A Semantic Carol

From issue 6 of Lab Horizons Digital Edition, Journey with Semantic Spirits to explore the past, present and future of the semantic web.

I hope that as you are reading this, you are feeling full of yuletide cheer and looking forward to the most wonderful time of the year: eating delicious food, celebrating with friends and family, and most importantly, finding out if Smithy is going to accept Nessa’s proposal! (If you know, you know.) 

As may be fairly obvious, unlike the infamous Scrooge, I absolutely adore Christmas. However, for this festive edition, I want to tell you a story about another love of my life: the Semantic Web. Recently, I spent some time catching up with one of my PhD supervisors—the person who taught me all about the Semantic Web. As we reminisced about the good old days and reflected on the past and the present, it got me thinking about what has changed since I started my Semantic Web journey, and what the future holds for this set of technologies. 

Setting the Scene 

Like all good stories, it’s important to set the scene. So, what is the Semantic Web? The Semantic Web, at its core, is essentially a way to bring context and meaning to data such that it can be understood and consumed by a machine. Ultimately, human beings are much better than computers at inherently understanding the context of data, such as words in the English language that are spelled the same but mean different things—like “tears” (crying) and “tears” (ripping something), or “Reading” (the city) and “reading” (a book). People use lots of different terminology to mean the same thing, such as “cat,” “kitty,” “fluffy,” “precious baby,” and much of the time we can just intuitively interpret it as intended. Computers can’t easily do this, which is why technologies such as natural language processing and sentiment analysis are so complex. The Semantic Web provides a set of core standards to facilitate a richer modeling of our data to provide this context and meaning. 

The core data format of the Semantic Web is RDF (Resource Description Framework), which is the linked data format that splits data into triples (subject → predicate → object), where the predicate defines the relationship between the subject and the object. Ontologies (which are basically fancy taxonomies at their core) define real-world concepts, hierarchies, and relationships within a given domain, and these are used alongside RDF to mark up data with the relevant terminology. The primary ontology language, OWL, also enables a level of logic to be added; we can set the domains and ranges of relationships and set different restrictions on them, and then run a reasoner to see what new information can be inferred that wasn’t explicitly defined in the data. A simple example of this would be: if we know that I, Sami, am allergic to juniper, and we know that gin has a botanical of juniper, then we can infer that I am allergic to gin. This is obviously a trivial example but could be very useful in identifying additional knowledge.

 Ultimately, in my humble opinion, this is all incredibly useful. This set of technologies allows us to create rich, interoperable, machine-understandable datasets and define common shared vocabularies that not only lay out the context in a clear fashion but also enable users to definitively know that they are talking about the same thing (or not) using agreed-upon vocabularies. This, in turn, enables users to query data based on concepts rather than text (using the Semantic Query Language SPARQL), and data can also be validated against a set of concepts using SHACL (Shapes and Constraints Language) to ensure that it fits a set of determined requirements. 

So why am I telling you this story? Well, I feel like somewhere along the way the true message of the Semantic Web got lost, and even today, not everyone appreciates its value. So I want to set the record straight and cast away the ghosts of semantics past and present to look to a brighter future. 

The Ghost of Semantics Past 

Let us cast our minds back to 23 years ago, to the beginning of the noughties in 2001, when Blockbuster was still a thing and when smart devices had only been conceptualized in movies. This was the year that Tim Berners-Lee, James Hendler, and Ora Lassila published their famous article about the Semantic Web in the Scientific American magazine. Incidentally, I was 11 years old, and my interaction with the web was predominantly to help me with my homework if a book wouldn’t do the trick. 

Unfortunately, the seminal articles that brought the Semantic Web fame also brought misinterpretation in their wake. Some of the explanations from these early papers (as shown in the images below) led to a misinterpretation that the Semantic Web would be some magical AI system, with quotes such as “agents roaming from page to page can readily carry out sophisticated tasks for users.” Some took this to mean that it would form some intelligent system, even if the full quote from the paper is actually “The Semantic Web will bring structure to the meaningful content of Web pages, creating an environment where software agents roaming from page to page can readily carry out sophisticated tasks for users”—making it clear that the Semantic Web would facilitate these software agents by providing the structure and content. 

This is unfortunate. 

The Ghost of Semantics Present (ish) 

However, if we fast forward to 2008 (that was basically yesterday, right?), when I embarked on my computer science undergraduate degree, I fell under the spell of the Semantic Web and remain under it to this very day. However, I was sad to find that particularly within my own discipline (computer science) I was very much in a minority with this opinion. Now, of course, this doesn’t mean that all computer scientists hate the Semantic Web—far from it—and over the years I have met wonderful computer scientists who work in this area. But in my own personal echo chamber and at the events I attended in this sphere, it was frequently met with negativity or lack of interest. 

However, interestingly, when I made my move into the world of chemistry in 2014, I was pleasantly surprised to find that the Semantic Web received a much warmer reception, and I am delighted to have been involved in a number of Semantic Web projects in the last decade in areas spanning agriculture, social sciences, drug discovery, and the physical sciences. 

Since then, over the last decade, there has been an obvious rise in the number of people talking about and utilizing these technologies. I attend many conferences that have entirely dedicated semantic tracks, looking at creating new ontologies to provide common vocabularies for different datasets. The FAIR Data principles (Findable, Accessible, Interoperable, Reusable) have emphasized the importance of interoperability in data, and above all, there has been a clear push in the metadata community to create semantic metadata records for datasets using the RO-Crate (Research Object Crate) specification. 

The Ghost of Semantics Future 

So what is the moral of our story? Ultimately, from my perspective, I see a bright future for the Semantic Web! But it is imperative that we remember this: 

THE SEMANTIC WEB IS NOT PROMISING AI, BUT THAT DOESN’T MEAN WE DON’T NEED IT! 

This is a tad ironic, as the technologies underpinning it are often referred to as symbolic AI. However, building on the examples in this article, it can—and should—be used to facilitate AI and ML. Everyone talks about making their data “AI ready,” and when I ran the AI 4 Scientific Discovery Network, at every meeting without fail someone would mention that one of their major issues was their datasets being in a malformed, inconsistent state. This is a major issue for humans trying to interpret the data, and also for the machines! Garbage In–Garbage Out is a well-understood concept—that even if you have a beautiful, efficient algorithm, if you feed it horrible data, you won’t get decent results—so it stands to reason that we want to start with the data first! 

And to those of you who might say LLMs and Gen-AI are going to take over and make the Semantic Web obsolete, I say, humbly, that I think you are wrong. I think whatever magical AI system we are currently using will ALWAYS be improved by feeding it structured semantic data over heterogeneous, unstructured, messy data. Case in point: I recently attended the 20th International Conference on Semantic Systems in Amsterdam. Now, I attended this conference back in 2018, when it was the Semantics and Machine Learning Conference. Back then, there were many talks demonstrating how ML systems could be improved by feeding them semantic data. This year, there was an entire track dedicated to LLMs and how they could be improved by semantic data. However hard we try and subvert the need for this rich structured data with AI-esque systems, there will always be an attempt to improve these with structured data, and chances are this will keep demonstrating superior results. 

So there you have it, my twisty, turny tale of Semantic Christmas cheer. I hope you remember your time with the Semantic spirits, and let’s bring a little of the semantic to every day. Merry Christmas, everyone!