An embedding model selection decision is being made for a RAG system that will serve multilingual queries (English, Spanish, Mandarin, Arabic) against a document corpus that is 90% English. Users query in their native language and expect documents in English to be retrieved. Which embedding model property is most critical for this use case?

B is correct because the core requirement is cross-lingual retrieval: a Spanish query must retrieve semantically matching English documents. This requires a multilingual embedding model trained on parallel or multilingual corpora such that the embedding space is language-agnostic — semantically equivalent content in different languages maps to nearby regions in the shared vector space. D is wrong because dimensionality affects representational capacity in general but is not the property that enables cross-lingual alignment; a high-dimensional monolingual English model cannot align Spanish queries to English documents regardless of dimensionality. B is wrong because context length affects the chunk size that can be embedded but does not determine whether cross-lingual alignment works; it is a secondary concern after selecting a multilingual model. C is wrong because cosine similarity vs. dot-product is an indexing and scoring choice that can be configured independently of the model; neither metric inherently provides cross-lingual capability, and the choice between them does not determine retrieval language generalization.