Natural Language Processing for Sensemaking

Making sense of unstructured, large, and diverse content collections requires identifying relevant information.

However, accurately and efficiently determining intended linguistic meaning involves much more than shallow text processing — it requires a deeper understanding of language and context.

But current text mining methods narrowly focus on:
-	isolated key words;
-	static text (instead of dynamic, interactive dialogue);
-	and broad but shallow content coverage.

PARC's approach resolves such issues, enabling efficient, context-sensitive, and meaning-based conversions between human language expressions and machine-interpretable representations.

Features

• Enables deeper processing of content
• Supports multiple levels of meaning extraction — allows analysis of entities and their relationships, to inference beyond the text
• Incorporates PARC security and privacy research ("intelligent redaction") — to develop secure, efficient, semi-automated, and deep content-based ways of protecting privacy
• Allows information to be tailored to user-defined interests

Applications

• Extraction of entities and relationships among them
• Text condensation for informative summarization and notetaking
• Fact-matching within or across documents
• Automated multilingual translations
• Question answering

Demonstrations

To see how content pre-processed with natural language processing technologies can be presented on the front end, please see Information Visualization & Interaction.

A long, ornate sentence of interest is highlighted from a content source.  The window on the right shows an abbreviated form of that sentence: "A dish was carried through the gates."	The window on the left shows an interface that allows the user to identify concepts or terms of interest. The condensed sentence on the right now incorporates those interests: "Domaradsky carried a dish of a culture of plague through the gates."

This work has been partially funded by the NIMD (Novel Intelligence from Massive Data) and AQUAINT (Advanced QUestion and Answering for INTelligence) programs in the U.S. Intelligence Community's/ DARPA's Advanced Research and Development Activity center.