Moses-support Digest, Vol 168, Issue 9

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. MOSES | Baseline | Can't get similar BLEU (Paul Pallaghy)
2. Re: MOSES | Baseline | Can't get similar BLEU (Hieu Hoang)
3. PhD position in Natural Language Processing at Universitat
Polit?cnica de Catalunya (Marta Ruiz)


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

Message: 1
Date: Tue, 20 Oct 2020 14:52:12 +1100
From: Paul Pallaghy <paul.k.pallaghy@gmail.com>
Subject: [Moses-support] MOSES | Baseline | Can't get similar BLEU
To: moses-support@mit.edu
Message-ID:
<CANmSjsTHjQrAXwSLxQmUPK+vqXeH_VJjh8JcqUEEziNX9BiOOg@mail.gmail.com>
Content-Type: text/plain; charset="utf-8"

Hi all
I'm trying to reproduce the Moses baseline training for fr->en
http://www.statmt.org/moses/?n=Moses.Baseline <moses-support@mit.edu>
using around 130K sentences (as an exercise) and it works BUT instead of a
BLEU of 23.50 like they do on that Moses webpage, I get quite low BLEU
(7.0-ish by memory).

Anybody else experience this?
Any suggestions?
Any other worked training examples with reproducible BLEU on the web?

Paul Pallaghy
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://mailman.mit.edu/mailman/private/moses-support/attachments/20201019/c583e54d/attachment-0001.html

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

Message: 2
Date: Wed, 21 Oct 2020 16:56:40 -0700
From: Hieu Hoang <hieuhoang@gmail.com>
Subject: Re: [Moses-support] MOSES | Baseline | Can't get similar BLEU
To: Paul Pallaghy <paul.k.pallaghy@gmail.com>, moses-support@mit.edu
Message-ID: <58514aef-1a91-18f4-a1c0-429c5c55affc@gmail.com>
Content-Type: text/plain; charset="utf-8"

That webpage is a good starting point to learn Moses. You should come
out roughly the same BLEU

double check every command you run. A little thing like not tokenizing
or truecasing correctly would kill your bleu.

Another example you can try is to reproduce the pre-made models
available on the website. Download the config file eg

http://www.statmt.org/moses/RELEASE-4.0/models/de-en/steps/1/config.1

Change paths so they point to your executables and data.

These example uses the Experiment Management System, which is described
here:

?? http://www.statmt.org/moses/?n=FactoredTraining.EMS


On Mon Oct 19,20 8:52 PM, Paul Pallaghy wrote:
> Hi all
> I'm trying to reproduce the Moses baseline training?for fr->en
> http://www.statmt.org/moses/?n=Moses.Baseline
> <mailto:moses-support@mit.edu>
> using around 130K sentences (as an exercise) and it works BUT instead
> of a BLEU of 23.50 like they do on that Moses webpage, I get quite low
> BLEU (7.0-ish by memory).
>
> Anybody else experience?this?
> Any suggestions?
> Any other worked training examples with reproducible BLEU on the web?
>
> Paul Pallaghy

--
Hieu Hoang
http://statmt.org/hieu

-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://mailman.mit.edu/mailman/private/moses-support/attachments/20201021/4a234889/attachment-0001.html

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

Message: 3
Date: Thu, 22 Oct 2020 12:20:32 +0200
From: Marta Ruiz <martaruizcostajussa@gmail.com>
Subject: [Moses-support] PhD position in Natural Language Processing
at Universitat Polit?cnica de Catalunya
To: Marta Ruiz <martaruizcostajussa@gmail.com>
Message-ID:
<CABEBqH+_Ob4RLBGpjUsdTgT1xhAi=+=8Bhr8cRCU9na4C19BVQ@mail.gmail.com>
Content-Type: text/plain; charset="utf-8"

Job openings: 1 PhD position in Machine Learning and Natural Language
Processing at Universitat Polit?cnica de Catalunya, Barcelona

Project: LUNAR ("Lifelong UNiversal lAnguage Representation"), ERC Starting
Grant (2020-2025)

Application deadline: November 15, 2020

Job requirements: Bachelor?s degree and a master?s degree in the field of
computer science/engineering or mathematics. Python programming. Experience
in Deep Learning.

Job description:

There is one open position for a PhD researcher in the scope of the ERC
project <https://deep-spin.github.io>LUNAR ("Lifelong UNiversal lAnguage
Representation"). This project addresses the low-resources problem and the
expensive approach to multilingual machine translation since systems for
all translation pairs are required. LUNAR proposes to jointly learn a
multilingual and multimodal model that builds upon a lifelong universal
language representation. This model will compensate for the lack of
supervised data and significantly increase the system capacity of
generalization from training data given the unconventional variety of
employed resources. This model will reduce the number of required
translation systems from quadratic to linear as well as allowing for an
incremental adaptation of new languages and data.

The role of this PhD is to focus on the common multilingual text
representation and the integration of partial resources.

This offer is a 3-year PhD position. If interested in this position, please
send an email with your CV to marta.ruiz AT upc.edu and the subject title
?[LUNAR] PhD Position?. Feel free to contact me for any further information.

Working environment:

The host institution (Universitat Polit?cnica de Catalunya) is a public
institution of research and higher education in the fields of engineering,
architecture, sciences and technology, and one of the leading technical
universities in Europe. Every year, more than 6,000 bachelor?s
<https://www.upc.edu/en/bachelors> and master?s
<https://www.upc.edu/en/masters> students and more than 500 doctoral
<https://doctorat.upc.edu/en> students graduate, more than 3,000 faculty
members, and a total of 12 ERC grants.

The principal investigator is part of the ELLIS network
<https://ellis.eu/units>. Collaborations with other ELLIS members are
encouraged.

Salary: 1,200-1,400 (free of tax) euros

More info about the research group at MT-UPC <https://mt.cs.upc.edu/>


--
Marta Ruiz Costa-juss?
http://www.costa-jussa.com
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://mailman.mit.edu/mailman/private/moses-support/attachments/20201022/1bbfdeea/attachment.html

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

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


End of Moses-support Digest, Vol 168, Issue 9
*********************************************

Tweet