Examples

Graphs and other frills are nice, but the main job of a natural language query system is to understand what you mean, and relate it to what's in the database. This is where a query product stands or falls.

We've added so many new comprehension features to this release, we'll have to deal with each one in a hurry.

Again we're featuring the Northwind database, because it's familiar to so many people around the world. But these capabilities are not in any way tailored to the demo. We're also making available some of the other databases we use to test our software (some containing hundreds of test questions) in the Downloadable Help section.

How many orders were placed in each 3 week period from 1/1/92 thru 5/4/94?
You can ask about any period measured in days, weeks, months, quarters or years. We're especially pleased with this one; you can't do this even working directly with Microsoft's graph wizards.

Which orders were shipped on Tuesday?
Which orders will be shipped on Tuesday?
It's not the same Tuesday, and ELF can tell which one you mean!

How many orders were placed 5 years ago today?
How many orders were placed 5 years ago yesterday?
Which orders are scheduled for shipment the day after tomorrow?
We've built in an understanding of many highly sophisticated time concepts. For instance,

Which customers have orders more recent than 6 months ago?
Seems simple enough. Till you try to write the SQL.

Which customers have placed 2 tofu orders since 1991?
ELF interprets this to mean "exactly two".
Which customers haven't placed 2 tofu orders since 1991?
This is understood as "haven't placed two or more".

Show customers that ordered only confections during May 1991.
Placed orders during 5/91, ordered sweets, didn't order anything else. We all know what this is supposed to mean. Now, so does ELF.

During 1994, which customers were handled exclusively by Fedex?
Another tough question; ELF needs to determine when the date part of the condition should be "inner" (customers not handled by Fedex in other years? Wrong!) or as in this case, "outer" (looking only at the year 1994, which customers shipped with Fedex and no one else).

Let's get away from dates!

List products by unit price from highest to lowest.
All manner of new syntax for ordering; sorting "by value", "chronologically", "A to Z", "Z to A", "backwards".

These fancy examples are sort of fun, but the fact is that nine-tenths of the effort of the December release was not in working out these obscure syntactic puzzles. Natural language software's hardest job is understanding the ordinary, everyday words you use, and not making mistakes that a child would sidestep. For example, Microsoft did an excellent job, both in "normalizing" their Northwind database (making the relationships tidy) and in "rationalizing" the data (avoiding ambiguity in the names and data references). But everyone makes mistakes. For instance, why have an entry for Dairy Products, when all the others are just "Produce", "Seafood" etc. It's a tiny thing you'd never notice; but software must be taught how to understand that when you ask for "seafood products" you don't mean to see anything from categories with the word SEAFOOD (like "seafood") or categories with the word PRODUCT (like "dairy products"). We've been able to measure the increased comprehension of our software day-by-day via the thousands of queries posted to our Web demos. The kinds of customization techniques we refer to on nearly every page of this Help Web are an important part of making an English Language Frontend both responsive and accurate.

For more information on example queries, see the Viewing Samples section of the Query topic. The Tutorial page also has a link to the downloadable BOOKS.MDB database, which contains an elfTestSuite table which can be skimmed to give you a quick idea of VB ELF's capabilities.