We see 3 basic approaches to making an NLP system - a chat bot, a search engine, NER, a classifier... - that works well for one language, like English, work well for more languages:

1. machine-translating at inference (or query) time

2. machine-translating labelled training data (or search indices), and training a multilingual model

3. zero-shot approaches with a multilingual LM like BERT or LASER

When to use which approach?

We at ModelFront have shared some an open guide on multilingual search: https://modelfront.com/search.

In March, Nerses Nersesyan from Polixis and I will give an online workshop on this at Applied Machine Learning Days.

https://appliedmldays.org/events/amld-epfl-2021/workshops/how-to-make-your-nlp-system-multilingual

We'd be happy to have your input and feedback and hope it can help you with your own work, though unfortunately the workshop won't be in time for the ACL/EACL submission deadline.

Thanks and regards,
Adam

--- SoupGate-Win32 v1.05
* Origin: fsxNet Usenet Gateway (21:1/5)

On Saturday, January 16, 2021 at 7:22:59 AM UTC-8, Adam Bittlingmayer wrote:

We see 3 basic approaches to making an NLP system - a chat bot, a search engine, NER, a classifier... - that works well for one language, like English, work well for more languages:

1. machine-translating at inference (or query) time

2. machine-translating labelled training data (or search indices), and training a multilingual model

3. zero-shot approaches with a multilingual LM like BERT or LASER

When to use which approach?

We at ModelFront have shared some an open guide on multilingual search: https://modelfront.com/search.

In March, Nerses Nersesyan from Polixis and I will give an online workshop on this at Applied Machine Learning Days.

https://appliedmldays.org/events/amld-epfl-2021/workshops/how-to-make-your-nlp-system-multilingual

We'd be happy to have your input and feedback and hope it can help you with your own work, though unfortunately the workshop won't be in time for the ACL/EACL submission deadline.

Thanks and regards,
Adam

In the concept-based Ghost Perl AI with Natural Language Understanding --

https://ai.neocities.org/NLU.html --

multilingual NLP is achieved for English and Russian by various design elements such as

https://ai.neocities.org/MindBoot.html -- innate Englsih and Russian basic vocabulary.

--- SoupGate-Win32 v1.05
* Origin: fsxNet Usenet Gateway (21:1/5)