It's been a while and the last time I did something like this it was within my own wysiwyg editor so I had full access to all font characteristics needed to calculate which word was clicked by the mouse.
Maybe there's a n easy way to do this if all your apps are .NET or com or share some other framework which provides a way to retrieve this directly.
Via the API, I would look into hooking the keyboard and mouse messages so that your app can pre-process every mouse click on other applications - start with SetWindowsHookEx and read everything you can about hooking messages.
After getting your app to pre-process the messages, you then need to grab the text being clicked. Since text can be painted onto a device context in many different ways, you may be best off doing a screen scrape of the clicked area because the text may only exist as a bitmap. If this is the case, you have to perform some OCR to translate the scraped bitmap back into text. In other cases, the text may reside in the window as text - the WM_GETTEXT message may return this text from some types of windows (e.g. textboxes, buttons, etc.) but for normal windows, this message only return the title in the caption bar.
Sorry I don't have any definite answer, but this may get you started in the right direction.