When tabbing past the language field for captions in the upload wizard, the focus is lost and returns to the start of the page, making it effectively impossible to fill in the form by tabbing through it.
The same happens on the structured data tab for monolingual text statements.
When focusing the language selector, the ULS language selector is opened and focus is transferred to the search field of the selector. As tab unfocus'es from the search field, the focus is not returned to the originating point, but to the start of the document (likely because the ULS language selector is a top floating window at the beginning of the document structure, instead of inline of the document structure where it is being presented.