princeton-nlp/MultilingualAnalysis
Repository for the paper titled: "When is BERT Multilingual? Isolating Crucial Ingredients for Cross-lingual Transfer"
This project helps researchers understand how different properties of language, such as script or word order, impact the performance of multilingual language models. It takes natural language text and applies various transformations to create synthetic language pairs, which are then used to train and evaluate models. The output helps computational linguists and NLP researchers isolate which linguistic factors are most crucial for language models to transfer knowledge across languages.
No commits in the last 6 months.
Use this if you are a computational linguist or NLP researcher seeking to deeply understand the underlying linguistic factors that enable cross-lingual transfer in large language models.
Not ideal if you are looking for a ready-to-use multilingual model for a specific application or a tool to simply improve an existing multilingual system without extensive research into its linguistic properties.
Stars
13
Forks
—
Language
Python
License
MIT
Category
Last pushed
Nov 10, 2021
Commits (30d)
0
Get this data via API
curl "https://pt-edge.onrender.com/api/v1/quality/transformers/princeton-nlp/MultilingualAnalysis"
Open to everyone — 100 requests/day, no key needed. Get a free key for 1,000/day.
Higher-rated alternatives
LoicGrobol/zeldarose
Train transformer-based models.
CPJKU/wechsel
Code for WECHSEL: Effective initialization of subword embeddings for cross-lingual transfer of...
yuanzhoulvpi2017/zero_nlp
中文nlp解决方案(大模型、数据、模型、训练、推理)
minggnim/nlp-models
A repository for training transformer based models
IntelLabs/nlp-architect
A model library for exploring state-of-the-art deep learning topologies and techniques for...