I have a recollection of seeing a tool or script that could analyse a legacy Magento project (custom theme and modules) to check for non-i18n code.
The concept of a regex to find strings that are not double-underscore wrapped __('abc') crosses my mind but seems likely to get a lot of false positives.
Any other suggestions?
I've developed an extension that you can use to collect the texts that should be translated from a magento module (or a set of modules). It is based on the same regex used by the tool that @Fooman talkes about in his answer.
It also let's you identify texts from a locale that are not translated (key = value) but for this false positives may occur.
The extension is free, give it a try. I've used it on my other custom modules and it seams to work.
Magento once had a download in their Enterprise section called 'translation-tools' which gives you a bunch of scripts to create translation csv files. Maybe this is what you are recollecting?
The code uses regexes to find __('abc') but does not perform the reverse (finding strings that are not run through Magento's translations).
I would currently have to agree with the assessment that you would get a lot of false positives when trying to find this programmatically since you are parsing xml, html, php, javascript and potentially some sql for good measure.
