Mapping Transformer Leveraged Embeddings for Cross-Lingual Document Representation

This research focuses on overcoming the language barriers in recommendation systems by using Transformer Leveraged Document Representations (TLDRs) to represent documents across languages. The team evaluated four multilingual pre-trained transformer models using three mapping methods across 20 language pairs. The results show that mapped TLDRs are effective in creating cross-lingual representations, suggesting a promising direction for expanding beyond language connections.

Publication date: 15 Jan 2024
Project Page: t.m.tashu@rug.nl, ediraul2001@gmail.com, m.sabatelli@rug.nl, m.a.valdenegro.toro@rug.nl
Paper: https://arxiv.org/pdf/2401.06583