Natural Language-like Queries

Though keyword querying remains standard practice on the Web, savvy users have been typing more detailed queries for years, and Web search engines have greatly improved their ability to handle long queries. Research has shown that people prefer natural expression of queries over keywords, 3, 30 and Web search engine query length continues to increase. According to Experian Hitwise, 22 a global online competitive intelligence service, when comparing queries over a four-week period (August–September 2010) to the same four-week period in 2009, found that searches of from five to eight words were up 10%, while searches of from one to four words were down 2%. The growth of query length suggests a desire to express one’s information needs more thoroughly and may pave the way toward full-sentence queries. Spoken queries are also likely to be full sentences when speech recognition is faster and more accurate.

Longer queries are also being helped by the online use of colloquial language. When most content is technical or scientific (as was characteristic of the early Web), there is less likely an easy-to-find match between a lay user’s words and the words used in the informative documents. Popular question-answering sites (such as Answers.com, Quora, and Yahoo Answers) that store user-generated content bridge colloquial and formal language directly in relevant documents; for example, if a searcher needs a device to connect both a Wii and a DVD player to a TV, but does not know what that device is called, a keyword query could fail. But the query “how do I connect wii and dvd to my tv” turns up a nearly perfect match on a question-answering site, with the solution being a product called either “video selector” or “two-way A/V switcher.” The point is that, though the searcher lacks the vocabulary to look up what is needed, the searcher has the same vocabulary as other people in the same cognitive situation. The combination of text worded colloquially and search engines that do a good job with sentence-length queries helps resolve the vocabulary problem. Considerable work has focused on how

though
observational
studies have
found that people
often search
collaboratively,
tools have only
recently been
developed to
explicitly support
people searching
together.

to search question-answering sites1, 2; ranking algorithms that make use of these mappings will continue to improve results for difficult queries.

Another technical development that may help users who express themselves through long queries is systems that support quasi-natural language interfaces. The new syntax is tolerant of variations, relatively robust, and “exhibit[s] slight touches of natural language flexibility.” 25 These interfaces are seen in Web search engines supporting various wordings for certain kinds of questions that retrieve answers from a database, as in “Istanbul time,” “What is the time in Istanbul?,” and “What time is it? Istanbul.” Blekko allows query modification through a simple slash notation to refine results to predefined categories (such as “istanbul /tech” for search results about technology and “istanbul /people” for results labeled relevant to people).

Miller et al. 23 developed tools for “sloppy commands,” meaning users have a lot of flexibility as to how they express the command, so memorization is not required to make use of them. The “linguistic command line” of Enso (later Ubiquity) 8, 33 experimented with leniency in operating system command lines. The Quicksilver application lookup tool for Apple operating systems supports a hybrid command/GUI interface, using continuous feedback to whittle down the available choices to include what the user has typed so far that still matches available commands.

The Wolfram Alpha search engine provides a range of predefined query types that mix structured forms with some flexibility in word order, along with a knowledge base and computational back-end able to handle certain combinations of these inputs. For instance, the query “ 2 slices of pizza with pepperoni” is decomposed into the base information need (information about pizza) refined by units (slices), the quantity (two), and modifications of the baseline concept (with pepperoni). The result is a table listing calorie and nutrition information. However, the system’s interpretive range is limited; the query “recipe for pizza with pepperoni” returns the same measurement information as “pizza with pep-