Datasets:

Modalities:
Text
Formats:
parquet
Languages:
Hebrew
ArXiv:
Libraries:
Datasets
Dask
License:

Using vowels in the dataset

#3
by tarasfrompir - opened

There is a desire to make a recognition model in Hebrew with vowels.
This is a question I would like to have an answer to.

Can you explain the request?
I do not understand.

Use this simbols in text part of dataset
https://en.wikipedia.org/wiki/Niqqud

ivrit.ai org

Thank you for your interest in developing a recognition model in Hebrew with vowels. I'd like to provide a two-part response to your query:

  1. Currently, we do not have plans to develop such a model in our project roadmap. This decision is based on our current priorities and resource allocation.

  2. However, I'd like to suggest a potential workaround. There are several automatic vowel predictors available, such as Nakdimon and the Nakdan by Dicta [https://nakdan.dicta.org.il/]. These tools could be valuable in your endeavor. You might consider integrating our corpus with these tools to achieve the desired functionality.

I hope this information is helpful, and I wish you the best in your project.

Thanks for the answer.
I'm currently in the process of developing a model using vowels.
From my experience - Nakdan by Dicta - is practically very difficult to use, their limitations have a very bad effect on the work, they change the entered text, individual letters disappear, etc.
Nakdimon - requires some improvement. But I really like it as a base.
That is why the following question was asked: I don’t know what will happen, but as an experiment it is very interesting to make a model using vowels.

I would say that what you call Nikud would be well-defined for 95% of the words, and 99.9% of the words in a given context.
I am relatively confident you can tag that automatically using something like Nakdimon + some basic GPT (3.5 for example).

Sign up or log in to comment