Originally Posted by itdoesntmatt (Post 1496351)

Ciao a te :) e grazie tante per il vostro impegno!

Originally Posted by spidernik84 (Post 1496352)

1) it's getting more complicated than I thought :D

2) It is a bit better:

Originally Posted by itdoesntmatt (Post 1496445)

eber sorry my ignorance but what do you mean for input corpus? however is it possibile to show both dall and dall' when swyped d-a-l-l? thanks

Originally Posted by itdoesntmatt (Post 1496445)

PS: anything about eventual solution to avoid prediction bar get hidden from whatsapp input field in dalvik full screen mode?

Originally Posted by ljo (Post 1496366)

1) it usually is.
2) Well, this is what I did yesterday:
which folds but keeps as much of capitalisation as possible.

Originally Posted by mautz (Post 1496620)

Your corpus file is way too small. How many sentences does it include? And on the other hand your dictionary is way too big. Even if it does compile OkBoard will crash with such a huge dictionary.

My corpora has a filesize about 200MB and contains around 2000000 sentences.
My dictionary has a size of nearly 1MB and contains around 100000 words. I tried a dictionary with 17 million words(size was around 30MB i think) and OKBoard crashed everytime i started it.

Originally Posted by mautz (Post 1496652)

I would start with a smaller dictionary, compile it and see if it works :-)

Originally Posted by spidernik84 (Post 1496642)

I'll try with the corpus sent by ferlanero, it's probably bigger.

Originally Posted by ferlanero (Post 1496685)

Did you recibe it? Could you download it?

Originally Posted by itdoesntmatt (Post 1497534)

it works great, thanks to ferlanero, spidernick84 and of course, Eber.

PS: i see that it miss a feature. When i swipe a word it would be better to put automatically a space...right now i have to tap space button between a word and the next...and is not that comfortable to swipe-tap-swipe

Originally Posted by itdoesntmatt (Post 1497537)

now issue seems to be risolved. i dont know why, maybe a bug, but now it works just fine.

edit: the problem is still there but ,as long as i can see, only in android application , not in native apps.
Seems like the autospacing doesnt work in dalvik

Originally Posted by ferlanero (Post 1497362)

Hi guys! I was checking the performance of the Spanish language for OKBoard and I have noticed just an issue: the words with letter "ñ"

In Spanish there is a lot of words containing the letter "ñ". After applying the "clean_corpus.py" script, all the words with that letter, remaining intact, but after doing "db/build.sh es" and finish the process correctly, the keyboard doesn't recognise the words with the "ñ" inside. I want to ask: Is there any solution for this? How can I do in order to OKBoard recognises words like "España"?

Thanks in advance!

Originally Posted by mautz (Post 1497617)

You can add words manually by editing

/home/nemo/.local/share/okboard/nl-user.txt

First word is the word you want OKBoard o show the second word is the one you swype, the numbers are always the same.

Originally Posted by rob_kouw (Post 1497627)

Could it be that this is the problem? The word "ik" is double in there.

Originally Posted by spidernik84 (Post 1497873)

@ferlanero: what os did you use to compile the files?

Originally Posted by ferlanero (Post 1497875)

ArchLinux

Originally Posted by ferlanero (Post 1497875)

I can't write words with letter "ñ" like España :(

Originally Posted by eber42 (Post 1498085)

Make sure you use the real "N" key and not "Ñ" (this is a known limitation that has been discussed earlier, and I have some ideas on how to fix it).

If it still does not work, send me your language files and logs (you have to enable them in the application and they can be found in ~/.local/share/okboard). And do not expect a quick answer (I did not even start to work on the weeks old transparency issue).


BTW, last week I worked on the engine that processes swipes. Now it is not that much more accurate, but it does no more assign bad scores to expected results. Next week I will start to re-implement the prediction engine (this is the part which handle language model and learning. At the moment it is completely broken) and after I will go back to regular bug fixing.

Originally Posted by ferlanero (Post 1494006)

The steps to do that are these:

-You need a linux environment (I'm using Archlinux, but Ubuntu or some other works too)

- You need to download the tarball first: http://git.tuxfamily.org/okboard/okb...master.tar.bz2 and uncompress it at your /home directory

- You need the dictionaries. I take it from https://github.com/titoBouzout/Dictionaries but it needs to be adjusted, so I attach the file already processed (see Spanish.dic.txt.zip on this post)

-You need the corpora files of your language (e.g. Spanish)
http://corpora2.informatik.uni-leipzig.de/download.html
http://www.cs.upc.edu/~nlp/wikicorpus/
http://opus.lingfil.uu.se/OpenSubtitles2016.php
http://www.lllf.uam.es/ESP/Corlec.html
https://tatoeba.org/spa/downloads

- You need the "aspell-es" package (in case of Spanish) instaled from the repos of your distro.

- You need "lbzip2" package installed in your system too.

-You need "rsync" installed in your system.

-You need "QT5" installed in your system.

- Now you need to create a folder somewhere and put the dictionary inside (e.g. /home/username/okboard/langs)

-If you have several corpora files, then:

- Open a terminal window

- And set the two environment variables:

- You can see those variables with

if you're curious

- You need to compress the file (Spanish.dic.txt) you put before in /home/username/okboard/langs:

- Now should be named corpus-$LANG.txt.bz2 In our case: corpus-es.txt.bz2 because of Spanish

- There should be a single file inside.

- The next thing is to do is to move in okboard files inside the same Terminal window in our case "/home/username/okb-engine-master/". Here is the okboard's source code.

- In 'db' folder you must create a lang-es.cf file first. You can copy it from another .cf file in the same folder (e.g. copy lang-en.cf and rename it into lang-es.cf)

-And left only ASCII characteres on those files:

- Execute ("es" in case of Spanish)

- After this, the script create the dictionaries for OKBoard with next list of files:

add-words-fr.txt
es-predict.dict
lang-fr.cf
clusters-es.log
es-test.txt.bz2
lang-nl.cf
clusters-es.txt
es.tre
predict-es.db
corpus-es.txt.bz2
grams-es-full.csv.bz2
predict-es.ng
db.version
grams-es-learn.csv.bz2
predict-es.rpt.bz2
es-full.dict
grams-es-test.csv.bz2
predict-es.txt.bz2
es-full.tre
lang-en.cf
words-es.txt
es-learn.txt.bz2
lang-es.cf

- So, now we have the Spanish dictionary created.

After this. I don't know what to do with these files. So any help is welcome :)

Originally Posted by uggeli (Post 1502166)

5) I think Finnish spellchecking doesn't use aspell, but Malaga based Voikko: http://voikko.puimula.org/ and if I'm not misunderstood, voikko is used by ispell for example. But how to get that finnish dictionary file is somehow unclear to me.