Statistical tricks may not change the truth, if the questions do not mean the same issue for your respondents. Why don't you just use the most interesting single questions?

Unfortunately, the Cronbach's alpha is very much influenced by the number of items. With low numbers of item (i.e., < 15 items), a low alpha does not necessarily mean you can throw them out of the window; it just is not a very good measure. The Gutmann's Lambda 2 is a better measure than the alpha and should be used instead.

Thanks both. I'll try the Guttman measure. I could try individual questions too, though they come from a previous study where they worked well as a composite scale. If so I would need to report findings for each item separately, which might be cumbersome to read - or interpret.