In ciphertext only attack, if I assume that the ciphertext is obtained by an OTP, then for a short length of ciphertext I can search over a random keystream of the same length and discover a plaintext which is meaningful, say in the language of the plaintext. Hence I get a most probable plaintext/ciphertext pair. This can be repeated over other segments of ciphertexts and the corresponding plaintext can be recovered. Hence does this mean ciphertext only attack can always be transformed to known plaintext attack of a certain block length?
Any plaintext input with the correct OTP could result in the ciphertext you have. For example, suppose you have ciphertext:
45 F1 C3 29
The plaintext could be LOVE (or Love or love) with the right OTP. It could also be HATE (or Hate or hate). Or it could be any other four letter word. Or a three letter word with a space or a period. You get the idea.
This is what makes an OTP so powerful from a crypto perspective. The requirements are that the OTP be random and that be as long or longer than the plaintext.