Text

Podcast: Follow your Nose at Rijksmuseum

Dutch Digital Heritage Network interviewed Odeuropa team member Marieke van Erp for the Dutch “Paulus en De Nijs op Reis” podcast. For this episode, Kirsten (Paulus), Ronald (De Nijs) and Marieke followed their noses at Rijksmuseum Amsterdam to take in the Odeuropa perspective on the artworks. The podcast can be found Soundcloud or Spotify.

If you understand Dutch, start up the episode and scroll along. If you do not (yet) understand Dutch, find the podcasts highlights below.

Marten Soolmans, Rembrandt van Rijn, 1634 (Gallery of Honour)

Here we focused no Marten’s gloves – the leather tanning process was rather smelly so people used to perfume their gloves. In Odeuropa, we found a 16th century recipe (A verie good parfume for to trimme gloues):

Girolamo Ruscelli, The Secrets of the Reuerende Maister Alexis of Piemount, trans. Wyllyam Warde (London, 1558), 59v–60. Available via:
Open Knowledge Commons and Harvard Medical School
https://archive.org/embed/secretsofreveren00rusc

The Odeuropa team worked with scent makers from IFF to create an interpretation of this recipe, to give for example museum visitors an idea of the type of smells 16th century people used. Museums and other cultural heritage institutions can  also use smells to for example communicate visual objects to visually impaired audiences. In Odeuropa, computer vision algorithms have been trained to detect gloves in paintings as well as natural language processing algorithms to detect references to smells in texts such as the above recipe.

Intermezzo: Kirsten asked two teenagers how they thought the Nightwatch smelled. They thought it would smell musty and of old paint 🙂

Second stop: Cows in a Meadow near a Farm, Paulus Potter, 1653 (on display in room 2.28). In the Odeuropa Explorer, you can find that the computer vision algorithms have recognised several cows and sheep.

The Explorer allows anyone to search and browse through the texts and images that were processed using the Odeuropa text and image processing tools. Paired with the Olfactory Storytelling Toolkit, these Odeuropa tools can aid museum professionals in finding interesting smell objects and references related to artworks and then hands-on tips on how to go about pairing those with scents and how to present them to museum audiences.

Third stop:An extensive seascape with figures by a boat on a shore, Ludolf Bakhuysen, 1667 (on display in room 2.28) Almost next to Potter’s farm landscape, is a seascape, which now conjured up the briny smell of a day at the beach for us.

It is Odeuropa’s aim to increase the knowledge and appreciation for smells and for museum audiences to engage all their senses when interacting with art and cultural objects.

Tracking perceptions shifts in the olfactory domain

(written by Sara Tonelli)

One of the main goals of Odeuropa has been understanding how smell perception has changed over time. Using the system for olfactory information extraction developed within the project, this kind of analysis has now become possible. A first exploration was carried out using a collection of freely available corpora in English, covering a period between 1500 and 2000. Such collection includes London Pulse Medical Records, the Early English Books online and Wikisource, among others.  We selected some items which are particularly relevant to European olfactory history, such as incense, candles, gloves, tobacco and ozone, with the goal to analyse how the description of their smell has changed over time.

We first launched the Odeuropa system for olfactory information extraction on the corpora collection mentioned above, in order to capture all mentions of the items, as well as their association with some smells. An overview of this first analysis is displayed below:

 

For each item, the graph displays the percentage of mentions in our corpus that are labelled also as smell source. In short, a peak in the graph corresponds to a time period in which a term was strongly associated with the olfactory domain. For instance, history scholars showed that leather gloves in the 17th Century used to be scented with perfumes to temper their bad smell coming from compounds used to make leather softer. Thus, they were seen as strong olfactory objects at the time, while nowadays they are not considered ‘smelly’ items. Overall, incense is the item that is most associated with the olfactory domain, in particular around 1860 and 1970, when almost 40% of its mentions are smell-related. The graph for candle(s), instead, displays a growth after 1960, probably related to the widespread use of scented candles. As regards glove(s), the graph shows that it stops being perceived as an olfactory object after 1950, but that nevertheless it was characterised as smell-related only rarely before that date (less than 2% of the mentions). Finally, tobacco and ozone are more ‘modern’ smells, in particular the latter, which was first used to characterise the aroma resulting from experiments with electricity around 1840 and later started being associated with ozone depletion, losing its odorous connotation.

In a second analysis, we aimed at identifying perceptions shifts of different smell sources. Therefore, we created for each item in a given time span a vector embedding containing the PMI (pointwise-mutual information) values of association between such item and a fixed set of olfactory qualities (e.g. fragrant, pungent, sweet). Then, we ran a hierarchical clustering algorithm to identify groups of items that, in a given time span, were described in a similar way. The output is displayed below.

The dendogram shows that the vectors of the same item in different time periods are often far apart and belong to different clusters, as can be observed for gloves, ozone and incense / frankincense. The last two terms, in particular, were considered interchangeable in the past (see yellow and green cluster), but from the beginning of the twentieth century frankincense seems to be used in different contexts (red cluster).

This work shows a novel approach, which combines the power of olfactory information extraction in depicting semantic context and the tradition of semantic change detection to explore the evolution of olfactory language from a diachronic perspective.

A demonstrator of the system for olfactory information extraction can be accessed here: https://smell-extractor.tools.eurecom.fr/

The list of smell sources and the PMI-based vectors used to perform the clustering are available here: https://github.com/dhfbk/scent-change

For a full analysis description, see:

Teresa Paccosi, Stefano Menini, Elisa Leonardelli, Ilaria Barzon, Sara Tonelli. Scent and Sensibility: Perception Shifts in the Olfactory Domain. Proceedings of the 4th International Workshop on Computational Approaches to Historical Language Change 2023 (LChange 23).