7.1 Information Extraction: The Task

Simply put, we can characterize information extraction (IE) as follows:

• Take a natural language text from a document source, and extract the essential facts about one or more predefined fact types.

• Represent each fact as a template whose slots are filled on the basis of what is found from the text.

A template is a "form" that conveys a fact when filled. The form has labels, one of which says what type of fact it represents, and the others (the slots) identify the attributes that make up the fact. There are different types of facts. We will be interested here mainly in simple facts and events. Examples of simple facts are:

James Smith works for XYZ Co.

James Smith, Chief Research Scientist of XYZ Co.

Binding of hsp90 to the glucocorticoid receptor requires a specific 7-amino acid sequence at the amino terminus of the hormone-binding domain.

Examples of events are:

XYZ Co. announced the appointment of James Smith as Chief Research Scientist on 4th August 2005.

We hypothesized that retinoic acid receptor (RAR) would activate this gene.

Unless otherwise indicated, we use "fact" to cover both simple facts and events.

A template for the report of the death of a famous person could have slots for the name, the age, the place, the date, and the cause of death. Slots are typically filled by named entities (see Chapter 6), or, in more complex representations, by other facts. In the news...