Moses-support Digest, Vol 98, Issue 53

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: NoDaLiDa 2015: 1st Call for Demonstrations (Be?ta Megyesi)
2. CFP: CICLing 2015 / NLP - Egypt - Springer LNCS
(Alexander Gelbukh (CFP))
3. Re: LM interpolation with open source tools in EMS
(Kenneth Heafield)
4. Mgiza - lock contention? (Marcin Junczys-Dowmunt)
5. Re: Mgiza - lock contention? (Rico Sennrich)


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

Message: 1
Date: Thu, 18 Dec 2014 06:37:30 +0000
From: Be?ta Megyesi <Beata.Megyesi@lingfil.uu.se>
Subject: Re: [Moses-support] NoDaLiDa 2015: 1st Call for
Demonstrations
To: ACL 2014 <acl@aclweb.org>, "<corpora@uib.no>" <corpora@uib.no>,
"<elsnet-list@elsnet.org>" <elsnet-list@elsnet.org>,
"<nodali@helsinki.fi>" <nodali@helsinki.fi>, "<nordlingnet@uib.no>"
<nordlingnet@uib.no>, "<nealt-members@iln.uio.no>"
<nealt-members@iln.uio.no>, "<alla@gslt.hum.gu.se>"
<alla@gslt.hum.gu.se>, "<hpsg-l@mailman.stanford.edu>"
<hpsg-l@mailman.stanford.edu>, "<lfg@listserv.linguistlist.org>"
<lfg@listserv.linguistlist.org>, "<moses-support@mit.edu>"
<moses-support@mit.edu>, "<mt-list@eamt.org>" <mt-list@eamt.org>
Cc: "nodalida15@stp.lingfil.uu.se" <nodalida15@stp.lingfil.uu.se>
Message-ID: <AD2C2F12-D9DF-4558-A2AA-34970214B404@lingfil.uu.se>
Content-Type: text/plain; charset="utf-8"

[with apologies for cross-posting]

20th Nordic Conference on Computational Linguistics (NoDaLiDa)

Monday, May 11, to Wednesday, May 13, 2015
Institute of the Lithuanian Language, Vilnius, Lithuania

http://lki.lt/nodalida/
https://easychair.org/conferences/?conf=nodalida15


CALL FOR DEMONSTRATIONS

The Northern European Association for Language Technology (NEALT)
invites proposals for demonstrations at its bi-annual conference on
human language and speech technologies: NoDaLiDa 2015 will be held
between May 11 and May 13, 2015 on the campus of the University of
Vilnius, Lithuania. Please mark your calendars.

CONFERENCE HIGHLIGHTS

As part of the main conference, there will be three invited keynotes:
+Kevin Knight, ISI, University of Southern California, USA
+Catherine Pelachaud, CNRS, TELECOM Paris Tech, France
+Sebastian Riedel, University College London, UK

Preceding the main conference, on Monday, May 11, there will be four
topical workshops:

+ 4th workshop on NLP for Computer Assisted Language Learning (NLP4CALL)
+ Semantic resources and semantic annotation for Natural Language Processing and the Digital Humanities
+ Innovative Corpus Query and Visualization Tools
+ Constraint Grammar - Methods, Tools and Applications

Workshops will issue individual call for papers in mid-January and will
have a closing date for submissions in March.

SUBMISSIONS

NoDaLiDa addresses all aspects of speech recognition and synthesis,
natural language processing, and computational linguistics, including
work in closely related neighbouring disciplines (such as, for example,
linguistics or psychology) that is sufficiently formalized or applied
to bear relevance to speech and language technologies.

We invite submissions for software or resource demonstrations, e.g. of
systems, interfaces, infrastructures, data collections, or annotations.
Demonstration submissions should outline the nature and scope of what
will be demonstrated, including sufficient detail to allow evaluation
of its validity, quality, and relevance to the NoDaLiDa community and
topics of interest (see above).

Demonstration papers accepted for presentation at the conference will
be included in the NoDaLiDa 2015 proceedings, and published as part of the
NEALT Proceedings Series by Link?ping University Electronic Press.
The proceedings will be publicly available in the ACL Anthology:
http://aclweb.org/anthology-new/

As part of the main conference, demonstrations will be presented parallel
to the poster session(s), where each demonstration will have available
(at least) a table and large screen, as well as space for an overview
poster. In case there are technical needs beyond power and (wireless)
network access, these must be specified as part of the submission.

SCHEDULE

+ Monday, March 23, 2015: Submission of Papers
+ Monday, April 6, 2015: Notification of Acceptance
+ Monday, April 13, 2015: Camera-Ready Manuscripts
+ Monday, May 11, 2015: Pre-Conference Workshops
+ Tuesday & Wednesday, May 12 & 13, 2015: Main Conference

SUBMISSION FORMATS

All submissions must follow the NoDaLiDa 2015 style files, which are
available for LaTeX (preferred) or MS Word and can be
retrieved from the following address:

http://stp.lingfil.uu.se/~bea/nodalida15/

Submissions for demonstrations need not be anonymous. Demonstration
papers must be submitted digitally, in PDF, and uploaded through the
on-line conference system. Paper submissions that violate the format
requirements will be returned without review.

Demonstration papers can be up to three pages long,
excluding any additional pages with bibliographic references.

SUBMISSION MANAGEMENT

Submissions to the conference must be uploaded electronically, obeying
the above requirements and no later than (end of day, world-wide):

+ Monday, March 23, 2015.

NoDaLiDa 2015 utilizes the EasyChair conference management system for
submission, reviewing, and preparation of proceedings. Submission for
the conference is now open at:

https://easychair.org/conferences/?conf=nodalida15

PROGRAMME COMMITTEE

Reviewing of submissions and selection of the conference programme will
be managed by the NoDaLiDa 2015 Programme Committee. Submissions will
be reviewed by (at least) two members of the programme committee or by
other experts in the field.

+ Be?ta Megyesi (program chair), Uppsala University, Sweden
+ Filip Ginter, University of Turku, Finland
+ Kristiina Jokinen, University of Helsinki, Finland
+ Arne J?nsson, Link?ping University, Sweden
+ Violeta Meili?nait?, Institute of the Lithuanian Language, Lithuania
+ Costanza Navarretta, University of Copenhagen, Denmark
+ Stephan Oepen, University of Oslo, Norway
+ Oscar T?ckstr?m, Google Inc.

To inquire about the submission and reviewing process or generally the
scientific programme of the conference, please email
"nodalida15@stp.lingfil.uu.se".

LOCAL ORGANIZATION

The NoDaLiDa 2015 Organizing Committee is comprised of staff from the
Institute of the Lithuanian Language, Vilnius, Lithuania:

+ Jolanta Zabarskaite
+ Violeta Meili?nait?

For all practical inquiries, please email "lki@lki.lt".




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

Message: 2
Date: Thu, 18 Dec 2014 17:19:09 -0600
From: "Alexander Gelbukh \(CFP\)" <cfp2015a@cicling.org>
Subject: [Moses-support] CFP: CICLing 2015 / NLP - Egypt - Springer
LNCS
To: "'NLP group'" <cfp2015a@cicling.org>
Message-ID: <001601d01b19$0a0d71d0$1e285570$@cicling.org>
Content-Type: text/plain; charset="utf-8"

CICLing 2015

16th Conference on
Intelligent Text Processing and Computational Linguistics
+ 1st Workshop on Arabic NLP / sentiment analysis

Cairo, Egypt - April 14-20, 2015

Springer LNCS, journals

Deadline: Abstract January 25 then full text February 1

www.CICLing.org/2015

TOPICS:
All topics related to computational linguistics,
natural language processing, human language technologies,
information retrieval, opinion mining, sentiment analysis, etc.

PUBLICATION:
LNCS - Springer Lecture Notes in Computer Science;
poster session: special issues of journals

KEYNOTE SPEAKERS:
Lauri Karttunen (Stanford U.), Joakim Nivre (Uppsala U.),
Mona Talat Diab (George Washington U.), Erik Cambria (Nanyang Tech. U.)

CULTURAL PROGRAM:
Three days of cultural activities:
Giza pyramids, Old Cairo, Cairo Museum,
pre- and post-conf tours to Luxor and/or Aswan (for separate fee)

AWARDS:
Best paper, best student paper,
best presentation, best poster, best software

SUBMISSION DEADLINES:
January 25: registration of tentative abstract / title
(why not register your tentative abstract right now?),
February 1: full text of the registered papers
(contact us for late submissions)

SAFETY:
Yes, it is safe. More details on the webpage.

See complete CFP and contact on www.CICLing.org/2015





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

Message: 3
Date: Sun, 21 Dec 2014 15:27:53 -0500
From: Kenneth Heafield <moses@kheafield.com>
Subject: Re: [Moses-support] LM interpolation with open source tools
in EMS
To: moses-support@mit.edu
Message-ID: <54972D49.4070400@kheafield.com>
Content-Type: text/plain; charset=windows-1252

Hi,

I've been meaning to fix that. We started at MT Marathon with a proof
that you can't:

1. Linearly interpolate
2. Produce a single ARPA file interpreted as a backoff model
3. Have the union of n-grams in the combined model

SRILM fails on #1. Should implement what we designed. . .

Kenneth

On 12/21/2014 06:31 AM, Eleftherios Avramidis wrote:
> Hi,
>
> although there has been a shift to open source software such as IRSTLM
> and KENLM, the default example configuration script for EMS still
> requires SRILM for LM interpolation.
>
> There has been a discussion in this list a couple of years ago:
> http://comments.gmane.org/gmane.comp.nlp.moses.user/7227
>
> Has there been any progress on integration of any new tool in the
> section for LM interpolation?
>
> best
> Lefteris
> _______________________________________________
> Moses-support mailing list
> Moses-support@mit.edu
> http://mailman.mit.edu/mailman/listinfo/moses-support
>


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

Message: 4
Date: Mon, 22 Dec 2014 15:04:50 +0100
From: Marcin Junczys-Dowmunt <junczys@amu.edu.pl>
Subject: [Moses-support] Mgiza - lock contention?
To: Moses Support <moses-support@mit.edu>
Message-ID: <9cac644ba87d43f016eafc5368f86540@amu.edu.pl>
Content-Type: text/plain; charset="utf-8"



Hi,
In mgiza, I noticed that the HMM model is a lot slower with 8 threads
than with 4 threads. It becomes a lot worse for 16 threads or more.
Multi-threading is the main feature of mgiza, so this kind of defies the
whole sense of the thing to the extend that I would say mgiza is broken.
Haven't tested the higher models yet, but I guess it is not going to be
much better. Tried to link to tcmalloc, but that does not seem to help.

150,000 sentence corpus, first iteration of HMM:

2 threads: 683s

3 threads: 532s

4 threads: 521s

6 threads: 632s

8 threads: 777s

16 threads: 858s

This is quite horrible, at least on my system.
I will try to have a look what is going on there (I dread the code), but
if anyone has some ideas, the same experience or wants to help that
would be most welcome. It seems Model 1 is fine, and speed improves with
a greater number of threads, but the other models are off. Also htop
shows mgiza is mostly in kernel-mode (all red bars instead of green)
once the HMM iterations start.

Best,

Marcin

-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://mailman.mit.edu/mailman/private/moses-support/attachments/20141222/9857a991/attachment-0001.htm

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

Message: 5
Date: Mon, 22 Dec 2014 15:12:33 +0000 (UTC)
From: Rico Sennrich <rico.sennrich@gmx.ch>
Subject: Re: [Moses-support] Mgiza - lock contention?
To: moses-support@mit.edu
Message-ID: <loom.20141222T155042-90@post.gmane.org>
Content-Type: text/plain; charset=us-ascii

Marcin Junczys-Dowmunt <junczys@...> writes:

>
>
> will try to have a look what is going on there (I dread the code), but if
anyone has some ideas, the same experience or wants to help that would be
most welcome. It seems Model 1 is fine, and speed improves with a greater
number of threads, but the other models are off. Also htop shows mgiza is
mostly in kernel-mode (all red bars instead of green) once the HMM
iterations start.

Hi Marcin,

looking at hmm.cpp (lines 446-485), some counts get synchronized between
threads for every sentence. Maybe I missed something and the counts actually
need to be synchronized ASAP, but it looks to me as if every thread could
work with its own data structure, with a single synchronization between
threads after the loop.

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 98, Issue 53
*********************************************

Tweet