strip_accents should be None by default in WordPiece

[Xingjian Shi]

Description

@leezu @szha @xinyual I noticed that we may need to set strip_accents to None in

gluon-nlp/src/gluonnlp/data/tokenizers/huggingface.py

Line 564 in 223f1f6

 strip_accents: bool = False, lowercase: bool = False, 

so that it will be turned on when lowercase is True.

This may impact the performance.

Error Message

(Paste the complete error message, including stack trace.)

To Reproduce

(If you developed your own code, please provide a short script that reproduces the error. For existing examples, please provide link.)

Steps to reproduce

(Paste the commands you ran that produced the error.)

What have you tried to solve it?

Environment

We recommend using our script for collecting the diagnositc information. Run the following command and paste the outputs below:

curl --retry 10 -s https://raw.githubusercontent.com/dmlc/gluon-nlp/master/tools/diagnose.py | python

# paste outputs here

[Xingjian Shi]

However, accents may have certain meanings for lots of languages, e.g., mochte vs. möchte. Thus, we may try to turn it off in nlp_process.

[Leonard Lausen]

Thus, we may try to turn it off in nlp_process.

Do you mean exposing an option in nlp_process or changing the defaults in nlp_process? As English is a special case that doesn't care much about accents, I suggest we must keep the option to keep accents in nlp_process.

[Xingjian Shi]

Yes, I mean to always explicitly set “strip_accent” to False in nlp_process. Get Outlook for iOS<https://aka.ms/o0ukef>

…

________________________________ From: Leonard Lausen <[e-mail]> Sent: Monday, February 22, 2021 7:37:12 AM To: dmlc/gluon-nlp <[e-mail]> Cc: Xingjian SHI <[e-mail]>; Author <[e-mail]> Subject: Re: [dmlc/gluon-nlp] strip_accents should be None by default in WordPiece (#1528) Thus, we may try to turn it off in nlp_process. Do you mean exposing an option in nlp_process or changing the defaults in nlp_process? As English is a special case that doesn't care much about accents, I suggest we must keep the option to keep accents in nlp_process. — You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub<#1528 (comment)>, or unsubscribe<https://github.com/notifications/unsubscribe-auth/ABHQH3VYVYGTNOKCD23YH2LTAJ22RANCNFSM4X76IM4Q>.