Grammar-based codes or Grammar-based compression are compression algorithms based on the idea of constructing a context-free grammar (CFG) for the string to be compressed. Examples include universal lossless data compression algorithms.[1] To compress a data sequence <math>x = x_1 \cdots x_n</math>, a grammar-based code transforms <math>x</math> into a context-free grammar <math>G</math>.
The problem of finding a smallest grammar for an input sequence (smallest grammar problem) is known to be NP-hard,[2] so many grammar-transform algorithms are proposed from theoretical and practical viewpoints.
Generally, the produced grammar <math>G</math> is further compressed by statistical encoders like arithmetic coding.
The class of grammar-based codes is very broad. It includes block codes, the multilevel pattern matching (MPM) algorithm,[3] variations of the incremental parsing Lempel-Ziv code,[4] and many other new universal lossless compression algorithms.
Grammar-based codes are universal in the sense that they can achieve asymptotically the entropy rate of any stationary, ergodic source with a finite alphabet.
Practical algorithms
The compression programs of the following are available from external links.
Sequitur[5] is a classical grammar compression algorithm that sequentially translates an input text into a CFG, and then the produced CFG is encoded by an arithmetic coder.
Re-Pair[6] is a greedy algorithm using the strategy of most-frequent-first substitution. The compressive performance is powerful, although the main memory space requirement is very large.
GLZA,[7] which constructs a grammar that may be reducible, i.e., contain repeats, where the entropy-coding cost of "spelling out" the repeats is less than the cost creating and entropy-coding a rule to capture them. (In general, the compression-optimal SLG is not irreducible, and the Smallest Grammar Problem is different from the actual SLG compression problem.)
