Particular Case of RAR Password Cracking

[TmC]

Hi there,
I'm here to ask some advice for a particular case of RAR Password Cracking.

I'm using ElcomSoft ARCHPR 4.54 for this purpose but the inner workings of the program are slowing down the process, or, at least, I don't know it it can be set properly for my case.

The situation is the following:

1. I have many RAR archives, whose files are protected with a single password. (only the files are protected, the archive can be opened and I can see the contents, only extraction needs password)

2. I know the "philosophy" which was used to generate the passwords. They are all Latin Words, only the first letter is capital, no numbers, nothing else. 50% of the files contain a file which name contains this password. Another 25% of the files has a password which is not present in the files name but it is a common latin word. All the passwords are 2 to 8 character maximum.

3. At the moment, the first 50% of the files have been cracked manually and the second 25% of the passwords have been cracked with the auxilium of ARCHPR and the Dictionary attack with the whole list of Latin words.

The problem is the following:

A Latin word (actually it is not Latin but a language derived from Latin which uses conjugation), can have many forms and the dictionary attack only has the "base" form. (ex. base form: habere conjugted: habeo, habes, habet, habebus, habetis, habent)

Since the "logic" behind the passwords is always the same, I suspect that the 25% of the non cracked passwords do use conjugated words that, of course, ARCHPR is unaware of since they are not present in the dictionary supplied.
There is also a remote possibility that it might use proper people's name, like Jesus, Mary, Joseph and so on, but this is not an issue because with a bruteforce attack the meaning of a word is useless

I am trying to figure out how to instruct ARCHPR to perform a brute-force attack based on these premises.

The problem is that, with ARCHPR, if I specify A->Z and a->z, the software tries every possible combination. This arises 2 problem:

1. Out of feasability of cracking time.
2. A lot of wasted time, since only the first letter is capital, and the others are not, so we have that only the first letter is A->Z + a->z and the subsequent others are only a->z.

So far, I have been unable to figure out a solution using ARCHPR...

Do someone of you have a clue on how to do this thing, or can suggest other, more configurable software, to achieve this goal?

Thankyou.