Scaling Multilingual Model Size for Zero-Shot Cross-Lingual Transfer Performance

Multilingual BERT (mBERT), a language model pre-trained on large multilingual corpora, has impressive zero-shot cross-lingual transfer capabilities and performs surprisingly well on zero-shot POS tagging and Named Entity Recognition (NER), as well as on cross-lingual model transfer. At present, the mainstream methods to solve the cross-lingual downstream tasks are always using the last transformer layer's output of mBERT as the representation of linguistic information. In this work, we explore the complementary property of lower layers to the last transformer layer of mBERT. A feature aggregat

Research goal: Does scaling the size of the multilingual model (e.g., from 1B to 175B parameters) improve zero-shot cross-lingual transfer performance more significantly when trained on multilingual intermediate tasks versus English-only tasks, as measured by relative gains in TyDi QA and XNLI accuracy?

Autonomous synthesis report generated by Assignee Research. Tribunal consensus score: 8.5/10.