Salience Engine is a multi-lingual text analysis engine that is currently integrated into systems for business intelligence, (social) media monitoring, reputation management, automated trading, survey analysis, customer satisfaction, and more. 

We're currently on release 4.4 of Salience, with Salience Five currently scheduled for release in late September/early October 2011.

Salience is built for easy integration into your system - written in C/C++ for performance, we have API wrappers for Java, PHP, .NET/C#, and Python.  Full API documentation is available here.

Salience supports a number of text processing, natural language processing, and text analytics technologies... These include:

• Sentiment Analysis
• Named Entity Extraction
• Theme (Context) Extraction
• Entity-Level Sentiment Analysis
• Summarization

What languages are supported?  English (both British and American) is included with the basic software license.  Non-English support is discussed here.

Salience configuration is extremely sophisticated, allowing for tuning at arbitrarily fine levels.  We have provided a number of different pre-configured data directories for your processing pleasure.  We have one that's optimized for Twitter, and one for longer-form content.   Languages are also treated simply as data directories, integrating very cleanly with the rest of your installation.  We also have a number of lists of famous people, companies, sports figures, etc. already in place.  You have complete flexibility to do a whole bunch of lists and taxonomies of your own, but it's really important to understand that you Do Not Need To.   We've got great statistical models in place for entity extraction.  Why don't you wander over here and read more about it?

Sentiment Analysis is, of course, equally tunable.  You could have a thousand different configurations for tuned sentiment covering lots of different verticals and customers.  That's fine.  Or you can just let it run without any tuning - many of our customers do just that.

Salience is fast, with performance scaling linearly across a multi-core system.  Typical performance with all operations enabled (processing each document for all of that stuff in the above bullet list.)

• Tweets:  50-200/core-second  (so, if you have 2 cores, you'll get 100-400 processed per second)
• Blog/News Articles:  3-5/core-second
• Large Secondary Research Reports (100-200 pages):  3-5 core-minutes/document  (yes, we have several companies processing this sort of content)

To put those performance numbers into context, we're processing the entire Twitter firehose with 16 cores.

Summing up:  Salience is an engine that you integrate into your own application (or we can build one for you) to handle text in deep, fundamentally meaningful ways.