Having explained a little about how great Rikaichan is, I’d like to introduce an excellent tool for many of the instances where Rikaichan is unusable: StarDict. This open source and cross platform dictionary software functions similarly to Rikaichan, but is not bound to a browser, being stand alone software.
It can read interfaces of Japanese software, and any text therein, if it is displayed as normally encoded text data. Little elaboration on how useful this is is required, I think.
One of the best features is that separate user contributed dictionaries are downloadable, and not only are some of these excellent (superior in many cases to all my Wordtank’s volumes, and the equal of much tedious Google scouring), they are in a standard format and it is possible to edit them easily.
It can also function as a standalone dictionary, but frankly there are probably better free dictionaries available if you are going so far as to manually paste text into it, to say nothing of commercial PC versions of various Japanese dictionaries.
Certain unavoidable limitations are present in StarDict, as well as ones which simply stem from the fact it is originally a Chinese dictionary. The unavoidable limitations are that it cannot handle text not displayed through the OS as actual, readable text data; this means that most games, which are usually using some form of custom internal method of displaying text, do not produce recognisable output – so no usage on eroge and the like is generally possible. Lamentable – I did try this, to make sure you understand. Another offender is the PDF file, which often strips the actual text encoding in favour of displaying mere pictures of the text (so it can display or print on a wide variety of platforms); this means StarDict has nothing to hook onto (as it is obviously incapable of OCR). The same is true of much Flash.
The current major problem with respect to Japanese is that StarDict cannot interpret any Japanese grammar, so in the case of any inflected Japanese word, the most it can recognise is the stem, if that, which often has a different meaning to the actual word. So for long strings of kango, it works very well, but for more Japanese sentences it is unable to cope. Having it parse grammar is probably quite challenging development wise.
A few other minor quibbles are that its interface is rather clunky (often I find the dictionary steals the focus, and I end up typing into StarDict rather than the application being worked with), and I have also seen it being difficult to install. Otherwise, it is a program which is easily recommended for those instances outside a browser where Rikaichan is unusable, or a specialist dictionary is required.