cracking OTP

Suppose two english language messages A and B are encrypted with the same one time pad P: X=A^P and Y=B^P.

Then X^Y=(A^P)^(B^P)=A^B so we can try to guess which words are in A and B and xor it with that to see if we get a word back to chip away at decoding the cyphertext.

The difficulty is that the words overlap, they don't line up easily. For example the messages might be like this:

---- ---------- --- ----- -------- -------- -------
-- ------ ------- ---- ---- --- ------ -------- ---

>>13 not sure...

>>10 actually i still think this might even work...
If we forget about truly random numbers... and just expect a simple prng and two random seeds, then it should reduce to simply having a long enough message to crack ^^ tadaa
for example, two 8-bit lsfr (lfsr?) will repeat every 255 bits, and there will be 255*254 pairs of seeds..
then you just need to match the sequence =D

What's your OTP, /prog/? Mine is Cirno\Agni Cirno. The contest of fire and ice, I just love it!

...well, having a long enough message + resources to compute enough prng streams.. lfsr's are good because each seed is basically just the same sequence with a different offset...
and the attack gets better if the lfsrs aren't of the same size =)