Archive for January 2014

Anatomy of a database, part 2

The first part of this column, Anatomy of a database, part 1, discussed the first four years of research and development for Peacock Data’s new name database products:

pdNickname 2.0 is an advanced name and nickname file used by businesses and organizations to merge database records.

pdGender 2.0 is a gender coding database built on the same set of names. Users can match the data against the first names on their lists to establish male and female identification.

Both upgrades embrace a host of similar and compatible features including languages of origin and use for each name as well as fuzzy logic so information can be recognized even when lists have typographical errors or uncommon spellings. They were built during the same development cycle because both are extracted from the same master file.

To recap, the main product research and development began in early 2009 and was completed by late 2012. Then beta versions of the new products entered field testing in January 2013.

According to the company’s chief development coordinator Barbara Adair, “By 2013 early planning for version 3.0 of the products was already underway and included new fuzzy logic technology designed to work with typographical errors and uncommon spellings. Then development proceeded so well that in April 2013 the new technology was moved up to the version 2.0 cycle.”

Barbara pointed out, “The most complex fuzzy logic involves predicting likely misspellings or alterations. We look at numerous factors that may occur in the spelling of a name. Common examples are frequently reversed digraphs (a pair of letters used to make one phoneme or distinct sound), phonetic transcriptions, double letters typed as single letters, non-common characters, the number of letters in a name, where elements occur in a name, and hundreds of other possible factors.”

“A lot of research and field trials have gone into creating the fuzzy logic algorithms and their inclusion in our new products will substantially increase their power for users,” she added.

“The difference between a real name and a fuzzy version can be very slight and even difficult to notice at first glance,” Barbara said. “But they are different and can make a big difference in the success rate for businesses and organizations working with lists of names.”

Barbara notes, “A sizable majority of the Pro edition of both new products is built with fuzzy logic, but users not ready to dive into the new technology can purchase a Standard edition without fuzzy logic and easily add it later when they are ready by contacting the company for an upgrade.”

As for the easiest part of development, Barbara quickly cited the special precision gender coding information in pdGender filtered for languages, rare usage of unisex names by one gender, and other criteria.

By the time we had established the language information in the master file and flagged name types and rare unisex usages, it was actually quite easy to draw out the gender coding fields,” she said. “This is a testament to the quality of the information and how straightforward it is to work with.”

Barbara said, “The new products do have a learning curve but are ultimately very easy to exploit. It may take a few uses, but those working with the data will appreciate more and more how the information is organized and presented. A lot of thought and field testing has gone into this.”

One result of the decision to build pdNickname and pdGender from the same master file is the strong compatibility between the two offerings.

“While pdNickname and pdGender can easily be used separately, when used jointly they make excellent partners,” Barbara said. “They are comprised of the same set of names and can be linked together with little effort.”

On November 1, 2013, Peacock Data demonstrated the products in front of participants gathered in their Chatsworth, California offices. By this time the new releases were almost ready to go and the development team working under Barbara began tweaking the final layouts and authoring the product documentation.

pdNickname 2.0 Pro and pdGender 2.0 Pro were released on Monday, December 30, 2013 and the Standard editions (without fuzzy logic) made their debut two weeks later.

pdNickname 2.0 Pro has 3.9 million records, including 2.61 million with fuzzy logic, and is 2.9 GB counting all formats and files. pdNickname 2.0 Standard has 1.28 million records, does not have fuzzy logic, and is 964 MG.

pdGender 2.0 Pro has 140,000 records, including 80,000 with fuzzy logic, and is 80.6 MB. pdGender 2.0 Standard has 60,000 records, does not have fuzzy logic, and is 25.5 MB.

Product information

Anatomy of a database, part 1

According to Peacock Data, plans for two just released product upgrades were initially written up in January 2009 and development began in earnest mid-summer of that same year. The products were built during the same development cycle because both are extracted from the same master file.

One of the new products is pdNickname 2.0, an advanced name and nickname file used by businesses and organizations to merge database records. They can match the data against their lists to determine if two or more records are the same individual. It identifies first names that are the same even when they are not an exact match, but rather equivalent, such as a variation or nickname.

The other new package is pdGender 2.0, a gender coding database built on the same set of names. Users can match the data against the first names on their lists to establish male and female identification.

Both upgrades embrace a host of similar and compatible features including languages of origin and use for each name as well as fuzzy logic so information can be recognized even when lists have typographical errors or uncommon spellings.

According to the company’s chief development coordinator Barbara Adair, “Creation of the master name file these new products result from is the biggest venture our company has ever undertaken. There are thousands of sources for names in scores of languages, and our task was to compare and contrast all this data and create the ultimate first name resource.”

Information drawn from the sources includes variant spellings, relationships with other names, and the languages and gender associated with each name.

Barbara pointed out, “The language features have never been available before on this scale and required a sizable portion of the nearly five years of research and development.”

“From the start it was essential to identify the languages associated which each name in considerable detail,” she added. “This gives users previously unavailable ethnic demographics linked to the names already on their lists.”

Barbara showed some of the documents used in construction of the new offerings including a manuscript from 731 AD, written by a monk named Bebe, listing the earliest English names dating from the Anglo-Saxon era of the Early Middle Ages. The still common personal name “Hilda” is an example from the manuscript.

Because sources often give diverse information and use different spelling conventions, it was crucial not only to gather all the information possible but also to differentiate between the quality of sources,” Barbra explained. “Better information became easier to identify after working with the sources over the course of the first year.”

About half the database records are English and Spanish names, and international names originating and used in over 200 other languages make up the second half. This includes such languages as French, German, Chinese, Japanese, Vietnamese, Korean, Hindustani, Arabic, Persian, and Yiddish as well as Native American names and ancient Greek, Latin, and Hebrew names.

According to Barbara, “Special attention is paid to rare usages of unisex names like Kimberly, Hillary, Valentine, and even Maria. Names like these, while usually associated with one gender, are also occasionally employed by both genders. The new products identify rare usages so they can be considered separately. pdGender in particular employs this technology out-of-the-box allowing users to ignore rare unisex usages when assigning gender.”

“Beyond just identifying the languages of use, we also classify name origins, such as Old English opposed to Middle English opposed to modern English,” Barbara noted. “This adds value for those researching personal names or the relationships between languages, such as in the fields of anthroponymy, onomatology, ethnology, and linguistics.”

According to the product documentation, both packages identify five basic first name types:

  • Base Names
  • Variations
  • Short Form Nicknames
  • Diminutives
  • Opposite Gender Forms

Assigning a type identification to each name was a lengthy part of development, but it is significant because the added information permits more precise filtering and ultimately better results,” Barbara said. “Base names are characteristically the oldest because they are the original names all later formations can be traced back to. A lot of time was devoted to these. It is important they are identified as accurately as possible because the remainder of the database is dependent on them.”

Most of the main product development was completed by the end of 2012 and field testing of beta editions commenced in January 2013.

See Anatomy of a database, part 2 for the rest of the story.

Product information