The version of Thunderbird I have recently installed (v. 3.1.4) has a number of changes and improvements over previous versions, most notably Firefox-like tab navigation across messages. However, a feature that has been a bit surprising for me is the forgotten attachments function. When you write an email, Thunderbird uses the text on it in order to predict if you have forgotten an attachment, as it can be seen in the next figure (figures in Spanish):
The text of the hint itself gives you the information of why Thundebird thinks you have forgotten the attachment; in the previous case, it is because you have written the word ".doc". The words that fire this prediction are configured at the Options > Redaction > General > Check forgotten attachments:
The words can be actually edited clicking on the Keywords... button:
I find this feature quite pretty (although not too useful, to be honest).
This feature has reminded me a paper that I went across on 2006, during the Third Conference on Email and Anti-Spam. The paper reference is:
Mark Dredze, John Blitzer, Fernando Pereira. "Sorry, I Forgot the Attachment:" Email Attachment Prediction. CEAS, 2006.
The paper used a set of automatically derived n-gram features along with specific features that included e.g. the proximity to important words like "attach" (!). I wonder how much useful and difficult would be to include this learning-based approach into Thunderbird... Another idea for a personal project!
Besides, I promote Thunderbird.