User:Junzay/Blank handling

From Apertium
Revision as of 20:06, 16 August 2016 by Junzay (talk | contribs) (→‎Repositories)
(diff) ← Older revision | Latest revision (diff) | Newer revision → (diff)
Jump to navigation Jump to search

GsoC 2016 project

Code at /

What works currently

The deformatter and the reformatter work. There's still more testing that needs to be done. The fst processor works fine to distribute the tags efficiently and correctly to the words. The pretransfer works fine with testing phase completed. The transfer, interchunk and post-chunk are completed, but still more testing needs to be done. This is how the chain works as of now:


The deformatter links every word with its inline tag before the word

Before deformatter:

<p><i>Hello brother</i> How are you <u>doing</u> Do you see <b>the point</b> I <u>couldn't</u> do it</p>

After deformatter:

[5][{1}]Hello brother[] How are you [{2}]doing[] Do you see [{3}]the point[] I [{4}]couldn't[] do it[6]


lt-proc distributes the tags efficiently to all the words and also handles the inline tags across MWE's

After lt-proc:

[5][{1}]^Hello<ij>$[{1}]^brother<n><sg>$[] ^How<adv><itg>$ 
^be<vbser><pres>$ ^prpers<prn><obj><p2><mf><sp>$ 
[{2}]^do<vblex><ger>$[] ^Do<vbdo><pres>$ 
^prpers<prn><subj><p2><mf><sp>$ [{3}]^see<vblex><pres># the point$[] 
[{4}]^can<vaux><past>+not<adv>$[] ^do<vbdo><pres>$ ^prpers<prn><subj><p3><nt><sg>$[6]


The tags before a word lu involving '#' or '+' are distributed to the other words as well- eg [{4}]^can<vaux><past>$ [{4}]^not<adv>$

After pretransfer:

^How<adv><itg>$ ^be<vbser><pres>$ ^prpers<prn><obj><p2><mf><sp>$ 
[{2}]^do<vblex><ger>$[] ^Do<vbdo><pres>$ ^prpers<prn><subj><p2><mf><sp>$ 
[{3}]^see# the point<vblex><pres>$[] ^prpers<prn><subj><p1><mf><sg>$ [{4}]^can<vaux><past>$ [{4}]^not<adv>$[] ^do<vbdo><pres>$ ^prpers<prn><subj><p3><nt><sg>$[6] 


The inline tags are linked with each word inside the chunk

After transfer:

^adv<adv><itg>{^Cómo<adv><itg>$}$ ^verbcj<SV><vbser><pri><p2><sg>{^ser<vbser><3><4><5>$ ^prpers<prn><subj><p2><mf><sg>$}$ 
^ger<SV><vblex><ger><PD><ND>{[{2}]^hacer<vblex><3>$[]}$  ^prnsubj<SN><tn><p2><mf><sg>{^prpers<prn><2><p2><4><sg>$}$ 
^verbcj<SV><vblex><pri><PD><ND>{[{3}]^coger<vblex><3><4><5># la gracia$[]}$ ^prnsubj<SN><tn><p1><mf><sg>{^prpers<prn><2><p1><4><sg>$}$ 
^mod<SV><vbmod><cni><PD><ND>{[{4}]^poder<vbmod><3><4><5>$[]}$  ^adv<adv><NEG>{[{4}]^no<adv>$[]}$   ^prnsubj<SN><tn><p3><m><sg>{^prpers<prn><2><p3><4><sg>$}$ [6]


In interchunk all the superblanks corresponding to every chunk are output before the reordering of the chunk so as to avoid Superblank Reordering

After interchunk:

[5]^default<default>{[{1}]^Hola<ij>$[]}$  ^Nom<SN><PDET><m><sg>{[{1}]^hermano<n><3><4>$[]}$ 
^adv<adv><itg>{^Cómo<adv><itg>$}$ ^verbcj<SV><vbser><pri><p2><sg>{^ser<vbser><3><4><5>$ ^prpers<prn><subj><p2><mf><sg>$}$ 
^ger<SV><vblex><ger><PD><ND>{[{2}]^hacer<vblex><3>$[]}$    ^verbcj<SV><vblex><pri><p2><sg>{[{3}]^coger<vblex><3><4><5># la gracia$[]}$  
^mod<SV><vbmod><cni><p1><sg>{[{4}]^poder<vbmod><3><4><5>$[]}$  ^adv<adv><NEG>{[{4}]^no<adv>$[]}$  ^prnsubj<SN><tn><p3><m><sg>{^prpers<prn><2><p3><4><sg>$}$ [6]


After postchunk:

[5][{1}]^Hola<ij>$[]   ^El<det><def><m><sg>$ 
[{1}]^hermano<n><m><sg>$ ^Cómo<adv><itg>$ ^ser<vbser><pri><p2><sg>$ 
[{2}]^hacer<vblex><ger>$  [{3}]^coger<vblex><pri><p2><sg># la gracia$     
[{4}]^poder<vbmod><cni><p1><sg>$[]  [{4}]^no<adv>$[]  ^prpers<prn><tn><p3><m><sg>$ [6]


After generator

[5][{1}]Hola[] El [{1}]hermano Cómo eres [{2}]haciendo  [{3}]coges la gracia  [{4}]podría[] [{4}]no[] él [6]


The libtidy module beautifies the input and reformats it to give the output

<i>Hola</i> El 
<i>hermano Cómo eres</i> 
<b>coges la gracia</b> 
<u>no</u> él</p>

Reordering Superblank issue, The problem on this page is dealt with.

On running the example through the chain

<p><i>Perro</i> <b>blanco</b></p>

The output:




See also