Targeting is nothing the brain can't fix

Linguist David Crystal believes strongly in the power of words. Here are some lessons he’d like to share with marketers on semantic targeting.

Prolific author and linguist David Crystal brought his expertise and knowledge of the English language and semantics to the Tuesday morning keynote at iMedia’s Breakthrough Summit, and he quickly caught the undivided attention of those in attendance.

After thoroughly waking up the audience with a few lighthearted jokes, Crystal dove into one of the most pressing and promising areas in all of digital marketing – semantic targeting.

“Increasingly, you guys have been coming to linguistics and asking what we’re up to,” he said. “Semantics is the science of meaning in a language. It’s mainly to do with vocabulary and the way that the languages work.”

In the semantic web, however, you’re dealing with a much broader set of meanings and a general awareness of the world, he explained.

The reason marketers are interested in semantics is that they have a problem, an easily identifiable one at that: poorly placed ads that aren’t only completely irrelevant to the content they run alongside, but also something their clients loathe to see.

The kind of mismatch between content on the page and the ads running above or on the sidelines often turns out to be a very displeasing experience, Crystal said. He cited a recent example about a news report on a stabbing in a Chicago street that was flanked by ads for knife sales. In the same vein, he was reading a news story about Natasha Richardson’s untimely death last week and there was an ad for the film “Happy-Go-Lucky” nearby. 

“We don’t do better these days,” he said. “Every day of the week, you can find examples on both ends of the extreme.”

The biggest culprit of this problem is the pervasive keyword approach, which uses “stupid software” to find the same word a few times and then matches completely irrelevant ads with that content, Crystal said, and they are usually highly ambiguous words.

So how do you guarantee relevance and protection? Semantics.

In his research, Crystal found that the average number of sentences used to describe a word in an English dictionary is 2.8.

“So the likelihood of ambiguity is going to be pretty strong,” he said. Not to mention the fact that there are anywhere from 100,000 to 150,000 words in the average college-level dictionary.

Beginning in 1985, Crystal began researching contextual-based advertising, which looks for keywords and the words that surround it on that page – typically in the same paragraph. If the word “knife” appeared multiple times on a page, and words like “fork” or “spoon” were nearby, then ads for cutlery would probably make sense, he said.

But it isn’t that easy. “For years I thought that was the solution,” he said. “Turns out, all was not well.”

The generally popular belief in contextual advertising began to die away four or five years ago – it’s still used by many companies, but it carries many problems.

“All these approaches are still out there… but a much more powerful approach was needed to solve the problems,” Crystal said. “Most webpages are multi-thematic, and that’s why the contextual approach cannot work.”

That brought him to semantic targeting.

“There are many themes going on in a typical webpage,” he said. “You’ve got to analyze everything on the page, you can’t leave anything out… you’ve got to adopt a broad-based approach. You’ve got to guarantee that you cover all the words that exist in the world.”

After working as the editor for a general encyclopedia at Cambridge, and then working for years to bring that work online, Crystal eventually determined that about 3,000 categories were needed to address the bulk of what digital marketers and advertisers do.

“You want just enough categories to be able to handle the advertising that you’re involved in,” he said. “This is the sort of research that can only be done by human beings. The brain gets it right.”

Computer approaches deliver either too much data or too little data, he added.

“You have to work out what the words are that relate to content on the page,” Crystal said. “This is what semantic targeting means.”

Equally important is the need to keep this process up to date. Marketers should always ask a semantic targeting firm when their data was last revised, and if it’s anything more than a few months, forget it, he said, adding that two to three new words come into the English language every day on average.

Solving the problems brought on by keywords and computer-driven data requires large teams of researchers to physically go through the dictionary looking for all the words that clients would not want associated with their brand. He added that there are about 1,500 words for saying nice things and about 3,000 words for saying bad things.

Summing it up, Crystal concluded: “For the moment, we’re stuck with that old familiar thing: the brain.”