Text

Answered

Recognition Problem

Hello Team,

I have created a Custom Resource File for Text.

File contains :

distance
distance is

distance is 10

islo

islo is

islo is nice

with many other words and sentences.

Many a times, when I write : "distance is 10", engine converts it into : "distance islo"

which i think is not in line with Natural language processing. Can you please check and let me know how can this be improved?


Thanks and Regards,

Vivek Saraswat



Best Answer

Dear Vivek,


thank you for the update.


based on the above, if you want to "recognize"misspelled words, something that may work is that you add the most common misspellings in your lexicon. Also, try toadd only one word per line, adding several words (even if they are usually written together) may indeed have side-effects.


Apart from this, we do not have better ways to proceed.


Best regards,


Olivier


Answer

Dear Vivek,


thank you for the update.


based on the above, if you want to "recognize"misspelled words, something that may work is that you add the most common misspellings in your lexicon. Also, try toadd only one word per line, adding several words (even if they are usually written together) may indeed have side-effects.


Apart from this, we do not have better ways to proceed.


Best regards,


Olivier

Dear Olivier,


I agree with you. I'll try to explain my constraint and hopefully that would allow a stronger clarity. 


Our mission is to educate those who couldn't afford it due to money, location or family upbringing. Those who have passed their teen age but aspire to do something for their village or community. Educate those who couldn't read or write earlier. 


So, our constraint:

1. Write poorly 

2. Improper grammar

3. Spelling mistakes

4. Hindi language words/names to be written in English.

 

 

What we need:

1. Write predefined phrases that we teach them.


So, we build resource file with all expected output that they would write which has actually yield us far better results than using default "TEXT". 

Also, it took us sometime to narrow down this problem and we are very hopeful that you will be able to help us.


Regards,

Vivek Saraswat

Dear Vivek,


thank you for contacting us.


Currently, my question is why don't you just use the default "TEXT" resource. For such use-case, it should work pretty well.


In your particular case, it depends on many parameters, among these the quality of the ink.


Currently, the best way to improve this is to have a "well written" ink and use the default TEXT recognition.


Best regards,


Olivier