The Scientists Using Machine Learning to Improve Whiskey Tasting | Wine Enthusiast
Wine bottle illustration Displaying 0 results for
Suggested Searches
Articles & Content

The Scientists Using Machine Learning to Improve Whiskey Tasting

Whether you sip wine or whiskey, how you describe what you’re tasting is inherently subjective. Personal experiences might lead someone from northern Europe to liken a flavor to lingonberries, a fruit unfamiliar to many in the Southern Hemisphere.

How, then, can professional reviewers ensure that their language is accurate and understandable to as many people as possible?

To answer that question, researchers at Virginia Tech’s Department of Food Science and Technology are using Natural Language Processing (NLP), a type of machine learning, to examine thousands of whiskey reviews. The goal is to build a lexicon of the terms used by whiskey reviewers not taken from books, articles or pre-existing tasting notes. Such a list could help consumers understand how experts taste and describe whiskey. It will also provide a more comprehensive view of the flavors that experts perceive.

The project is the dissertation research of Ph.D. student Leah Hamilton.

“We were talking about vocabulary, and how you define the words that people use to talk about flavor,” says Hamilton. She decided to explore one of the thornier challenges in the field of sensory science: how to describe taste.

Taste is highly subjective, and there’s no agreed-upon set of words that people use to describe characteristics like flavor, aroma, appearance and mouthfeel, Hamilton says.

“Taste is culturally constructed: the way we taste food, what tastes good and what doesn’t.”—Jacob Lahne, Ph.D., Virginia Tech

To build their library of flavor descriptors, Hamilton and her team selected reviews from two websites, WhiskyCast and Whisky Advocate. Both of the sites met their criteria of how their NLP algorithm works.

According to Chreston Miller, assistant professor and data and informatics consultant, the program sifts through each review and extracts words that relate to flavor. The data will be analyzed by Hamilton and will eventually form the basis of her Ph.D dissertation.

Hamilton’s project could be groundbreaking not just because it seeks to create a database of naturally occurring flavor descriptors, but also because those words can be problematic.

Hamilton and the dissertation committee chair, Jacob Lahne, Ph.D., say that taste is hard to describe because our perceptions are driven by not just experience, but culture.

“Taste is culturally constructed: the way we taste food, what tastes good and what doesn’t,” says Lahne. “But also, the perceptual dimension seems to be culturally constructed.”

If you ate birthday cakes with sour cherries throughout childhood, you might be adept to discern those flavors. You also might be fonder of sour cherries than someone who has never tasted them.

Plus, how we discern and describe flavors, what Lahne refers to as “flavor responses,” are influenced by the people around us. This is one of the reasons that guided tastings, reviews, and tasting notes can be useful.

“Tasting something with tasting notes, you may start noticing those things, and noticing them in other whiskeys,” says Lahne.

“You need to relate to things you have smelled or had before.”—Jamar Mack, founder, Kentucky’s Original Black Bourbon Enthusiasts

Having an authoritative lexicon of whiskey descriptors might have unintended consequences. It could make whiskey less approachable and inclusive.

When Jamar Mack, founder of the nonprofit Kentucky’s Original Black Bourbon Enthusiasts (KOBBE), leads whiskey seminars, he encourages participants to find familiar flavors in the spirits they encounter. Prescriptive tasting notes can inhibit newcomers, he says.

“It’s going to box people into, ‘Here’s what you should be smelling or tasting based on these words,’ ” says Mack. “You need to relate to things you have smelled or had before.”

A lexicon of whiskey descriptors is the first step, says Hamilton. There remain more nuanced analyses to be made, like how certain words are used to describe pricier whiskeys or how slang comes into play.

These findings could be invaluable to those in the drinks business. Even if specific flavors get associated with more expensive bottles, do drinkers actually taste those notes when they sip? Might they simply anticipate them due to the price tag? And, perhaps most importantly, will they pay more for such subjective, ephemeral pleasure?