Preserving European silk heritage through AI; historians and computer scientists are joining forces
Although silk trade is often linked to Asian origins, it was fundamental for the European culture, linked to our history, sociology, economy and art. From furniture and clothing to tapestries and formal attires, silk was present throughout history quite often as a symbol of luxury. Following the European silk road through the centuries will lead us to a better understanding of European history, unravelling the effects of the commercial and cultural exchanges around silk to modern society.
EURECOM’s Prof. Raphaël Troncy, expert in Knowledge Graphs, Semantic Web and Natural Language Understanding, will guide us through his EU research project SILKNOW, with the goal to provide insight about historic silk fabrics and AI technology applied to preserving cultural heritage.
Q. What is the idea behind the SILKNOW project?
RT. SILKNOW started three years ago, based on the idea that European silk heritage is in danger and that we would have to act in order to preserve it. There is a lot of historical information about silk fabric, mostly traded around the 15–17th century between China and Europe, making some cities in Europe quite rich, like Valencia, Lyon and some cities in Italy. Silk is a textile used at the time by high society, mainly for making clothes or furniture for different occasions. Now, historians are trying to understand the role of silk in societies of the past. This research is also expected to be used as an inspiration for today’s fashion, since it is known that fashion is always reinventing itself and very often it is influenced by the past.
Another aspect of the project is to bring back to life the original techniques used for weaving silk through history. Weaving machines used cards with punch holes, in order to create the texture of the textile, while mechanically pressing with your feet a pedal of a loom. However, there is no way to know the pattern of the textile just by looking at a “punchcard”; we would have to use the card with a weaving machine. Keeping in mind that there are thousands of punchcards, we would need to activate a machine for 10 hours, just to be able to see the motif of a textile. Instead, what we propose is to construct a virtual loom that would read these punchcards and simulate in the computer what would be the drawing of the textile.
Q. Who are your partners in SILKNOW and what is everyone’s contribution?
RT. We are a consortium of 9 partners representing cultural heritage, history, ICT and SME industries, all committed to keep silk heritage alive. We harvest a lot of data from plenty of museums, currently 25 museum-partners. All these museums gave us access to their silk collections, with additional valuable metadata, about production time, technique, material, location and several historical observations made by historians about the role of each silk object. Integrating various information from different museums in a common model was a challenge since descriptions were not uniformly provided, depending on the system of each museum. Now, on our side, as computer scientists, one of our contributions is to try and predict missing information on silk objects like production time and place etc., using AI algorithms to predict the missing features, based on similarities with known objects defined by historians. The common utter goal is to create a virtual museum, which would be the sum of several existing ones.
Q. What is the role of EURECOM in this project?
RT. The main contribution on our side at EURECOM, is ADASilk (Advanced Data Analysis for Silk Heritage), an exploratory search engine equipped with a spatiotemporal map in a user-friendly interface, built on top of SILKNOW’s knowledge graph, literally a graph representing all the information we know. It contains around 50K textile entries with images and metadata. With ADASilk tool you can search for textiles by any keyword, but you can also search techniques by other criteria. For example, if you type “Damask”, which is a type of weaving technique, you get around 20K results, with a lot of silk objects coming from different museums, that you can visualise and get more information about them. This exploratory engine, that we have developed at EURECOM, is really the integration point between a lot of technologies.
The project partner UVEG has developed a specific visualisation called spatiotemporal maps, which enables to visualise all textiles on a map, depending on where and when they have been produced. 3D maps are constructed, where every layer corresponds to a specific century and the user can see similarities and differences between silk objects. You can really see trends and evolution of textiles in terms of techniques along the time, which is a powerful tool for historians, using these rich visualisations to extend their study on silk.
Another feature integrated in ADASilk is the Virtual Loom, which was developed again by the project partner UVEG, which is acting as a digital memory of silk heritage, reconstructing the 3D image of a textile the way its weaving was done. On top of that, you are able to simulate the weaving process after changing a few parameters and finally generate a new textile, based on the old one. It is quite fascinating! Don’t hesitate to test this exciting tool.
Q. Could you give us an insight on real life applications of SILKNOW results?
RT. One of the project partners is a small Polish company called “Monkeyfab”, specialised in 3D printing; so, we wanted to print a dress using a 3D printer! The dress is designed by a fashion designer, using some of the textiles we have documented at our museum-partners. Monkeyfab is specialised in 3D printing of large objects, using big machines at the size of a room and the process of printing can take many days. The material used for 3D printing is a special type of plastic that looks like silk, ending up with a silk-like texture. This dress was worn by a real fashion model during a fashion show on February 27th, showing a collection by the Polish designer Patryk Wojciechowski, created with the participation of Paweł Twardo and his Monkeyfab team, using 3D printing technology!
Another aspect is that people will be able to see the evolution of textiles, virtually online on a global scale. We could make an impact on the fashion industry, if we can demonstrate the ability to print a dress for example, using this inspiration from distant past-centuries to influence the fashion of tomorrow.
Q. What are the future directions for SILKNOW?
RT. There are still outstanding challenges to overcome. For example, we are still improving how the machine can better understand domain-specific terms, since for now it is limited to encyclopaedic general background knowledge. Natural Language Processing (NLP) tools can accurately extract information from a text regarding a place or a person. But if you want to teach the machine how to extract more specific information such as the dimension of the object or the type of weaving being used, how much training data you would need to do that? In other words, we provide you with a search engine but you don’t know what queries to type, since you don’t know the domain specific terminology.
So, we would like to have the option to search by image, instead of typing a word that you most likely don’t know. Maybe you see a dress that you like or a piece of textile, so you could take a picture, upload it on the search engine and you will have more information about this. The algorithm will find results that are similar, for which we already have historical information and that we can display. SILKNOW project will finish in September 2021; now we have just completed the very first round of final evaluations and we are eventually beginning the final cycle of development until the end of the project.
Q. What is the message you want to spread through SILKNOW project?
RT. SILKNOW is a project about cultural heritage and I am part of many such projects e.g. ODEUROPA, a new project which is starting now. Those projects are intrinsically multidisciplinary, bringing together computer scientists, historians and other various disciplines. It is true that it takes a lot of time to understand each other, since we don’t share the same vocabulary, or the same timelines. For example, historians take the time to analyse things in detail and they often go more in depth. I really like these collaborations and I get very excited in understanding them and supporting them with useful tools. Our group consists of passionate individuals and also curious enough to open their minds and listen to other people’s perspectives.
For example, we have great collaborators in Lyon who are experts in silk and at the same time they are very open to technology aspects and wanting to understand it. Inside EURECOM, it is also important to recruit people who have this open mindset and I am very fortunate to have such people in my team, either with mixed backgrounds or very curious. For example, a PhD student who is a computer scientist, he is also in love with history, so he likes spending time discussing with historians and this is really an important part of the success.
by Dora Matzakou for EURECOM
Silk Exploratory Search Engine
International Conference on European Silk heritage
SILKNOW YouTube channel
Follow SILKNOW on Instagram!