In principle, you can align anything to anything given some kind of a substitution matrix that tells the program how to score the alignments. Depending on how many unique words are in your dataset, you can still simplify it down to single letters (yellow->Y, red->R, etc). It also depends on whether you are looking for absolute identity or if some color can substitute for each other even when non-identical (say, orange and red or orange and yellow). Again, a substitution matrix would specify if any imperfect matches are permissible.