Show HN: Gramara – English grammar correction via transformers

  • Been working on this for the past year. This is a grammar correction tool that uses modern (transformer-based) NLP techniques for fluency edits. Basically it helps you write more fluently if you're not a native English speaker.

    A lot of my family members are English-learners, and I found that they struggled to use Grammarly and similar tools - 1. These tools assume that your input is mostly correct, whereas for English-learners the entire sentence needs to be rewritten/re-interpreted 2. They don't parse your text for semantics, and often come up with nonsensical recommendations if your English is particularly broken 3. English speakers don't actually write in a completely grammatical way, but often in terse sentence fragments. To be more "fluent", some grammatical errors can actually help.

    I built this tool to address this specific niche. In the backend there is a vanilla seq2seq transformer model, trained on a small corpus of parallel data combined with data augmentation and backtranslation. It basically treats grammar correction as a NMT problem, although I don't think this is a new approach.

    here's an example of a sentence that requires fluency translation, which is not handled well by grammarly etc:

    "Because we are a few numbers of people at our store, it has been very inconvenienced for those who have been difficult to connect to the phone."

    The app has a toggle for English level, which switches the model depending on your English proficiency. For the higher level the app acts more as a paraphrase tool.

    some other real life examples:

    The link shared is not working. Please do the needful at the earliest.

    Please share me a new link. I used the last link you forward to me.