Hello, I have to use my own custom lexicon in my native android myscript application using Interactive Ink SDK.
I have generated a resource file named: "custom_lex.res" from this link: https://developer.myscript.com/support/recognition-assets
and copied the file to the "resource" directory under GetStarted asset folder. I have also added the command: AddResource en_US/custom_lex.res
to the en_US.conf file in "conf" directory.
Even after doing this I couldn't see any improvement in the recognition accuracy. What am I doing wrong or missing?
thank you for contacting us and your questions.
Currently, first let me confirm you created and added your lexicon the proper way to proceed! I am therefore surprised accuracy is not satisfying...
Nevertheless, please keep the following in mind:
-the Text resource (en_US-lk-text.res in your case) is a multi-purpose resource, i.e. it can be used to recognize any common word (e.g. first name, last name, number, standard word), but also unknown words (that we call out of lexicon)
-Attaching a custom lexicon should only be done to recognize "specific" words, that are not part of our lexicons, such as medical terms, custom expressions... Indeed, adding common words is indeed likely to decrease accuracy.
Also, what can be done to ensure your lexicon is working properly is to create a configuration without the text resource, write a single word of your lexicon and ensure it is recognized.
as said above in the topic, the GRM will allow you to "loop" in your lexicon.
By default, the text resource contains a GRM. So, if adding your lexicon to a configuration that is also using the TEXT resource, you will have a GRM.