> arrayfun(@(s) s{:}(end-((numel(s{:})-1)>=1):end),arrayfun(@(s) split(s).',S,'uni',0),'uni',0)
ans =
4×1 cell array
{1×2 string}
{1×2 string}
{["W" ]}
{1×2 string}
>>
"what if we want to extract last three words instead of two words."
Actually, there's a much easier way to compute the starting position than above...
nW=3;
arrayfun(@(s) s{:}(max(numel(s{:})-(nW-1),1):numel(s{:})),arrayfun(@(s) split(s).',S,'uni',0),'uni',0)
Set nW first; to pass as argument to arrayfun would have to make it an array since it won't accept anything except an array as an argument...unfortunately, no automagic argument expansion for such purposes.
As noted, the regexp solution is undoubtedly the more better route for stuff like this...altho there's a notable difference in the two results -- this returns a cell array of the words separated out; the regexp result has a single phrase containing the number of words.
An alternate way with string functions would be to find() the last blank in the string before the right number of words and extractAfter. That would also return the substring in one piece.
