2

I wish to find the CFG for a language on two symbols (say a and b) whose words begin and terminate with the same symbol, and have equal quantities of a's and b's. What is the thought process I should use for finding such a grammar? What is the most natural or simplest grammar for this language? I hope you'll explain your answer. Hopefully this will suggest some patterns I should look for when trying to synthesise a grammar for a specified language.

Here's the best solution I could come up with on my own:

$ S \to aTbbTa \mid bTaaTb$

$ T \to abT \mid baT \mid aTb \mid bTa \mid \epsilon$

I think this grammar is correct. Informal argument: I can see that if the word is $awa$ ($w$ being a substring), then there are at least two $b$'s in $w$ that are adjacent. This suggests the form $aTbbTa$ in the first production rule. (The argument holds if the roles of the two symbols are reversed). The second set of production rules is meant to generate every possible word of the language while keeping the number of $a'$s and $b'$s equal. Symmetry suggests that the rule $abT$ should be accompanied by the rule $baT$, and $aTb$ by $bTa$. Initially was wondering if any of the rules in the second set was redundant, but I don't think so - I can think of words that couldn't be formed if any of the second set of production rules was missing. Rather I need to be sure there aren't any words from the language that my grammar can't generate.

[I guess I would need induction to prove my grammar generates every possible word in the language. But right now I'm more interested in the thought process behind coming up with a grammar, and as far as I know, induction (in general) doesn't help much in synthesising a solution/rule/formula/etc.; it principally serves to verify a purported solution.]

D.W.
  • 159,275
  • 20
  • 227
  • 470
A.K.
  • 145
  • 5
  • I've edited the question to try to focus it on a core that avoids the check-my-answer part. I think part of the problem is that you might have been a bit too quick to post this question: because you're not sure whether your solution is correct, you end up with two concerns (is my solution correct? is there a better solution?), which introduces some problems. I think it would be better to first write out a complete proof that your solution is correct, so you can be confident whether it is correct or not -- then that will help you focus on the other part of the question. – D.W. Jan 26 '16 at 22:54
  • As far as the remaining question (about how to find a grammar), this is a common question -- so common that we already have written up detailed reference material on this here: http://cs.stackexchange.com/q/18524/755. Please read through that material and see if it answers your question. Note especially the part about closure properties (e.g., intersection of CFL and regular language). Is there any reason why this shouldn't be closed as a dup of http://cs.stackexchange.com/q/18524/755? – D.W. Jan 26 '16 at 22:56
  • @D.W. Thanks for the effort you put in to make my question more cs.SE worthy! – A.K. Jan 27 '16 at 02:53

1 Answers1

3

Here is the thought process I would use. I would notice that your language $L$ can be written as $L= L_1 \cap L_2$, where $L_1$ is the set of words that begin and end with the same symbol, and $L_2$ is the set of words that have equal quantities of a's and b's.

Then, I would note that $L_1$ is regular and so can be expressed by a simple DFA (with 5 states). Also, I would note that $L_2$ is context-free and can be expressed by a simple CFG: e.g., $S \to \varepsilon \mid aSb \mid bSa \mid SS$.

Finally, I would recall the standard closure property: the intersection of a regular language and a context-free language is context-free. It follows that $L_2$. The proof of this standard closure property includes a construction that shows how to construct a CFG for $L_2$, given a DFA for $L_1$ and a CFG for $L_2$. This gives a CFG for your language. The resulting CFG will have $5^2=25$ non-terminals; some of them are unreachable and can be pruned, but the result is still quite messy.

The resulting CFG isn't the simplest or smallest CFG for your language. Whether it is the most natural is open for debate. But since you asked for the thought process behind coming up with a grammar, this illustrates one general technique for constructing CFG's: separate out the part that requires something more than finite state (equal quantities of a's and b's) from the part that can be handled with a finite-state automaton.

D.W.
  • 159,275
  • 20
  • 227
  • 470
  • Thanks for the reply. In fact I had noticed that my language was an intersection of a regular language and a context-free language, but I hadn't thought about the ramifications of that. Does it imply that this language can be represented by a deterministic pushdown automaton, since (I know that) $L_2$ can be represented by one? Is there a standard construction I can use to design the pushdown automaton that recognises language L? – A.K. Jan 27 '16 at 02:40
  • 1
    (Sorry if this is bog-standard stuff - I've only been reading about computing theory since a few days ago, and about context-free languages and pushdown automata since yesterday, and I confess I've been skimming over sections and flitting between multiple sources. I intend to do a deeper and more thorough study once I've got a feel of the subject.) – A.K. Jan 27 '16 at 02:41
  • @A.K., yes, it does mean that -- but that could be a nice question for this site, if you've done some research, including checking a standard textbook or resource, before asking. You can convert any CFG to a pushdown automaton, which should be enough that you can work out how to design the DPDA. – D.W. Jan 27 '16 at 03:03