The Aroma Sensory Space of Chenin Blanc

My husband is continually reading about wines and how to make them. Our vineyard is mainly comprised of Chenin Blanc plantings ad recently, he sent me a link to this:
Modelling the sensory space of varietal wines: Mining of large, unstructured text data and visualisation of style patterns,¹ a freely available research publication from Scientific Reports.
What the authors of this research did was to mine the most comprehensive and data-rich sources of publicly available information on South African (SA) wines, notably John Platter's Wine Guide to South African Wines. Platter's wine guide began in 1978 and contains yearly entries from some 1,300 SA wine cellars and 15,000 individual wines.
This data mining study used a novel application of a data visualisation technique, known as formal concept lattices and mined Platter’s data from more than 2,500 Chenin Blanc wines that were produced over a 7-year period, 2008–2014. Their goal was to model the wine style patterns with a focus on the aroma sensory characteristics of the wine.
Chenin Blanc was entered into Platter’s in three style classes:

• unwooded dry (no noticeable wood sensory character and residual sugar content not exceeding 5 g/L)
• wooded dry (noticeable wood sensory character and residual sugar content not exceeding 5 g/L)
• wooded or unwooded semi-dry (RS content more than 5 g/L and not exceeding 12 g/L)

In this data mining experiment, the authors used Sauvignon Blanc as a control because the sensory characteristics of Sauvignon Blanc are well documented and based on established sensory and statistical methods. The dataset was mined for aroma attributes, all non-sensory words were removed, and duplicates were eliminated. The final dataset consisted of 266 different aroma sensory words for Chenin Blanc and 250 different aroma sensory words for Sauvignon Blanc wines.
The experiment revealed some sensory attributes located more towards the Sauvignon blanc side, and others more towards the Chenin blanc side of the lattice. The central region (labelled as Shared in Fig. 1) consisted of a large number of attributes that were common to both varietals. Attributes appearing further from the centerline or more to the sides of the diagram, are more unique to Sauvignon Blanc or Chenin Blanc.¹

One caveat about the research is that generated hypotheses are the result of algorithmic (and human) pattern-finding in the data, and non-existent patterns may be found.
To validate the results that the authors discovered through the use of formal concept lattices, the authors applied the standardized text data to traditional statistical methods. This standardized text data included terms that were mentioned 50 times or more. The following table shows a list of aroma sensory attributes associated with Sauvignon Blanc and Chenin Blanc.¹

The data showed that complexity was the key in Chenin Blanc aromas. The following descriptors were associated with unwooded dry, wooded dry, and wooded or unwooded semi-dry Chenin Blanc.

Descriptors	% Unwooded Dry	% Wooded Dry	% Wooded/ Unwooded Semi-Dry
Oak, rich, citrus, spice, vanilla, savory, creamy almond	18%	73%	9%
Tropical, fresh, apple, guava, apricot, lemon, honey, mineral, acid, juicy, melon, peach, floral, lime, pineapple, quince	56%	31%	13%

The above table shows the aromas associated with the different styles of Chenin Blanc wine. This can be a guide to the Chenin Blanc wine style that the winemaker is aiming to produce. For us, since we haven't used any oak in making our wines, we can attest to the lemon, quince, mineral, acid, honey, and I would also add passionfruit aromas that our wines exhibit. Another thing that this data mining research shows is the plethora of descriptors that have been used to describe the aroma of Chenin Blanc wines.
References:
1. Carlo C. Valente, Florian F. Bauer, Fritz Venter, Bruce Watson & Hélène H. Nieuwoudt, Modelling the sensory space of varietal wines: Mining of large, unstructured text data and visualisation of style patterns, Scientific Reports, Volume 8, Article number: 4987 (2018).