We address the problem of adaptive compression of natural language text, considering the case where the receiver is much less powerful than the sender, as in mobile applications. Our techniques achieve compression ratios around 32 % and require very little effort from the receiver. Furthermore, the receiver is not only lighter, but it can also search the compressed text with less work than the necessary to uncompress it. This is a novelty in two senses: it breaks the usual compressor/decompressor symmetry typical of adaptive schemes, and it contradicts the long-standing assumption that only semistatic codes could be searched more efficiently than the uncompressed text. Our novel compression methods are in several aspects preferable over the existing adaptive and semistatic compressors for natural language texts.

Abstract. In the last years, new semistatic word-based byte-oriented compressors, such as Plain and Tagged Huffman and the Dense Codes, have been used to improve the efficiency of text retrieval systems, while reducing the compressed collections to 30–35 % of their original size. In this paper, we present a new semistatic compressor, called Pair-Based End-Tagged Dense Code (PETDC). PETDC compresses English texts to 27–28%, overcoming the optimal 0-order prefix-free semistatic compressor (Plain Huffman) in more than 3 percentage points. Moreover, PETDC permits also random decompression, and direct searches using fast Boyer-Moore algorithms. PETDC builds a vocabulary with both words and pairs of words. The basic idea in which PETDC is based is that, since each symbol in the vocabulary is given a codeword, compression is improved by replacing two words of the source text by a unique codeword. 1

We describe a compression model for semistructured documents, called Structural Contexts Model (SCM), which takes advantage of the context information usually implicit in the structure of the text. The idea is to use a separate model to compress the text that lies inside each different structure type (e.g., different XML tag). The intuition behind SCM is that the distribution of all the texts that belong to a given structure type should be similar, and different from that of other structure types. We mainly focus on semistatic models, and test our idea using a word-based Huffman method. This is the standard for compressing large natural language text databases, because random access, partial decompression, and direct search of the compressed collection is possible. This variant, dubbed SCMHuff, retains those features and improves Huffman’s compression ratios. We consider the possibility that storing separate models may not pay off if the distribution of different structure types is not different enough, and present a heuristic to merge models with the aim of minimizing the total size of the compressed database. This gives an additional improvement over the plain technique. The comparison against existing prototypes shows that, among the methods that permit random access to the collection, SCMHuff achieves the best compression ratios, 2–4 % better than the closest alternative. From a purely compression-aimed perspective, we combine SCM with PPM modeling. A separate PPM model is used to compress the text that lies inside each different structure type. The result, SCMPPM, does not permit random access nor direct search in the compressed text, but it gives 2–5 % better compression ratios than other techniques for texts longer than 5 megabytes.

Semistatic word-based byte-oriented compressors are known to be attractive alternatives to compress natural language texts. With compression ratios around 30-35%, they allow fast direct searching of compressed text. In this article we reveal that these compressors have even more benefits. We show that most of the state-of-the-art compressors benefit from compressing not the original text, but the compressed representation obtained by a word-based byte-oriented statistical compressor. For example, p7zip with a dense-coding preprocessing achieves even better compression ratios and much faster compression than p7zip alone. We reach compression ratios below 17 % in typical large English texts, which was obtained only by the slow PPM compressors. Furthermore, searches perform much faster if the final compressor operates over word-based compressed text. We show that typical self-indexes also profit from our preprocessing step. They achieve much better space and time performance when indexing is preceded by a compression step. Apart from using the well-known Tagged Huffman code, we present a new suffix-free Dense-Code-based compressor that compresses slightly better. We also show how some self-indexes can handle non-suffix-free codes. As a result, the compressed/indexed text requires around 35 % of the space of the original text and allows indexed searches for both words and phrases.

The inverted index supports efficient full-text searches on natural language text collections. It requires some extra space over the compressed text that can be traded for search speed. It is usually fast for single-word searches, yet phrase searches require more expensive intersections. In this article we introduce a different kind of index. It replaces the text using essentially the same space required by the compressed text alone (compression ratio around 35%). Within this space it supports not only decompression of arbitrary passages, but efficient word and phrase searches. Searches are orders of magnitude faster than those over inverted indexes when looking for phrases, and still faster on single-word searches when little space is available. Our new indexes are particularly fast at counting the occurrences of words or phrases. This is useful for computing relevance of words or phrases. We adapt self-indexes that succeeded in indexing arbitrary strings within compressed space to deal with large alphabets. Natural language texts are then regarded as sequences of words, not characters, to achieve word-based self-indexes. We design an architecture that separates the searchable sequence from its presentation aspects. This permits applying case folding, stemming, removing stopwords, etc. as is usual on inverted indexes.

Positional ranking functions, widely used in web search engines, improve result quality by exploiting the positions of the query terms within documents. However, it is well known that positional indexes demand large amounts of extra space, typically about three times the space of a basic nonpositional index. Textual data, on the other hand, is needed to produce text snippets. In this paper, we study time-space tradeoffs for search engines with positional ranking functions and text snippet generation. We consider both index-based and non-index based alternatives for positional data. We aim to answer the question of whether one should index positional data or not. We show that there is a wide range of practical time-space trade-offs. Moreover, we show that both position and textual data can be stored using about 71 % of the space used by traditional positional indexes, with a minor increase in query time. This yields considerable space savings and outperforms, both in space and time, recent alternatives from the literature. We also propose several efficient compressed text representations for snippet generation, which are able to use about half of the space of current state-of-the-art alternatives with little impact in query processing time.

Word-based compression over natural language text has shown to be a good choice to trade compression ratio and speed, obtaining compression ratios close to 30 % and very fast decompression. Additionally, it permits fast searches over the compressed text using Boyer-Moore type algorithms. Such compressors are based on processing fixed source symbols (words) and assigning them variablebyte-length codewords, thus following a fixed-to-variable approach. We present a new variable-to-variable compressor (v2vdc) that uses words and phrases as the source symbols, which are encoded with a variable-length scheme. The phrases are chosen using the longest common prefix information on the suffix array of the text, so as to favor long and frequent phrases. We obtain compression ratios close to those of p7zip and ppmdi, overcoming bzip2, and 8-10 percentage points less than the equivalent word-based compressor. V2vdc is in addition among the fastest to decompress, and allows efficient direct search of the compressed text, in some cases the fastest to date as well. 1

Several actions are usually performed when document is appended to textual database in information retrieval system. The most frequent actions are compression of the document and cluster analysis of the textual database to improve quality of answers to users’ queries. The information retrieved from the clustering can be very helpful in compression. Word-based compression using information about cluster hierarchy is presented in this paper. Some experimental results are provided at the end of the paper.

Recent research has demonstrated beyond doubts the benefits of compressing natural language texts using word-based statistical semistatic compression. Not only it achieves extremely competitive compression rates, but also direct search on the compressed text can be carried out faster than on the original text; indexing based on inverted lists benefits from compression as well. Such compression methods assign a variable-length codeword to each different text word. Some coding methods (Plain Huffman and Restricted Prefix Byte Codes) do not clearly mark codeword boundaries, and hence cannot be accessed at random positions nor searched with the fastest text search algorithms. Other coding methods (Tagged Huffman, End-Tagged Dense Code, or (s, c)-Dense Code) do mark

Recent publications advocate the use of various variable length codes for which each codeword consists of an integral number of bytes in compression applications using large alphabets. This paper shows that another tradeoff with similar properties can be obtained by Fibonacci codes. These are fixed codeword sets, using binary representations of integers based on Fibonacci numbers of order m ≥ 2. Fibonacci codes have been used before, and this paper extends previous work presenting several novel features. In particular, the compression efficiency is analyzed and compared to that of dense codes, and various table-driven decoding routines are suggested.