (if this is possible).
Perhaps it would have been better to keep producing a value, but not
UNK, standing for "unknown token". Rather, for example, UND, standing
for "undefined factor".
A finalized solution would have been to really have all needed factors
defined for each token with a properly set value...
Le 15/06/2017 ? 15:27, Hieu Hoang a ?crit :
> thanks. Committed
> https://github.com/moses-smt/mosesdecoder/commit/4b0560b5c9bd95d7c55cb0451e8947de0eee1d6d
>
> * Looking for MT/NLP opportunities *
> Hieu Hoang
> http://moses-smt.org/
>
>
> On 15 June 2017 at 14:07, Etienne Monneret (LM)
> <em@lingua-et-machina.com <mailto:em@lingua-et-machina.com>> wrote:
>
> Manager.cpp
> OutputSurface(..)
>
> Replace :
>
> for (size_t i = 1 ; i < outputFactorOrder.size() ; i++) {
> const Factor *factor = phrase.GetFactor(pos, outputFactorOrder[i]);
> if (factor) out << fd << *factor;
> else out << fd << UNKNOWN_FACTOR;
> }
>
> By :
>
> for (size_t i = 1 ; i < outputFactorOrder.size() ; i++) {
> const Factor *factor = phrase.GetFactor(pos, outputFactorOrder[i]);
> if (factor) out << fd << *factor;
> // else out << fd << UNKNOWN_FACTOR;
> }
>
>
>
> Best regards,
>
> Etienne
>
>
> Le 07/06/2017 ? 17:23, Hieu Hoang a ?crit :
>> there's probably a bug somewhere in the server code
>>
>> * Looking for MT/NLP opportunities *
>> Hieu Hoang
>> http://moses-smt.org/
>>
>>
>> On 7 June 2017 at 11:02, Etienne Monneret (LM)
>> <em@lingua-et-machina.com <mailto:em@lingua-et-machina.com>> wrote:
>>
>> Hi !
>>
>> I just re-compiled a Moses server.
>>
>> Now, the "report-all-factors" option is marking all words as UNK:
>>
>> taking|UNK|UNK|UNK advantage|UNK|UNK|UNK of|UNK|UNK|UNK
>> the|UNK|UNK|UNK
>> |0-1| mushrooms|UNK|UNK|UNK |2-2| relaxante|UNK|UNK|UNK |6-6|
>> is|UNK|UNK|UNK an|UNK|UNK|UNK activity|UNK|UNK|UNK |3-5|
>> .|UNK|UNK|UNK |7-7|
>>
>> This is only in the XML-RPC reply, because, in the
>> mosesserver LOG, I
>> really get the good words marked as UNK:
>>
>> [moses/server/TranslationRequest.cpp:472] BEST TRANSLATION:
>> taking
>> advantage of the mushrooms relaxante|UNK|UNK|UNK is an activity .
>> [11111111] [total=-108.208]
>> core=(-100.000,-10.000,5.000,-6.851,-16.168,-4.675,-21.479,-8.000,-60.608)
>>
>> Is there a new way to do this ?
>>
>> Best regards,
>>
>> Etienne
>>
>>
>> _______________________________________________
>> Moses-support mailing list
>> Moses-support@mit.edu <mailto:Moses-support@mit.edu>
>> http://mailman.mit.edu/mailman/listinfo/moses-support
>> <http://mailman.mit.edu/mailman/listinfo/moses-support>
>>
>>
>
>
> _______________________________________________
> Moses-support mailing list
> Moses-support@mit.edu <mailto:Moses-support@mit.edu>
> http://mailman.mit.edu/mailman/listinfo/moses-support
> <http://mailman.mit.edu/mailman/listinfo/moses-support>
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://mailman.mit.edu/mailman/private/moses-support/attachments/20170616/2e9f4fe0/attachment-0001.html
------------------------------
Message: 2
Date: Fri, 16 Jun 2017 14:11:09 +0200 (CEST)
From: Ondrej Bojar <bojar@ufal.mff.cuni.cz>
Subject: [Moses-support] Metrics Task -- signup for manual evaluation
To: wmt-tasks <wmt-tasks@googlegroups.com>, moses-support
<moses-support@mit.edu>, nmt-developers
<nmt-developers@googlegroups.com>
Cc: Amir Kamran <kamran@ufal.mff.cuni.cz>, Yvette Graham
<graham.yvette@gmail.com>
Message-ID:
<488262356.2212752.1497615069568.JavaMail.zimbra@ufal.mff.cuni.cz>
Content-Type: text/plain; charset=utf-8
Dear all,
all participants of the metrics task are expected to spend 8 hours of annotations (about 16 HITs) for each submission they make.
In the end, we *will* use Appraise and we are searching for people able to evaluate translations into Chinese, Finnish, German, Latvian, and Turkish. Please consider asking your colleagues if you don't speak the languages yourself.
To join the manual evaluation, please sign up to the following Google Sheet:
http://tinyurl.com/wmt17-appraise-signup
(You can do this now, it is not related to your metrics paper submission.)
Please create one line per research group and provide:
- the number of submissions your group is making
- a contact person for your group
- the number of invite tokens, i.e. a rough upper bound on the number of people who will work on the evaluation
The contact person will receive the invite tokens from Christian Federmann and should distribute them to the evaluators. Please note that invite tokens are one-time use only.
The total number of HITs annotated by each group will be listed most likely at this URL:
http://wmt17.appraise.cf/group-status/
Thanks,
Ondrej.
----- Original Message -----
> From: "Ondrej Bojar" <bojar@ufal.mff.cuni.cz>
> To: "wmt-tasks" <wmt-tasks@googlegroups.com>, "moses-support" <moses-support@mit.edu>, "nmt-developers"
> <nmt-developers@googlegroups.com>
> Cc: "Yvette Graham" <graham.yvette@gmail.com>, "Amir Kamran" <kamran@ufal.mff.cuni.cz>
> Sent: Thursday, 15 June, 2017 10:31:10
> Subject: Metrics Task Last Call and a Small Deadline Extension
> Dear all,
>
> This is the last call for participation in the WMT17 Metrics Task and we are
> happy to offer you at least a small deadline extension:
>
> * Papers: Saturday June 17 (anywhere on Earth)
> * Metrics themselves: Wednesday June 21 AoE, yes, later; you can add improved
> results to the camera-ready version
>
> Looking forward to your submissions.
>
> Please remember that in order to be included in the metrics evaluation, you
> should complete the requested number of human evaluation hits (depending on
> numbers of submissions your team will make).
>
> This year the metrics evaluation will run via the DFKI-DA system (as opposed to
> Appraise). Instructions will be sent tomorrow. Languages that will be evaluated
> are Chinese, Finnish, German, Latvian, and Turkish, as the other languages for
> the metrics task are covered by crowd-sourcing. Please consider also reaching
> out to your foreign colleagues if you don't speak these languages. Any
> problems, please let us know.
>
> Deadline for manual evaluation: already Friday June 23.
>
> Your metrics task organizers,
> Ondrej Bojar
> Yvette Graham
> Amir Kamran
> Raphael Rubino
>
> --
> Ondrej Bojar (mailto:obo@cuni.cz / bojar@ufal.mff.cuni.cz)
> http://www.cuni.cz/~obo
--
Ondrej Bojar (mailto:obo@cuni.cz / bojar@ufal.mff.cuni.cz)
http://www.cuni.cz/~obo
------------------------------
Message: 3
Date: Fri, 16 Jun 2017 14:43:12 +0100
From: Hieu Hoang <hieuhoang@gmail.com>
Subject: Re: [Moses-support] Moseserver "report-all-factors" marks all
words as UNK
To: "Etienne Monneret (LM)" <em@lingua-et-machina.com>
Cc: moses-support <moses-support@mit.edu>
Message-ID:
<CAEKMkbjTUOuEDwKry8e_-VvyMHEQgJXZSbe_vf=8Fsz6o=pM=g@mail.gmail.com>
Content-Type: text/plain; charset="utf-8"
i think the change you gave me was ok.
*It is not required to fill all the space in the array with a factor. Some
can be null.*
*If you have more code changes, always welcome to send it over again*
Hieu Hoang
http://moses-smt.org/
On 16 June 2017 at 07:55, Etienne Monneret (LM) <em@lingua-et-machina.com>
wrote:
> The idea of this code was certainly to produce an output of each factor,
> for each token.
>
> With the functioning of outputFactorOrder[i], it's not a good idea to
> remove this line : it produces empty factor output for many tokens, and it
> could produce a bad factors order, due to the lack of one of them for a
> given token (if this is possible).
>
> Perhaps it would have been better to keep producing a value, but not UNK,
> standing for "unknown token". Rather, for example, UND, standing for
> "undefined factor".
>
> A finalized solution would have been to really have all needed factors
> defined for each token with a properly set value...
>
>
> Le 15/06/2017 ? 15:27, Hieu Hoang a ?crit :
>
> thanks. Committed
> https://github.com/moses-smt/mosesdecoder/commit/
> 4b0560b5c9bd95d7c55cb0451e8947de0eee1d6d
>
> * Looking for MT/NLP opportunities *
> Hieu Hoang
> http://moses-smt.org/
>
>
> On 15 June 2017 at 14:07, Etienne Monneret (LM) <em@lingua-et-machina.com>
> wrote:
>
>> Manager.cpp
>> OutputSurface(..)
>>
>> Replace :
>>
>> for (size_t i = 1 ; i < outputFactorOrder.size() ; i++) {
>> const Factor *factor = phrase.GetFactor(pos, outputFactorOrder[i]);
>> if (factor) out << fd << *factor;
>> else out << fd << UNKNOWN_FACTOR;
>> }
>>
>> By :
>>
>> for (size_t i = 1 ; i < outputFactorOrder.size() ; i++) {
>> const Factor *factor = phrase.GetFactor(pos, outputFactorOrder[i]);
>> if (factor) out << fd << *factor;
>> // else out << fd << UNKNOWN_FACTOR;
>> }
>>
>>
>>
>>
>> Best regards,
>>
>> Etienne
>>
>>
>> Le 07/06/2017 ? 17:23, Hieu Hoang a ?crit :
>>
>> there's probably a bug somewhere in the server code
>>
>> * Looking for MT/NLP opportunities *
>> Hieu Hoang
>> http://moses-smt.org/
>>
>>
>> On 7 June 2017 at 11:02, Etienne Monneret (LM) <em@lingua-et-machina.com>
>> wrote:
>>
>>> Hi !
>>>
>>> I just re-compiled a Moses server.
>>>
>>> Now, the "report-all-factors" option is marking all words as UNK:
>>>
>>> taking|UNK|UNK|UNK advantage|UNK|UNK|UNK of|UNK|UNK|UNK the|UNK|UNK|UNK
>>> |0-1| mushrooms|UNK|UNK|UNK |2-2| relaxante|UNK|UNK|UNK |6-6|
>>> is|UNK|UNK|UNK an|UNK|UNK|UNK activity|UNK|UNK|UNK |3-5| .|UNK|UNK|UNK
>>> |7-7|
>>>
>>> This is only in the XML-RPC reply, because, in the mosesserver LOG, I
>>> really get the good words marked as UNK:
>>>
>>> [moses/server/TranslationRequest.cpp:472] BEST TRANSLATION: taking
>>> advantage of the mushrooms relaxante|UNK|UNK|UNK is an activity .
>>> [11111111] [total=-108.208]
>>> core=(-100.000,-10.000,5.000,-6.851,-16.168,-4.675,-21.479,-
>>> 8.000,-60.608)
>>>
>>> Is there a new way to do this ?
>>>
>>> Best regards,
>>>
>>> Etienne
>>>
>>>
>>> _______________________________________________
>>> 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
>>
>>
>
>
> _______________________________________________
> Moses-support mailing list
> Moses-support@mit.edu
> http://mailman.mit.edu/mailman/listinfo/moses-support
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://mailman.mit.edu/mailman/private/moses-support/attachments/20170616/95e9bcbd/attachment.html
------------------------------
_______________________________________________
Moses-support mailing list
Moses-support@mit.edu
http://mailman.mit.edu/mailman/listinfo/moses-support
End of Moses-support Digest, Vol 128, Issue 21
**********************************************
Subscribe to:
Post Comments (Atom)
0 Response to "Moses-support Digest, Vol 128, Issue 21"
Post a Comment