Unverified50% confidenceFactExact time
During training, another embedding model was used to select hard negatives, and English and Russian question-passage pairs were mixed in to prevent the model from forgetting these two languages.
1
Sources
50%
Confidence
Long-term
Relevance
11/3/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedDuring training the model is required to pick a single next token, but language itself is ambiguous, so the model learns to make a choice no matter what.76% similarUnverifiedTraditional n-gram models and neural language models trained on a single language assign extremely low probability scores to cross-lingual word sequences, causing the decoder to tend to forcibly map foreign-language words to phonetically similar words in the target language.73% similarUnverifiedThe study conducted a dual evaluation of 8 large language models, having them serve both as question setters generating misleading options and as answer solvers resolving the problems.72% similarUnverified多语言预训练模型如mBERT、XLM-R展现出跨语言零样本迁移能力,即便某些语言训练中从未与其他语言平行语料共同出现72% similarUnverifiedThe study proposes that enhancing a model's ability to distinguish between different languages during pre-training can narrow and in some metrics even eliminate the multilingual gap.71% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1129324API
curl https://kongchang.com/api/v1/knowledge/claims/1129324MCP
get_claim(id=1129324)