JHU/APL continued to explore the use of knowledge-light methods for multilingual retrieval during the CLEF 2004 evaluation. We relied on the language-neutral techniques of character n-gram tokenization, pre-translation query expansion, statistical translation using aligned parallel corpora, fusion from disparate retrievals, and reliance on language similarity when resources are scarce. We participated in the monolingual and bilingual evaluations. Our results support the claims that n-gram based retrieval is highly effective; that fusion of multiple retrievals is helpful in bilingual retrieval; and, that reliance on language similarity in lieu of translation can outperform a high performing system using abundant translation resources and a less similar query language.