A multilingual assistant retrieves English passages well but fails on supported non-English customer questions. Which two actions best investigate the gap? Select 2 answers.

1. Evaluate the embedding and retrieval configuration on representative queries in each required language.
2. Evaluate generation quality separately using correctly retrieved context for those languages.
3. Assume all embedding models are equally multilingual.
4. Translate only the user interface labels and skip answer evaluation.
5. Report only the aggregate English-dominated average.