Consider these modifications to tol33t/2:
tol33t([], []).
tol33t([Code|Codes], Remainder) :-
translate([Code], Translation), !,
tol33t(Codes, Rest),
append(Translation, Rest, Remainder).
tol33t([Code|Codes], [Code|Remainder]) :-
tol33t(Codes, Remainder).
The first clause is the base case.
The second clause will succeed iff there is a translation for the current Code via translate/2, as a list of characters of arbitrary length (Translation - note you had [Lower] instead, which restricted results to lists of length 1 only). The cut (!) after the check for a code translation commits to finding the Rest of the solution recursively and then appends the Translation to the front, as the Remainder to return.
The third clause is executed iff there was no translation for the current Code (i.e., the call to translate/2) in the second clause. In this case, no translation for the Code means we just return it as is and compute the rest.
EDIT:
If you don't have cut (!), the second and third clauses can be combined to become:
tol33t([Code|Codes], Remainder) :-
tol33t(Codes, Rest),
(translate([Code], Translation) ->
append(Translation, Rest, Remainder)
; Remainder = [Code|Rest]
).
This (unoptimized) version checks, at every Code in the character list, if there is a translate/2 that suceeds; if so, the Translation is appended to the Rest, else the Code is passed through unchanged. Note that this has the same semantics as the implementation above, in that solutions are commited to (i.e., simulating a cut !) if the antecedent to -> (translate/2) succeeds. Note that the cut in both implementations is strictly necessary; without it, the program will backtrack to find solutions where Code bindings are not translated where there exists an applicable translate/2 clause.