2

I'm looking for an algorithm to construct a grammar which, given a set of words which can have multiple identical symbols, represents a compressed version of this set, that is, I can generate only the words of the set but the grammar will take less memory than the set himself.

Besides, I'm looking for an algorithm which can update the grammar when I want to remove a word of the set.

What type of algorithm is able to do that ?

I give a concrete example:

Consider a string S="abcdefghij", and then consider the finite set of words "cdhij", acdef", "fghi", "bcfgij", "defi".

I would like to construct a grammar which generates only this set of words (words which can be viewed as concatenation of various substrings of any length from the original string S).

Finally I would like to remove a word in the set and update subsequently the grammar.

Thank you.

user7060
  • 475
  • 5
  • 12

2 Answers2

2

it is not clear what you mean by "words which have multiple identical symbols". it sounds like you mean certain symbols are interchangeable. in that case, just replace all cases of an interchangeable symbol with a single representative symbol.

as to the more general question, there is an area of CS not-so-widely studied, called grammar based codes which is, roughly, focused on the idea of creating compressions of strings based on grammar expansions. in this way classic compression algorithms such as eg Lempel Ziv can be regarded as special cases of grammar compressions.

your problem is not so well defined unless you specify the type of grammar also, eg say CFG. if the grammar is generated by a recursive machine, then this problem is actually similar or almost identical to finding the Kolmogorov complexity of a string, which is undecidable in general.

vzn
  • 11,162
  • 1
  • 28
  • 52
0

The keyword is "CFG induction": given several examples of words from a CFG, find the smallest/simplest context-free grammar (CFG) that generates those words. If you do a literature search with that phrase, you'll find prior work on this problem. It is a highly non-trivial problem, but you should be able to find some algorithms, e.g., the TBL algorithm. You might also find papers under the phrase "learning a CFG".

Another approach is to use regular languages (e.g., finite-state automata). There are known algorithms for learning regular languages, i.e., given one or more words from a regular language, find a small finite-state machine that accepts those words. See, e.g., the seminal algorithm by Angluin and this survey on the hardness of finding the absolutely smallest possible finite-state machine. For more, look for papers on "DFA learning" and related topics. You might also be interested in MERLIN and this book on learning grammars.

D.W.
  • 167,959
  • 22
  • 232
  • 500