Training-Free Text Embedding
A text embedding approach that does not fine-tune model parameters and instead directly uses intermediate hidden states or outputs of a pre-trained language model as sentence representations. Typical practices include taking the last-layer token hidden states, mean pooling, or guidance through prompt templates.
Timeline (last 90 days)
Existing training-free text embedding methods almost exclusively focus on English, and multilingual scenarios have largely been neglected.
Training-free text embedding refers to methods that, without fine-tuning model parameters, directly use the intermediate hidden states or outputs of pretrained language models (such as LLMs) as sentence representations.
This work is the first to systematically reveal the language-conditioned geometry problem in multilingual training-free text embeddings.
All Facts (3)
Existing training-free text embedding methods almost exclusively focus on English, and multilingual scenarios have largely been neglected.
50%UnverifiedTraining-free text embedding refers to methods that, without fine-tuning model parameters, directly use the intermediate hidden states or outputs of pretrained language models (such as LLMs) as sentence representations.
50%UnverifiedThis work is the first to systematically reveal the language-conditioned geometry problem in multilingual training-free text embeddings.
50%