OpenNMT Forum

Unexpected translations using the pretrained model

Hi,

I was testing the existing pre-trained model with below  input , all looked good except for 

'8.0 as input which came out as .8.0 as output.

Please find the output :
administrator~/OpenNMT$ python translate-ende.py
http://0.0.0.0:7785/translator/translate
Enter the sentence ==> “0.8”
“0.8”
Do you want to continue: y or n ?y
Enter the sentence ==> 8
8
Do you want to continue: y or n ?y
Enter the sentence ==> '8.0
.8.0
Do you want to continue: y or n ?y
Enter the sentence ==> '8.0
.8.0
Do you want to continue: y or n ?

I also used the phrase_table option with the phrase_table.txt file as follows:
‘|||’ . Even this did not work out.

Please assist me to resolve this issue as early as possible.

Thank you,
Kishor.

Hi,

I am using the pre-trained model for my PoC from GER - ENG.  At several whenever there are 

capital letter words we get VOTE as output. If there are lowercase letter word, then the correct output is
repeated several several more times in the output.

administrator@:~/OpenNMT$ python translate-deen.py
http://0.0.0.0:7785/translator/translate
Enter the sentence ==> Erledigt
VOTE
Do you want to continue: y or n ?y
Enter the sentence ==> erledigt
done done done done done done done done done done done done done done done done done done done done done done satisfactorily done done done done done done done done done done done done done done done done done done done done done done done done done done done done done done done done done done done done done done done done done done done done done done done done done done done done done done done done done done done done done done done done done done done done done done done done done done done done done done done done done done done done done done done done done done done done done done done done done done done done done done accomplished accomplished processes accomplished responsibly carried done done carried done carried done carried done carried roughshod care illegally done efficiently skilfully efficiently skilfully efficiently .
Do you want to continue: y or n ?

Please suggest how could I resolve this VOTE and repetition of translated words in teh output?

Thank You,
Kishor.

Hi,

This model was trained on sentences, not words. A model will always have unexpected results on unexpected inputs.

Hi,

How do we then resolve the above issues ?

Could you please give some suggestion on this ?

Regards,
Kishor.

You can train a model on data containing isolated words such as your example.