Grammar-based compression is a well-studied technique to construct a context-free grammar (CFG) deriving a given text uniquely. In this work, we propose an online algorithm for grammar-based compression. Our algorithm guarantees O(log² n)- approximation ratio for the minimum grammar size, where n is an input size, and it runs in input linear time and output linear space. In addition, we propose a practical encoding, which transforms a restricted CFG into a more compact representation. Experimental results by comparison with standard compressors demonstrate that our algorithm is especially effective for highly repetitive text.

Practical data structures for the edit-sensitive parsing (ESP) are proposed. Given a string S, its ESP tree is equivalent to a context-free grammar G generating just S, which is represented by a DAG. Using the succinct data structures for trees and permutations, G is decomposed to two LOUDS bit strings and single array in (1+\epsilon)n\log n+4n+o(n) bits for any 0<\epsilon <1 and the number n of variables in G. The time to count occurrences of P in S is in O(\frac{1}{\epsilon}(m\log n+occ_c(\log m\log u)), whereas m = |P|, u = |S|, and occ_c is the number of occurrences of a maximal common subtree in ESPs of P and S. The efficiency of the proposed index is evaluated by the experiments conducted on several benchmarks complying with the other compressed indexes.

We consider the problem of {\em restructuring} compressed texts without explicit decompression. We present algorithms which allow conversions from compressed representations of a string $T$ produced by any grammar-based compression algorithm, to representations produced by several specific compression algorithms including LZ77, LZ78, run length encoding, and some grammar based compression algorithms. These are the first algorithms that achieve running times polynomial in the size of the compressed input and output representations of $T$. Since most of the representations we consider can achieve exponential compression, our algorithms are theoretically faster in the worst case, than any algorithm which first decompresses the string for the conversion.