Most of the summarizers available today rely on simple extraction of significant text fragments to produce summaries. They can be classified into two general categories: domain dependent approaches use knowledge of the specific domain and text structure (financial, medical, etc.) to achieve high quality summaries. On the other hand, domain independent approaches employ various statistical and linguistic techniques to identify key sentences of the document. The problem with text extractors is that they often produce inconcise, incoherent, or even misleading summaries. Text abstractors aim to overcome these limitations by parsing the original text and finding new, shorter concepts to describe it. Rather than simply extracting sentences, these systems automatically transform the extracted sentences so they are more concise and coherent. An interesting article by Slate's Karenna Gore clearly describes inner workings of a summarizer built into Word 97: First, AutoSummarize identifies the most common words in the document (barring "a" and "the" and the like) and assigns a "score" to each word--the more frequently a word is used, the higher the score. Then, it "averages" each sentence by adding the scores of its words and dividing the sum by the number of words in the sentence--the higher the average, the higher the rank of the sentence. Of course, more specialized products (especially text abstractors) employ much more advanced lexical and statistical algorithms. Each product listed bellow is usually accompanied with articles describing technical details in more detail.

Extractor is a software for automatically summarizing text, developed by the Interactive Information Group of National Research Council of Canada (NRC). Extractor takes a text file as input and generates a list of key words and a list of key sentences as output. Applications include text summarization, generation of a "back-of-the-book" index (an index for people to read), generation of a machine-readable index (for search engines to read), and document preprocessing for further automated text analysis. It is available as an add-in for Internet Explorer or stand-alone executable. Core engine is available as a DLL, and is used by several commercial products listed at their Web site. One of the most popular packages using this engine is Copernic Summarizer ($69.95). It summarizes English, French, German and Spanish texts in any field of interest, from Web pages, Word documents, e-mail messages or many other formats. In addition, proprietary WebEssence technology is used to remove irrelevant text - navigation, banner ads, etc.- from Web pages and focus only on the essentials.

Megaputer Intelligence offers quite a few interesting and innovative products for data, text and Web mining. Their ML Annonator ($99) and TextAnalyst for IE 2.0 ($79) are tools for semantic analysis, summarization, and natural language querying of encountered Web pages (only English language is supported at this time). Their approach use semantic networks, special stochastic models and dictionaries to determine the significance and semantic independence of source text sentences for processing the source text into a summary of a given length. Another interesting product is TextAnalyst COM, a software development kit implementing a set of functions for automated semantic analysis of English texts. Developers will also appreciate Inxight's Summarizer SDK, another tool that can be "trained" to find key sentences based on the structure of specific document types. It currently supports a large number of foreign languages, which may be very useful if you are developing multilingual solutions.

Mac users will like Data Hammer ($24.95), "the world's first summarizing Web browser". Whenever Data Hammer encounters a page that appears to be an article, it automatically flips into summary mode. All unnecessary navigational links are stripped away, leaving only a short abstract. Then, you can simply use Data Hammer's slider to increase or decrease the page's level of detail. This product employs Microword Tree Trimming (MTT) - high performance, small footprint engine written in standard C++. MTT is not based on or at all dependent on English. MTTs design is such that summarization is actually executed independently from the operating language, and it is estimated that an engineer could port it to the previously "unknown" language in less than a week.

infoGIST Suite ($249.95) is a set of applications for finding and using on-target online information. It contains several modules - Search Director, Viewer, Watcher, and Router, that can automatically provide relevant search results on the Internet, intranets or stand-alone PCs. The summarizing is performed in Viewer, an integrated document reading and analysis environment, providing speed reading previews, key point highlighting, fuzzy match searching, and a variety of analysis reports on text, Microsoft Word, WordPerfect, and HTML documents.

Text summarizers are often available as a key component in an integrated text mining or knowledge management solution. This approach is taken by IBM (Intelligent Miner for Text), Autonomy and Verity. However, these tools require significant investments in both time and money, and will be attractive for developers or end users that require powerful and scalable end results.