Re: [Presage-user] Standard location of presage data

[Matteo V. <mat...@ya...>]

Hi Weng,

Presage is designed to provide support for any natural language. One 
core design feature of presage is the ability to make use of different 
sources of information and predictive algorithms to generate 
predictions. The sources of information and predictive algorithms used 
to generate a prediction are controlled by presage configuration. The 
default build of presage, for example, is configured to use three 
predictors:
- the smoothed n-gram predictive algorithm with a language model that is 
generated from a single training text (the novel "The picture of Dorian 
Gray" by Oscar Wilde)
- the abbreviation expansion predictor
- the recency predictor

The Presage.PredictorRegistry.PREDICTORS configuration variable controls 
what predictors should be instantiated. The elements nested within the 
Presage.Predictors element control the configuration of each instance of 
the predictors.

By default, presage currently builds language models for the English, 
Italian, and Spanish languages. These language models are generated 
during the build from source and are installed in the share/presage/ 
directory of your presage installation.

In order to switch to a different language, you need to modify the 
configuration used by presage to suit your needs. You can achieve that 
by editing your presage.xml configuration file. The configuration file 
is read from the following locations:
  1. /etc/presage.xml if it exists
  2. <installation_directory>/etc/presage.xml if it exists
  3. ~/.presage.xml if it exists ($HOME/.presage/.presage.xml on Unix, 
%USERPROFILE%/.presage.xml on Windows) [1]

Configuration values in 3. take priority over configuration values in 2. 
and 1.

You can also modify the configuration programmatically through the 
Presage::config(...) calls or a variant of the Presage class constructor.

The Predictor.SmoothedNgramPredictor.DBFILENAME config variable controls 
the language model database used by the Smoothed Ngram predictive 
plugin, i.e.:

<Predictors>
<SmoothedNgramPredictor>
<DBFILENAME>/home/matt/presage_local_install/share/presage/database_en.db</DBFILENAME>

For example, suppose that our goal is to enable support for the German 
language.
In order to achieve that, you will want to modify this variable to point 
to a German language model.
presage provides the tools to generate predictive resources required by 
the predictors to suit any natural language or specific users' needs.
You can generate a German language model database using the text2ngram 
tool. The text2ngram tool generates n-gram language models from a given 
corpus of text.
Ideally, you would collect a representative set of text (text that the 
user has produced or text that matches the writing style and context of 
the user) and then feed that to the text2ngram tool to generate a n-gram 
database (1-gram, 2-gram and 3-gram tables, but you can also generate 
higher order n-gram and set the SmoothedNgram DELTAS values accordingly).

Obviously, the higher n-gram order used, the higher the risk of 
overtraining the model and overfitting the training corpus (wikipedia 
has good articles about machine learning and overfitting).

Assuming you have a German training text corpus in a file named 
training_de.txt, you can generated a 3-gram language model database for 
use with the smoothed n-gram predictor by running the following commands:

      for i in 1 2 3; do text2ngram -n $i -l -f sqlite -o database_de.db 
training_de.txt; done


So, to sum it up, Presage does not provide an API to query what language 
models are available. This is controlled through configuration by the 
application invoking on presage.


Hope this helps,
- Matteo


[1] It is likely that future presage releases (starting from presage 
0.8.8) will use ~/.presage/presage.xml instead of ~/.presage.xml.


On 10/05/12 06:20, Weng Xuetian wrote:
>
> Hi,
>
> Currently I'm using presage to implement word completion in my input 
> method, I found presage it self provides several languages, but it 
> doesn't provide some API like enchant to list supported language, I 
> could do some manual search in my input method, but I'm not sure the 
> path will changed in the future or not.
>
> So would presage provide some standard "supported language" API for 
> doing this? Thanks!
>
> Regards
>
>
> ------------------------------------------------------------------------------
> Live Security Virtual Conference
> Exclusive live event will cover all the ways today's security and
> threat landscape has changed and how IT managers can respond. Discussions
> will include endpoint security, mobile security and the latest in malware
> threats. http://www.accelacomm.com/jaw/sfrnl04242012/114/50122263/
>
>
> _______________________________________________
> Presage-user mailing list
> Pre...@li...
> https://lists.sourceforge.net/lists/listinfo/presage-user
>