It usually pays to simply re-read the definition you start with when constructing regexps. In this case:

1.) only a word
If we use the old IBM definition of a "word": A word is a sequence of non-blank characters, separated by blanks. ; we end up with something like:

Code:

grep '[<b><tab>]*[^<b><tab>][^<b><tab>][<b><tab>]*'

We search for an (optional) blank/tab character, followed by one or more non-blanks/non-tabs, followed by an optional tab/blank.

2.) which may start with a lowercase letter or underline,
Ok, we fine-tune our definition of the word:

Code:

grep '[<b><tab>]*[^_a-z][^<b><tab>][<b><tab>]*'

We search for an (optional) blank/tab character, followed by one underline or lowercase letter, more non-blanks/non-tabs, followed by an optional tab/blank.

3.) the following characters may contain numbers and dashes aswell,
more fine-tuning on what we mean by "word" here, i think it is self-explanatory now:

Code:

grep '[<b><tab>]*[^_a-z][-_a-z0-9][-_a-z0-9]*[<b><tab>]*'

4.) and the last character can be a dollar sign.
still more fine-tuning:

Code:

grep '[<b><tab>]*[^_a-z][-_a-z0-9][-_a-z0-9]*\$*[<b><tab>]*'

Probably we could drop the ending "[<b><tab>]*" now, because it might be superfluous - you will have to decide that by running the regexp against your data. Replace "<b>" and "<tab>" with literal blanks/tabs when you enter the code, i just used this to make them visible.

I hope this helps.

bakunin