Fran Alexander: Democratizing Taxonomy Practice – Episode 202

Podcast: Play in new window | Download

Subscribe: Apple Podcasts | Spotify | Android | TuneIn | RSS

There is always more taxonomy work to be done than there are practitioners to do it. Fran Alexander’s solution to this imbalance is to democratize taxonomy practice.

Fran’s work actually spans the full range of semantic practices, from simple term lists to taxonomies, thesauruses, and ontologies and knowledge graphs. Wherever she’s working in this span of activities, she’s always happy to bring other practitioners along with her.

We talked about:

• the transition in the taxonomy world from building taxonomies to help machines understand humans to building taxonomies to help humans understand machines
• he the rise of AI and LLMs has highlighted the importance of well-structured knowledge and good semantic layers
• the hierarchical progression of knowledge organization from simple lists to full-blown ontologies
• how her efforts to democratize taxonomy and semantic practice are jump-started by humans’ innate organizing schemes
• how the fact that there is always more semantic work to be done than there are taxonomists drives the need to democratize the craft
• how to evaluate the effectiveness of a taxonomy
• fascinating taxonomic edge cases like boundary objects that highlight the artistic aspects of taxonomy science
• how an ontology “is a map of important ideas”
• two main types of taxonomies – descriptive taxonomies and operational taxonomies
• how to assess whether a taxonomy is doing the work you need it to
• typical uses cases for a taxonomy: tagging, indexing, discovery, retrieval, recommendation systems, personalization, etc.
• some advanced taxonomy practices that enterprises could benefit from
• the connections between taxonomies, ontologies, and knowledge graphs
• the Taxonomy Bootcamp London Bite-sized Taxonomy Boot Camp series

Fran’s bio

Fran started her career as a writer and editor of dictionaries and thesauruses in the UK, and, as technology evolved, she specialised in information architecture, search systems, and digital archives, and more recently, the use of semantics in knowledge graphs and LLM applications. Having worked on reference publications including the Collins English Dictionary, and as Taxonomy Manager for the BBC Archive, she now lives in Montreal, Canada, and is the Senior Taxonomist for Expedia Group. She was Taxonomy Bootcamp London’s Taxonomy Practitioner of the Year 2023.

Connect with Fran online

• LinkedIn

Resource mentioned in this interview

• Bite-sized Taxonomy Boot Camp

Video

Here’s the video version of our conversation:

Podcast intro transcript

This is the Content Strategy Insights podcast, episode number 202. As we write our shopping lists and organize our thoughts, we all practice taxonomy and think semantically every day. Some of us get to do these practices professionally. Fran Alexander is a taxonomist based in Montreal. She’s also the reigning Taxonomy Bootcamp Practitioner of the Year. As she’s realized that there will always be more semantic work to be done than there are trained practitioners, she added “democratizing taxonomy” to her to-do list.

Interview transcript

Larry:
Hi, everyone. Welcome to episode number 202 of the Content Strategy Insights podcast. I am really delighted today to welcome to the show Fran Alexander. Fran is a taxonomist based in Montreal, in Canada. She’s also the reigning Taxonomy Boot Camp Practitioner of the Year, so she’s a rock star in the taxonomy world. Welcome, Fran. Tell the folks a little bit more about what you’re up to these days.

Fran:
Well, thanks, Larry. It’s a pleasure to be here. It’s a pleasure to talk to you about taxonomies and democratizing taxonomies. One thing that’s top of mind for me, and it gives a bit of context, I think to the work that I’ve done and the changes that I’ve seen over, well, the last couple of decades really, is that we used to build taxonomies to help machines understand humans. Now we’re building taxonomies to help humans understand machines. We’ve done a lot of foundational work to set up intelligent systems and organize our concepts and our context to create knowledge basis, knowledge graphs and so on. Now AI is coming in and is so elaborate and sophisticated. We, as humans, are having to step back and try and understand and explain and work with the complexities of that world. There’s really been a huge shift and it’s fascinating. It makes it a really great field to be working in.

Larry:
The way you just said that, is business booming these days for the taxonomy folks? Has AI been-

Fran:
Absolutely.

Larry:
Yeah, I would assume so.

Fran:
Absolutely. There’s a lot of excitement obviously over the last few years with the rise of AI and LLMs, but that’s also brought into focus the importance of knowledge and the importance of well-structured knowledge and the importance of really good semantic layers is what people are talking about now to support those AI applications and to support those LLMs. If you’ve got really good, well-described knowledge in a way that machines can process and understand, then that artificial intelligence, those applications that sit over the top are so much more effective and accurate and explainable and so on. I really feel that taxonomy and semantics are going to go in and out of the spotlight, but there’s huge, huge interest at the moment. It really is a boom time for taxonomy and taxonomists.

Larry:
I was at the semantics conference a couple of weeks ago in Amsterdam and there was palpable excitement and even giddiness about the need for semantic help. A lot of it goes to what you were just saying about AI and the need to go and that progression you described of well, from AI and LLMs and then that illustrating the importance for well-structured knowledge. That’s almost like not the, when you look at a maturity model or a, I don’t know what you call it, but that evolution of ways to manage descriptions of categorizations and descriptions of things. Can you talk a little bit about like we talked before, we went on the air a little bit about everything from folksonomies to a full-blown ontology, but there’s lots of stuff in between too.

Fran:
Sure, absolutely. Well, it’s interesting because actually, some of the earliest forms of writing known to history are lists and catalogs. That’s like Hittite tablets and Sumerian King List and so on, and they’re kind of inventories, stock lists. Those were the first things that people, at least as far as we know, those were the first things that people were writing down. That really is the first point of knowledge organization is that you make a list, and we’re all very, very familiar with lists and how useful they are, but a list is your first step. As you get more and more information, once your list starts to get bigger, you need to divide your list up to make it usable and manageable into categories and subcategories. At that point, you’re starting to build a taxonomy.

Fran:
Then you need to go one step further as you get more and more content that you want to understand and manage. You want to move away from a simple monohierarchical taxonomy to start building graphs and have concepts that can belong in more than one subcategory on your list and at that point, you’re starting to build a graph. Taxonomy is obviously focused on the hierarchy, broader and narrow it, to allow you to zoom in and out and find things through that kind of browse mechanism. It’s the fundamentals of taxonomy. Then you start to want to add in synonyms and related concepts, so you start to build up a thesaurus, and thesauruses are where you have all the different words for the same concept.

Fran:
That’s a very fundamental point in semantics, it’s why we need semantics because we have so many different words for the same thing and we need to organize that. Then ontologies come in as you start to define those relationships. How do your lists taxonomies relate to each other? You start to put in some logical rules and your ontology then starts to build up your whole worldview and your perspective. At that point, when you’ve got your ontology and you start to have the actual data that you’re working with, the actual information, you’ve got a knowledge graph. That is something that you then need to support your LLMs and work with your LLMs to get really good results. That’s how you evolve from starting off with lists and tags and labels and build up through to the sophisticated knowledge ecosystems that we’re working with now.

Larry:
As you set out that escalation of organizations schemes, I realize that I’m reminded of something else we talked about earlier, this notion that we’re all already doing a lot of this stuff already. You just happen to do it professionally and have very precise and ISO standards and stuff that you’re working to, but everybody does shopping lists and then goes, “Well, I need to organize it by aisle,” or whatever. Can you talk a little bit about, because the whole theme of this conversation is the democratization and that’s like pre-democratization. People are already poised for this. Assuming we have a slam dunk here in terms of democratizing this.

Fran:
Yeah, absolutely. We are all taxonomists kind of all the time. One example that I like to use is supermarkets, mentioning shopping lists, the way that a supermarket is organized, that’s effectively a taxonomy. You’ve got the dairy section and then within the dairy section you’ve got the milk and you’ve got the different types of milk and cheese and so on. Well, that’s really, that’s a taxonomy. Whenever you organize your folders on your desktop or you go into your email inbox and you start tagging your emails and organizing those into a folder structure, you are doing taxonomy work. In the digital age, we put hashtags on our Instagram posts, all that kind of work is effectively taxonomy work. You may not think of yourself as a taxonomist. I think the word taxonomy often either intrigues people or scares them a little bit, and I’m convinced because it’s got an X in the middle of it. It makes it sound like a very interesting word.

Fran:
Building those lists and categories and subcategories and so on, we do it all the time. As you say, there’s a big difference between what you do in your own personal inbox. For example, take your own personal email inbox, and what we do at scale for knowledge organization for a large enterprise or a large business. If you do, and as an amateur taxonomist, you can certainly start off with your own email, is if you get that wrong, if you make a complete mess of that and the categories don’t make any sense and it’s really hard for you to find things, well, that’s only really going to impact you.

Fran:
Once you’re working professionally for an organization and there’s lots and lots of people involved and the taxonomy is maybe being expressed in the information architecture of a site and that’s how you’re making your money and so on. Or the knowledge assets that you’re organizing, I’m thinking maybe in the legal knowledge management, that can be really important and really critical. Lawyers have to be really sure they’ve got good knowledge and information systems for the work that they do. Doing that professionally, I’d say that’s a lot of what taxonomy professionals are doing are just working at a size and a scale and with knowledge that has a really huge impact and commercial value importance and so on. That’s the difference, but it all starts off the same thing in the same way.

Larry:
As you mentioned that, there’s the intermediate step between that scale you just mentioned, like organizing your desktop or your mailbox. In between that, I see it all the time, I’ve done it a lot, is this notion like you’re working on a product or a project or something like that and you’re go like, “Oh, we have a lot of things here. We need to list them out.” That seems like that’s often a starting point. At some point in that I’ve often got to the point where like, “Ooh, I need professional taxonomy help here.” How do you know when to call Fran, I guess?

Fran:
Well, yeah, so that’s a good question. I would say if you find you’re getting stuck, if things are not working and you’re not sure how to fix them, if you need to do something that has a large scale in it and has a criticality to it and you are justifying investment and expense and so on, you can have huge impacts. Then it is important to think about what you’re doing and to have a plan and to have governance and to have policies and to have processes. This is where we make the difference between just putting a few concepts together in an Excel spreadsheet and actually thinking through, is this taxonomy really working? What’s its function? Is it serving its users? Does it match the mental model? This is one of the questions that I often find myself asking, do the concepts that we are putting together and the structure that we are putting together in a taxonomy match the mental model of the people who are going to be using and engaging with that taxonomy?

Fran:
In terms of democratizing the process, I think it’s a really interesting one for organizations. There’s so much taxonomy work, there’s so much knowledge management work that needs to happen. There’s so much semantics that the more you can democratize good practice across everyone in the organization, the better your knowledge and information will flow, the better organized you’ll be. Taxonomy practice is really good. The theory behind taxonomy work, it’s really good at encouraging clear communication and clear thinking. What makes a good taxonomy, whether it is for your emails in your inbox or for something that’s driving a say, shopping on a website that lots of people are going to interact with, is it’s that clarity. Have you got good clear unambiguous labels? Have you really thought about what this concept is and what it’s doing and what words you should use to express that? Can people look at that and understand it?

Fran:
Then the mutual exclusivity of categories sounds very grand, but actually, what we’re doing when we look at that is we’re saying, “Is it clear that this thing belongs here and this thing belongs somewhere else?” If people are confused and they go, “I don’t know which category this thing should belong in.” Maybe the categories are not clear. Maybe there’s something wrong there. This is a little bit when should you call in Fran. If you’ve got something that really doesn’t seem to be working, people can’t figure out how to navigate your website, people don’t know where to look for the assets. You’ve got categories, but nobody knows how to apply those tags because it may well be that your categories are not clear. They’re not well expressed, they’re not differentiating what you need to differentiate for the particular purpose and for the particular use case, what you’re trying to do.

Larry:
Yeah, that’s it. You just used the first half of that thing mutually exclusive, collectively exhaustive, like a place for everything in its place. That’s one way you’re like, “Oh, shoot, there’s this thing off to the side. I don’t know where to put it.” Or you have a bunch of empty boxes or something like that. That’s when you call Fran or one of the places that you’d call Fran.

Fran:
Then you can look at, well, what are you trying to do? Why are you trying to separate these? Do you need to separate these things out? Have you expressed clearly? Can you articulate what the difference is between these things or are these things that really sit in between categories? I’m fascinated by boundary objects, the things that fall between two categories and how do you make them findable? I’m thinking of things like where you get things that serve multiple purposes like when you’re shopping for things that could be one thing or the other. Could be used in different ways or types of footwear that might be good hiking shoes, but are also good for walking around town. You wouldn’t necessarily want to categorize them in one place or another, but can you make them findable in these different ways that people think about them or use them and so on?

Fran:
That’s somewhere where the art of taxonomy work comes in. There’s a lot of science in taxonomy work. There’s a lot of things that we can do a lot of number crunching that we can do, a lot of UX work and information, architecture work where you can put categories and concepts in front of people and understand their mental model and how they understand them and see if you’re getting things right and people are able to navigate through this system. You can test that. There’s also the art of coming up with the wording and coming up with the ways of dividing up your content in a way that’s really useful in the particular context that you’re working in.

Larry:
Because so much of that is people stuff. Like you said earlier, agreeing on the terms and getting and understanding their mental models. That reminded me of a thing. I don’t know if you follow Jorge Arango, the information architect. He’s one of the authors of the Polar Bear book. He did this brilliant blog post a couple years ago about different kinds of models. Because I’m a modeler, so apologies for going off on this, but he talked about mental models, conceptual models where you’re like, those are the sharing things where you say, “Hey, do I understand this correctly? Do we agree on what we’re doing with?” Then implementation models. Is there a similar thing going on in taxonomy practice? You mentioned the mental models, how important to understand those. Do you work with a conceptual model and then ensconced in some kind of a system?

Fran:
Yes, absolutely. I think that, I would say really a lot of taxonomies are particularly conceptual models. In fact, one of the ways that I like to describe ontologies is I say an ontology is a map of important ideas. What you’re doing is you’re looking for the conceptual model and you can use those conceptual models to make sure you’ve got clear communication. When you build a taxonomy, you should know what the function or the purpose of that taxonomy is, and there are different types of taxonomies that can do very different things. I talk about descriptive taxonomies and operational taxonomies a lot. Descriptive taxonomies tend to be taxonomies where you’re doing things like content management. You’re wanting to say tag images or tag videos so you can find them again. The set of concepts that you use for those tags should match the mental model of the people who are doing the research and who are trying to retrieve and find those assets.

Fran:
Those are very descriptive. You can look at, what’s a conceptual model that makes sense to us from the point of view of the assets that we’ve got? These are all videos of types of animals, then we’re going to want lots of concepts around animals and different types of animals. Do we know how the people who are searching for those different types of animals, what are they doing? Are they vets or are they animal rescue people and so on? They’ll have different perspectives and different viewpoints. You want to test your conceptual model and make sure it aligns with their conceptual model. When you get those into alignment, then you’ve got a successful system. Then operational taxonomies tend to be quite technical, and they’re interesting because they might be ones that are still working behind the scenes to organize functions and maybe process business rules in code, do logical operations to do data quality checks and validations and so on. Those tend to be a little bit different because you can test those in much more functionally. Are they working, are they achieving the results we expect to achieve through this process?

Larry:
That’s interesting.

Fran:
Again, it’s a question of lining up the results with what you’re trying to achieve and making sure everything is making sense.

Larry:
That sets up a question that just occurred to me. How do you know that a taxonomy is working? That sounds like one way that you have these operational metrics and things. Are there other metrics or criteria that you look at to make sure taxonomy is doing its job?

Fran:
There’s a lot. Again, depending on the different types of taxonomy. With descriptive taxonomies, it really is really looking at that kind of function. What am I trying to do by describing my assets? Who am I trying to serve with this taxonomy? What’s the business reason for having this taxonomy? Is it trying to help people find products when they’re shopping on a website? Is it to help lawyers access the right documents they need? Can they find the legal case precedence, and so on? Are you getting relevant results in the search? You can test that. You can set up reference questions. That’s something that comes into ontology design a lot. What questions do I want this ontology to answer? Do I have a good understanding of the domain? What am I trying to do? Then you can build metrics around that depending on the particular function. Am I retrieving the right assets? If I’m looking for images, am I getting the right set of images and am I getting all the relevant images?

Fran:
Those things are testable and scorable. If it’s a website, if it’s a shopping website, you can make change the taxonomy and you should be able to see an improvement in people being able to find things and buy things. There should be a commercial, an increase in profit that’s coming through that more increased sales, those kind of metrics. There are other things, if you’re doing something more operational, is it making a system more efficient? You can go and talk to people. Is this helping this system run better? Are we getting better data quality? Can we compare states of data and measure that? Do people like using it? That’s a really good, and that’s the same whether it’s how you’ve organized your email inbox or whether how you’ve organized some knowledge organization structures in a large organization. If people enjoy engaging with the taxonomy and if the model that you’ve created fits their mental model and are simplified all the complexity that we’re working with and people like using it, then that’s a huge measure of success.

Larry:
Because when I think of the people I know who listen to these podcasts that there’s a variety of like designers, editorial people, operations managers, all of them, and they all have slightly different ways they would regard the efficiency of the system. One of the things you might look at, a lot of them are working in a CMS and the way they might be exposed to a taxonomy is through tagging or categorizing something as they enter it. I guess that gets into a little bit like the tooling. How are people, assuming the taxonomy is just there and has been done, there’s the use of taxonomy, the consumption of a good taxonomy, what does that look like typically, do you think?

Fran:
Certainly, in the kind of systems that I’ve worked with, you’re usually building a taxonomy for some kind of tagging or indexing system, and it’s usually driving some kind of retrieval in some way. Like you say, CMS or a DAM system, you want to be able to tag an index, the assets that are in that CMS and that DAM in order that you can go and grab and fetch just the right assets that you need, the relevant assets at the right moment. That covers everything from publishing use cases, putting content elements onto a page to publish them, search and research use cases through all sorts of other use cases, recommendation systems, personalization. That’s becoming a big one now, that kind of relevance. The huge vast amounts of content and digital assets that we have now, how do we decide what’s relevant for a particular user to personalize, what can we recommend?

Fran:
Using those similarities, the links between concepts, being able to line up your user profile that says, this person’s interested in these things. This person’s interested in art and music. We have these assets. The taxonomy tells us that we have these assets tagged, art and music, tells us that there’s a connection between art and art galleries. We can then make a connection and say, “Oh, well, there happens to be an art gallery we know about that’s nearby, so let’s surface up this art gallery. Let’s recommend this art gallery to this user while they’re interacting.” Let’s put a nice image of that art gallery and say, “Maybe you’d like to go to this art gallery while you’re in the neighborhood.” Those links and connections, recommendations, personalizations, the way things fit together in the world, it’s very, very powerful.

Larry:
The way you described the personalization use case, it’s hard to imagine doing that without taxonomy. I don’t know how you could. You have to have it tagged on both sides, and especially, at scale. Many of the personalization use cases I’ve seen are just around marketing, and I see a lot of opportunities in other places like ecommerce. I guess in ecommerce you do see a lot of recommendation engines, but I guess do you see other places where taxonomy is underutilized? Like other benefits that you see every day that just aren’t being exploited well in enterprises or other places?

Fran:
I think there’s a couple of things. Obviously, the big hot topic at the moment is improving LLMs, results from LLMs and AIs and intelligence systems and really scaling up, that ability to have… Really, what you’re talking about doing is giving those AI systems really good conceptual models of the world that are human models of the world so that they can then do all their clever processing and come up with results that really work in particular contexts and particular ways. I’m also actually very interested in the logical business rules, encoding logical business rules. Something that is a little bit new to me is not just having the descriptive taxonomies that have the classic taxonomy. This is art, what comes into art and impressionism, expressionism, who are the artists and so on. Actually, going, okay, we can do a little bit more. We can give the developers and the engineers more information about how these concepts should relate to each other, how you should use them.

Fran:
We can encode bits of logical rules into those taxonomy graphs to give instructions about when to use certain concepts in relation to other concepts, how to order results, how to wait results, how to associate, look for changes when changes in one taxonomy, how does that affect other taxonomies? How can we use those ontological rules to say, this class of things needs to behave in a certain way, or this class of data has certain qualities to it, so you can then compare data sets conceptually and create inferences and improve data quality, improve performance, identify anomalies, identify things that are incomplete, create links and associations that you might not have seen before. With lots of business operations, sometimes that can be very small scale and technical, and sometimes it can be very, very wide ranging. That, to me, is a really interesting new development, something people didn’t have.

Larry:
That ensconcing of business logic and rules in these. The way you just described it, it sounds like a full-blown ontology. This kind gets beyond, I actually do a whole other podcast on knowledge graphs, so we should probably talk over there sometime about that. I do think people are going to be exposed more to that kind of stuff in the emerging RAG architectures and there’s a lot of content stuff in and around that. Can you talk maybe just a little bit about the connection between ontologies and knowledge graphs?

Fran:
Some people say that ontologies are knowledge graphs in their own right, but I think the common definition is that your ontology, as I said before, it’s like a map of important ideas. One way that I explain how ontologies and taxonomies work together is that your taxonomies tend to be those nice hierarchical lists. A familiar one is like countries of the world, geography. You’ve got your continents, you’ve got your countries, you’ve got your capital cities, and you can organize that in a hierarchical structure. Your ontology, what your ontology is doing is saying how things like that, how countries might relate to say people or how literature might relate to people and places. It’s the ontology that explains how the taxonomies relate to each other, what relationships they can have, what logical rules can be applied.

Fran:
That’s kind of a schema, it’s a bit like a schema. It is a map, it’s a plan of how will that information connects up. When it really becomes a full knowledge graph and comes to life to be able to answer real questions for your business is when you bring in the actual instance data. These are our particular customers. This customer was born in this place and they’ve bought these products and they like to read these books, and their favorite genre is this. When you’ve got that individual instance data, that’s what gives you a full knowledge graph. The ontology explains how all that fits together, is the organizational principle. Your taxonomies are giving you your options, almost like your dropdown list you can pick from. Then the whole thing when you’ve got that real data coming in of actual transactions, actual customers. That’s what makes your knowledge graph.

Larry:
Fantastic. Because I think that term is popping up more and more, so that’s a great overview, I think. Hey, Fran, I can’t believe that we’re coming up on time. I warned you that this go way too quick. Before we wrap up, is there anything last, anything you want to make sure we share before we close?

Fran:
Well, I would particularly like to give a shout out to Taxonomy Bootcamp London, Bite-sized Taxonomy Bootcamp. The next one is on October the 9th. They’re really good. If you’re interested in learning more about taxonomies, they’re really good. Highly recommend those. We’ll be discovering who the next Taxonomy Practitioner of the Year is at the next one. I really enjoy going to their events and it’s a really great part of the community. Taxonomy Bootcamp London, check that out.

Larry:
I’ve been to many of those, I can also recommend it. I just took a sneak peek at my calendar. This episode will drop the day before. We’ll, probably I’ll drop this, aim to drop this on October 8th so people will still have time to register.

Fran:
Fantastic.

Larry:
Well, oh, and one very, very last thing, Fran, if people want to connect or follow you online, what’s the best place to find you?

Fran:
I’m on LinkedIn. There are quite a few Fran Alexanders on LinkedIn, but if you look for Fran Alexander, taxonomist, Montreal on LinkedIn, you will find me. That’s a good way to get in touch.

Larry:
Excellent. I’ll put that in the show notes as well, of course. Well, thanks so much, Fran. Always fun to talk. This was a really great conversation.

Fran:
Well, thank you, Larry. Appreciate it. Always enjoyed talking to you too.