Moses-support Digest, Vol 184, Issue 3

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

To subscribe or unsubscribe via the World Wide Web, visit
https://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: hierarchical-to-string translation doesn't produce string
output (Hieu Hoang)
2. Re: hierarchical-to-string translation doesn't produce string
output (Oleg Parashchenko)
3. Re: hierarchical-to-string translation doesn't produce string
output (Lane Schwartz)


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

Message: 1
Date: Thu, 17 Mar 2022 23:09:25 -0700
From: Hieu Hoang <hieuhoang@gmail.com>
To: Oleg Parashchenko <olpa@uucode.com>
Cc: moses-support <moses-support@mit.edu>
Subject: Re: [Moses-support] hierarchical-to-string translation
doesn't produce string output
Message-ID: <15730ae4-e203-55b8-d0c5-379987b3926b@gmail.com>
Content-Type: text/plain; charset=UTF-8; format=flowed

the training does add glue rules for non-syntax models but you're using
a variant with syntax where you have to do it yourself.

DOwnload the small sample model and look in the tree-to-tree example to
see how it's done

?? http://www.statmt.org/moses/download/sample-models.tgz

Also, I hope you realise that these are research models. They don't
produce good results in practise and have largely been overtaken by
neural models

On 3/17/2022 10:16 PM, Oleg Parashchenko wrote:
> Hello Hieu,
>
>> ... You always need a glue rule
> I assumed that training should create one. If I need to write one self,
> where can I read the details? The page
> https://www.statmt.org/moses/?n=Moses.SyntaxTutorial mentions only the
> using of glue rules.
>
>> Can I have a look at your moses.ini file
> I've put the whole model at
> http://uucode.com/download/tmp/mini-hie-to-string.tar.gz,
> here is the ini file:
>
> ```
> #########################
> ### MOSES CONFIG FILE ###
> #########################
>
> # input factors
> [input-factors]
> 0
>
> # mapping steps
> [mapping]
> 0 T 0
>
> [cube-pruning-pop-limit]
> 1000
>
> [non-terminals]
> X
>
> [search-algorithm]
> 3
>
> [inputtype]
> 3
>
> [max-chart-span]
> 20
> 1000
>
> # feature functions
> [feature]
> UnknownWordPenalty
> WordPenalty
> PhrasePenalty
> PhraseDictionaryMemory name=TranslationModel0 num-features=4
> path=/share/mini-hie-to-string/model/rule-table.gz input-factor=0
> output-factor=0 KENLM name=LM0 factor=0
> path=/share/mini-hie-to-string/lm.en order=2
>
> # dense weights for feature functions
> [weight]
> # The default weights are NOT optimized for translation quality. You MUST tune the weights.
> # Documentation for tuning is here: http://www.statmt.org/moses/?n=FactoredTraining.Tuning
> UnknownWordPenalty0= 1
> WordPenalty0= -1
> PhrasePenalty0= 0.2
> TranslationModel0= 0.2 0.2 0.2 0.2
> LM0= 0.5
> ```
>
> Regards,
> Oleg
>
> On Thu, 17 Mar 2022 21:09:10 -0700
> Hieu Hoang <hieuhoang@gmail.com> wrote:
>
>> Can I have a look at your moses.ini file. You always need a glue rule
>> phrase take for scfg models to create the final output
> ...

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



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

Message: 2
Date: Fri, 18 Mar 2022 08:38:53 +0100
From: Oleg Parashchenko <olpa@uucode.com>
To: Hieu Hoang <hieuhoang@gmail.com>
Cc: moses-support <moses-support@mit.edu>
Subject: Re: [Moses-support] hierarchical-to-string translation
doesn't produce string output
Message-ID: <20220318083853.a274f150606aac4e4bef6a10@uucode.com>
Content-Type: text/plain; charset=UTF-8

Hello Hieu,

On Thu, 17 Mar 2022 23:09:25 -0700
Hieu Hoang <hieuhoang@gmail.com> wrote:

> the training does add glue rules for non-syntax models but you're using
> a variant with syntax where you have to do it yourself.

I didn't want to use a variant with syntax, I want to try "hierarchical
phrase-based: no linguistic syntax". Please suggest what I'm doing wrong
with my corpus and training parameters.

Here is the corpus:

```
<tree label="N">nhoj</tree>
<tree label="N">yram</tree>
<tree label="V">olleh</tree>
<tree label="V">eyb</tree>
<tree label="VP"><tree label="V">olleh</tree> <tree
label="N">hnoj</tree></tree>
```

Here is the training step:

```
opt/moses/scripts/training/train-model.perl \
-corpus corpus \
-f ne -e en \
-lm 0:2:$(pwd)/lm.en \
-alignment grow-diag-final-and \
-external-bin-dir /opt/tools \
-mgiza \
-hierarchical \
```

>
> Download the small sample model and look in the tree-to-tree example to
> see how it's done
>
> ?? http://www.statmt.org/moses/download/sample-models.tgz

Unfortunately, the samples contain only the resulting models, without
the data for the training step.

> Also, I hope you realise that these are research models. They don't
> produce good results in practise and have largely been overtaken by
> neural models

That's ok, for now I'm experimenting with rule-based approaches for low
resource languages.

Regards,
Oleg


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

Message: 3
Date: Fri, 18 Mar 2022 09:17:33 -0500
From: Lane Schwartz <dowobeha@gmail.com>
To: Oleg Parashchenko <olpa@uucode.com>
Cc: Hieu Hoang <hieuhoang@gmail.com>, moses-support
<moses-support@mit.edu>
Subject: Re: [Moses-support] hierarchical-to-string translation
doesn't produce string output
Message-ID:
<CABv3vZ=qup2k6A09QFBvgXRL33_E_JBrxAuDR7m8b3=X3RBSWg@mail.gmail.com>
Content-Type: text/plain; charset="utf-8"

Oleg,

If you?re looking for rule based approaches, I would also encourage you to
look at explicitly rule based systems like Apertium

On Fri, Mar 18, 2022 at 2:42 AM Oleg Parashchenko <olpa@uucode.com> wrote:

> Hello Hieu,
>
> On Thu, 17 Mar 2022 23:09:25 -0700
> Hieu Hoang <hieuhoang@gmail.com> wrote:
>
> > the training does add glue rules for non-syntax models but you're using
> > a variant with syntax where you have to do it yourself.
>
> I didn't want to use a variant with syntax, I want to try "hierarchical
> phrase-based: no linguistic syntax". Please suggest what I'm doing wrong
> with my corpus and training parameters.
>
> Here is the corpus:
>
> ```
> <tree label="N">nhoj</tree>
> <tree label="N">yram</tree>
> <tree label="V">olleh</tree>
> <tree label="V">eyb</tree>
> <tree label="VP"><tree label="V">olleh</tree> <tree
> label="N">hnoj</tree></tree>
> ```
>
> Here is the training step:
>
> ```
> opt/moses/scripts/training/train-model.perl \
> -corpus corpus \
> -f ne -e en \
> -lm 0:2:$(pwd)/lm.en \
> -alignment grow-diag-final-and \
> -external-bin-dir /opt/tools \
> -mgiza \
> -hierarchical \
> ```
>
> >
> > Download the small sample model and look in the tree-to-tree example to
> > see how it's done
> >
> > http://www.statmt.org/moses/download/sample-models.tgz
>
> Unfortunately, the samples contain only the resulting models, without
> the data for the training step.
>
> > Also, I hope you realise that these are research models. They don't
> > produce good results in practise and have largely been overtaken by
> > neural models
>
> That's ok, for now I'm experimenting with rule-based approaches for low
> resource languages.
>
> Regards,
> Oleg
> _______________________________________________
> Moses-support mailing list
> Moses-support@mit.edu
> https://mailman.mit.edu/mailman/listinfo/moses-support
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://mailman.mit.edu/mailman/private/moses-support/attachments/20220318/9fd5de44/attachment.htm>

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

Subject: Digest Footer

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


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

End of Moses-support Digest, Vol 184, Issue 3
*********************************************

Tweet