psawa / gecko-app Goto Github PK

We should add a readme.md which should address installation and some more general info

Current behaviour: I don't know exactly why, but if you want to correct several sentences, in some cases punctuation will be removed after correction.

Expected behaviour:
The input text should be processed sentence by sentence, and punctuation should remain when needed.

. Hi we are using your code for GEC but we could see below sentence been corrected even though its correct

'Inpatients and Outpatients are allowed ' been modified as 'patients and patients..Could you please look on this

We want that in the output box, token-wise differences with input text be highlighted.

The css classe of the highlighting style already exist, they are called delta-* in the css file.

I believe all we need is either comparing the input and the final output token-wise, or keep track of the changes during the correction process, then adding the css tags where needed..

When a demo is launched, on the longest sentences, the prediction is triggered before the sentence is completely filled in the input box. This causes the output to be corrupted.

Issue occuring on Firefox. Not occuring on: Brave, Chrome, Chromium, Edge.

File involved: https://github.com/psawa/gecko-app/blob/master/application/static/js/demo.js

Delete all the predicted files related to the query in paragraph folder

Improve CSS rules so that the website is nicely displayed on mobiles

Add somewhere that no data is ever stored

Add a demo button with the sentence of the recipes to show all the functionalities.

I noticed that on some computers, the website is displayed very "big". I mean that all the content seems to be very zoomed-in. Among the undesired effects:

• The demo-area is pushed below the visible part of the screen
• The about container is displayed vertically (The team is above the project, instead of having them side-by-side)

Potential track: I develop on Ubuntu, and I noticed that the system is scaled-out in comparison to when on Windows. The website with ubuntu displays well, even if I zoom to 150%.

My screen resolution: 1920:1080

Currently there is no visual indicator that the text has been re-ordered. I don't exactly know under which form it could be indicated. But I think this is very important, to let the user know what transformations have been applied to their text.

Hi,
Thank you for the interesting solution.
Can you please provide an image on https://hub.docker.com ?

https://github.com/shrimai/Topological-Sort-for-Sentence-Ordering

https://paperswithcode.com/task/sentence-ordering

Try to implement a sentence reordering model in addition to the grammar corrector

Currently, if some additional formatting has been copied from another source (Like highlighting, font styling), it will be pasted as is.

When the website appears in google, the description provided is the description of Eduardo's profile. It should show the decription of the application itself (about section)

Understand how to effectively separate DEV environment (offline) and PROD environment (online) when the docker image is online.

Currently, the title displayed in the tab of the browser is GECko+ - Gobbles up your mistakes.
This + - is kinda odd, so I suggest to use a pipe instead of the hyphen.

GECko+ | Gobbles up your mistakes

We have various UX improvement to make, suggested by some students in cognitive science:

• Set key bindings to trigger correction by typing enter, and go back to line by typing shift+enter

• Check the quality of our profile descriptions. (Edu: "at Samsung", Leo: ", and", Thi: "which, I am convinced, can")

• Change the About part adding the discourse feature description and citing the other model we'll implement

During the docker build there was an arror that the "pysdb==0.3.4" is not found.
I got it running by changed the version to "pysdb==0.0.3".

The docker build brocke on:
/bin/sh: 1: cd: can't cd to application/models/gector/data/model_files

Please fix the issues.
Even better, please upload the docker file to hub.docker.com

Not sure what is causing this, but I've noticed two related bugs:

• When selecting a text to paste something over it, the selected text remains and the pasted text is added at the end.
• When selecting a place in the text to paste, the pasted text is added at the end too.

The link is broken.
http://tts.speech.cs.cmu.edu/sentence_order/nips_bert.tar
can't download the nips_bert.tar model while running the following command

mkdir -p application/models/sentence_reorder && cd application/models/sentence_reorder/ && curl -O http://tts.speech.cs.cmu.edu/sentence_order/nips_bert.tar && tar -xf nips_bert.tar && rm nips_bert.tar && mv nips_bert/ model/

In the artcle:
https://arxiv.org/pdf/2105.13318v1.pdf
Authors show improved results by using better synthetic-dataset-for-grammatical-error-correction:
https://github.com/google-research-datasets/C4_200M-synthetic-dataset-for-grammatical-error-correction
Maybe you can use the same approach to improve your results?

Try to re train the czech model to get better accuracies + implement it

Dataset: https://lindat.mff.cuni.cz/repository/xmlui/handle/11234/1-3057

Results of the model for now:

train: accuracy: 0.8452, loss: 0.9536
validation: accuracy: 0.8121, loss: 1.3333

Test set:
Precision : 0.6565
Recall : 0.2397
F_0.5 : 0.4871

I don't remember exactly how it was, but when the app was still working online, output sentences were super weird, i.e. mixed up one with the other.

I suspect that this is due to the fact that there is only 1 tsv file which is overwritten at each query. So if more than 1 user tries to predict, there will be a conflict.

Current behavior: the TreebankWordDetokenizer fails to group back together hyphenated words.
Expected behavior: the output text does not contain any additional spaces compared to the input text.
E.g.: An out-of-the-box feature -> An out - of - the - box feature