Rob Hanna and Lance Cummings: Information Typing in DITA

Speaker A: This is the Content Strategy Insights podcast, episode number 215. When you're trying to communicate with either human beings or computers, it helps to align the kind of information you're imparting with the way the recipient expects to receive it. Technical writing experts call this information typing the practice of organizing content around its intent, not its format, which better aligns the information to human cognitive preferences. It turns out that this practice also helps AI systems deliver better results.

Speaker B: Welcome to the Content Strategy Insights podcast where accomplished content strategy experts share their wisdom with our friends in the content community. Our mission is to democratize content Strategy to make its principles and practices accessible to everyone. And now here's your host, Larry Swanson.

Speaker A: Hi everyone. Welcome to episode number 215 of the content Strategy Insights podcast. I am really delighted today to welcome to the show, uh, Rob Hanna and Lance Cummings. Rob is the CEO and the founder of Precision Content, a technical content um, consulting uh and company. Uh, and Lance is a professor of professional writing at um, UNC Wilmington University of North Carolina at Wilmington. Uh he's also a well known content creator. We'll talk a little bit more about that. So welcome to both of you and um, maybe start with Rob. Rob, tell the folks a little bit more about what you're up to. These.

Speaker C: Perfect. Thanks Larry. Yes, so we've been um, pioneering a writing methodology um, that uh, we believe makes content easier for um people to use and uh, as uh it turns out easier for machines to use. So there's continuous development and research in this field on how we um, bring uh, bring these together and create better future proof content. So I've been uh, embroiled in that for the last uh number of years.

Speaker A: Nice. And Lance?

Speaker D: Yeah. So I've been teaching and researching how we write and collaborate in diverse linguistic and technological environments and I've been writing about AI um since before chatgpt and online. And um, kind of found my niche with technical writing. Was uh, introduced to, did a um, probably back in 201718 something like that at a conference and been trying to figure it out ever since. Um and uh, really ran into uh, Rob's stuff at a conference and the information typing kind of clicked for me. Whereas like um, I understand Dita, uh and the use cases but haven't necessarily had the time to really dig in and use it on my own or, or to teach a class on it. Um, but uh, the information typing really uh, really got me thinking about structured uh, writing in a different way especially in concerns with AI um which I've been exploring. On my blog, Cyborgs Writing on Substack, um, been looking at structured approaches, how you take structured approaches, combine that with rhetoric to make AI work better. And I've really found information typing to be a pretty big part of that recently and been working on uh, developing approaches, um, in AI writing systems, uh, and things like that.

Speaker A: Nice.

Speaker D: Yeah.

Speaker A: And this. And you're reminding me of the reason this conversation came together. One of you, I think I can't remember which of you posted it, but you had jointly created um, this document about, called dita. Information Types aren't Templates, which I think is really um, one of the key concepts behind dita. Um, I wonder maybe Rob, maybe you can talk about that a little bit. Like what led to that article, um, and a little bit about uh, the difference between uh, types and templates.

Speaker C: Sure. Perfect. Thank you. Yeah. Um, this has been a crusade of mine for a very, very long time. Um, understanding what the information typing is. Uh, I like to say it's at the heart of dita, the ITA Information typing. It's right there. But there's so little discussed or um, talked about about what information typing actually is. Um, in my work with the uh, oasis, uh, Subcommittee, Business Documents Subcommittee back in the aughts, um, we uh, were looking at how we extend uh, beyond um, product support and into the rest of the business. So um, I studied a lot of different models, uh, that we could look to for extending data beyond simply concept, task and reference and landed upon information mapping. And so that's where I discovered sort of the genesis of um, information typing as defined by Robert Oram back in the late 60s at Stanford University, um, where he came up with these ideas around um, these information types based on patterns that he had found in textbooks. His task was to try to find a methodology that would uh, help to build a better textbook. And so through his analysis of a lot of um, text, he found these patterns in that content. And from there a number of other uh, folks in pedagogy and other uh, areas started to look at why do these patterns emerge? And so where we've landed with that is that information types represent the function of information. In other words, how do our brains work with that information? How do we lower cognitive load for the reader? By aligning information by function and not mixing content together that have different, um, purposes, different functions, different outcomes. So we can sort that apart, tease them apart and signal to the reader what type of information it is that they're looking for. We can dramatically reduce the cognitive load and improve performance on that content. So it's well beyond what a template was ever sort of designed to do. This is more a template for your brain. This is how your brain works and how we template the information. Go in here and actually do something with that information.

Speaker A: Yeah. As you said that it's like, it's a higher level than just like um, WYSIWYG thinking. It's like, what are you doing here? What's the function versus the form that it takes in its expression? And one of the things that makes it so powerful, I think, is the, its alignment with how humans think. Maybe, maybe you can talk a little bit about this, Lance, is how, um, the cognitive science that supports information typing and sort of a lot of other DITA stuff as well.

Speaker D: Yeah. So I mean, I'll start by saying like one of my hobbies has been to take the idea behind DITA and try to apply it to my education, my education materials that I use with students and also my own like notes, structures and things like that. Because I see the value of reuse and uh, modular thinking. And that's one of the things that I did in my blog when AI first came out is to try to apply those. But a lot of times I, you know, trying to use the data labels, they didn't quite fit and uh, I had a lot of trouble and sometimes just make m. Make. Make up your own labels and things like that and really, uh, and tried to teach that structured thinking to students in that way, which, you know, with some mild success. But then when I ran into information typing, it just all kind of clicked. And uh, and I think it's because it matches uh, those ways of thinking that we're already intuitively following but we don't always explicitly talk about. And I would say that's one job of a, Of a RH teacher or a writing teacher or even a literature teacher is uh, is to make explicit the templates that we're using when we're writing. So uh, there's a study, um, uh, I can't cite it off the top of my head, but of literature classes where um, uh, they find that literature teachers are expecting certain templates when they're, when they're looking grading papers, but they don't necessarily make those templates explicit. Uh, and so they listed. So they figured out what some of these templates are and then taught them to students and then the grades went up. So it's really the same kind of thing. And um, I really connect this back with my background in rhetorical theory all the way Back to Aristotle, where he's talking about what he calls topoi. Um, and these are, uh, and literally it means little rooms in your head. So they're like patterns that speakers can access when they have to give an impromptu speech. Um, because that was what you did back there. Back then, you didn't write blogs to, you know, to convince people. You had to go up and make speeches and you'd be like, okay, this argument needs a process argument. This argument needs a comparison argument. And you could construct a speech just like that. Um, and that's really followed us through the history of communication, uh, communication, rhetoric and writing. Back, um, in the 19th century, there was the modes of discourse, um, which, you know, comparison, definition, process. So I already kind of had an idea of these information types already, but hadn't quite connected them, uh, to structure quite the way that Robert Horn had in his research in the 60s, um, and 70s. Uh, and I think that's why, uh, and I found teaching students, when I introduce it in this more methodical way, um, they. It clicks better for them because. Because it's more visible and explicit.

Speaker A: I gotta say, I'm loving. I don't know that either of you knew this before this call, but my first career before I got into digital content was in academic textbook publishing. And we thought a lot about this from that angle. And I love that some of the original research, like you were saying, Rob, that goes into this comes from a guy trying to write better textbooks. And then what you say in Lance, applying these principles to your educational materials, I love this alignment. But you both mentioned in your last few minutes this notion of patterns. And when you start to articulate patterns of stuff that happens, you end up with something that looks like a design discipline. Um, is that a good way to think about design, uh, infotyping as a design discipline?

Speaker C: Ah, absolutely. It's not simply a metadata label. We'll slap on a piece of content, right? We need to be able to telegraph to the reader what type of information this is based on how we title this block, this topic, this chunk of content, um, what sort of structures we provide to this information so that visually they can look at something and they can get a, uh, signal as to what type of information this is and then how we write it. Second, third person, active voice, whatever the information type requires, we want to be consistent in how we're writing and structuring that information, ensuring that the content is concise, it's well structured, and it is, um, on it's relevant for the type of information that it's trying to convey.

Speaker B: Yeah.

Speaker A: Um, that line. I think I pull a quote from that. I think your paper. Our brains work according to defined functions that aid in comprehension and application of information. That's what you just said, basically. Um, but can you talk a little bit about. Because I think so many. I think a lot of people. Because, like the way CMSs work and things like that, that a lot of people conflate structure and the intent of a thing.

Speaker C: Absolutely.

Speaker A: And this teases that out. Can you talk a little bit about, um, either of, or both of you about how that teasing out and being clear in your mind helps, uh, uh, communicate ideas better?

Speaker C: Well, uh, if you don't mind, I'll start on this one. I mean, uh, Robert Horn was one of the first to coin the term structured authoring. And if you go on to Google Scholar and you look for Robert Horn, you'll find some of his earliest works where he structured content using a typewriter. It looks terrible, but, uh, all is to say is that structured authoring, uh, preexisted before X amount did. Right. So structuring content is not about the tooling you use or the format behind the content that you're using. It has everything to do with using, um, repeatable patterns and language. And that's really what it is. It's creating these patterns that I recognize. And while the dita, um, uh, tag set may be somewhat verbose and complicated, but our brains naturally acclimate to these patterns if we use them regularly, very, very quickly. So once I see a pattern and I understand that that's a task, I'm going to see another task. Um, it's being telegraphed before I even read the words, what it is. And I know how to handle that. I know what part of my brain I need to turn on. I love what Lance was talking about, rooms in your head. Because we believe that there's, on the cognitive science side of things, is that these rooms represent different types of memory that we need to access to provide context to information as we're reading it, as we're working with it, as we're using it. So, you know, it really is about how do I activate the right part of the brain to be able to work with this piece of information most effectively.

Speaker A: Nice. Oh, I was going to say, Lance, I was going to ask you, uh, just say what you were going to say because I had a question as well.

Speaker D: Oh, okay. No, I was just going to say, like the idea of design. I think one of the things that attracted me to the information typing system system is that it's, it is more than just structure and it is rhetorical in the sense that, you know, you are trying to put the intent into the, into this, into the structure, into the piece that you're, that you're creating. And that requires intentional design of the content, not just, you know, labeling it like a certain kind. And now with AI, it's more uh, important than ever because, uh, if that intent's not there, then the AI just kind of guesses what the intent is or, you know, or, you know, just forgets about it, you know.

Speaker A: Yeah, hey, I'd love to at this point, kind of like, because we've talked, we've alluded to and talked about a little bit about the types, um, and there's three kind of main types in dita and then there's extensions where you can add more and you know, just the fact that they align so well with your rhetorical understanding and approach to things and the needs and technical communication. Um, maybe M. Rob, maybe you can talk a little bit about just the overview of the types, the basic types in dita.

Speaker C: Yeah, you bet. So, um, as I said before, Robert Horn really sort of pioneered uh, this area and he originally identified seven different types of information as he sort of, um, boiled the ocean of textbooks, sorting information not by what it said, but how it said it. Um, those seven types were procedure, fact, um, process, concept, structure, ah, and principle types of information. Now the only word that's familiar with dita, there is concept. Well, there's a concept in DITA and there's a concept in um, uh, information mapping. But if you break it down further, where a fact is reference and procedure is task, um, you have other information types, um, that appear in business documents that are broken down in this model. And so with precision, content, we've adopted uh, concept, task and reference. But we've introduced principle and process as being, um, two additional archetypes from which we would create, uh, content for. And we would create specializations around if we needed to extend those further. So piece of principle information is really talking about, um, uh, advising the reader what to do or not do and when. So this may be in the form of an admonition, uh, like a caution, a warning, it could be a tip, it could be a best practice, it could be ah, anything that sort of like I say, is telling somebody, don't do that or do do that, um, and when, like I say, um, so there's no real structure for that. We have notes in DITA that we'll just pop a Note in the middle of a task and say, don't stare at the laser. Um, we've got a little bit more sophisticated uh, uh, admonitions with uh, for machinery. Um, but by and large there's no topic type for that. So we created a topic type for that and a writing methodology around principle and process. Wow. Process is the big missing archetype for get in technical communication right now where we're talking about how things work. When I say things I mean how did mail get delivered, how are my taxes processed? You know, how, how does the print, like how does the printer work? Or um, how does, how do planes fly? Right. So each one of these have actors. Each of these actors have actions that they perform. These identify accountabilities and who's responsible for what and in what sort of process or what order do these get uh, performed in to accomplish a particular outcome. We don't see a lot of this described in technical documentation typically because the processes are not well defined on how would I use a piece of commercial software to do a thing? There's no well defined process on how that, that works, that naturally flows out of it. But when you're looking at banking procedures or you're looking at something that's much larger, process starts to take on a much more important role. And when we look to agentic AI, agentic AI is really looking for process information far more than it's looking for task based information because it needs to understand what are the outcomes and what are the types of things that need to happen in order to reach that particular outcome. So anyway, I'm big on process these days. They really help to unlock and make it easier to get into context for tasks that need to be performed. But again, they're going to have a huge impact I think on the uh, agentic AI side.

Speaker A: Yeah, no, I can totally see that. Lance, I'm curious now about how. Well all of what Rob just said aligns with like your rhetorical approach to things, your rhetorical background. But also you've done a lot with AI and a lot of experimentation and stuff and maybe how the um, like especially the principles and I think maybe all of the types, how they contribute to this notion of context that everybody has become so much more aware of with the arrival of AI.

Speaker D: Yeah. So you know, really what rhetoric comes down to is, I mean the traditional word is of definition, uh, is the available means of persuasion. Um, but you know, really it has many different, different niches now depending on what you want to do with it. But really it's about shaping the world to, to m. To make things, get people to do things that you want them to do or to convince them to see the world in a certain way. Um, and uh, and in machine rhetorics was what I call it be convincing the AI or giving the AI the vision to actually do what it is that you want it to do. And so, um, being able to lay out the, the actual, uh, the actual different kinds of information, like that really makes it especially just like if you were to give it to a student, for example. Um, you know, a lot of times our worst assignments are because we don't make those things explicit. We give them the task but not the process. Right. Um, or we give them the process, but we don't give them the concepts they need to know to do the process, understand the process, that kind of thing. And it's really the, the same thing with, with agents, um, or with the documentation that those agents, uh, create. Um, and I've really found that, uh, increases the, the uh, the AIs perform better. Oh, it helps you really, uh, expand on your documentation too. So for example, I've like built documentation for a study abroad bot, for example, and you know, which can run on my notes, like, which are just kind of random scattered things. And it can kind of pick and pull things and then kind of guess at what m I might need in an answer. But if I go through my notes, uh, and say, okay, this is actually a process. Let me actually describe that as a process in more detail. And then that gives you like, questions that somebody might ask about that process. And then you can add things to that that you was missing because, um, they. Somebody asked that question. Or you can have the AI be like, okay, what questions might somebody ask about this. This topic? And then you can add those things to it. And then it really helps you expand the documentation. Not just categorize everything, but realize what's missing, ah, what's mislabeled. Uh, and that, that, that really improves what the AI has available for its outputs. Um, not just, not just as a way of making the outputs better, but more complete and accurate.

Speaker A: Rob, I see you nodding vigorously. I wonder if you have anything you want to add to that.

Speaker C: Well, uh, you know, I think we, we lose the connection between how our brains work and what content is. Right. Content comes from our brains. Our brains shape these patterns. These patterns show up in content. They're maybe not visible when it's right in front of you, but when you step back to 50,000ft, you see these patterns just like Robert Horns other's patterns. By pulling apart all those textbooks, AI sees these patterns in our content that we don't see up close. And that's what makes AI able to better retrieve information by its general shape based on the human corpus of content that it is consumed. So it's really bringing content back to human cognition, to machine cognition. Machines understand it well because they're being fed off of our human corpus of content. That's where the connection really lies. Finding those patterns and exploiting those patterns.

Speaker A: Yeah, and that's really interesting that I've read a number of articles recently on the differences in how machines do cognition and how humans do. Like there's this famous, um, uh, the epistemological disconnect between human and computer, uh, language abilities. And I've seen others as well. Um, but it sounds like in Lance, I know you've seen evidence of this and you've both seen evidence of how structuring content in this way and making that meaning explicit, and not only making the meaning explicit, but structuring it consistently. Um, is there any science or um, studies that show X percent better retrieval, uh, or understanding question answering ability or things like that with structured content? Either of you?

Speaker C: Well, okay, so, uh, we've done some experimentation, um, with retrieval, uh, augmented generation of rag on taking our content and breaking it down into these chunks and then looking at how we use RAG with the metadata layer and with the content broken down into what we call micro content pieces and looking at the um, uh, frequency in which it returns accurate results. So there's a very small scale, um, test. We don't have a large corpus of content to test against. But our very earliest experimentation is showing very, very positive uh, results for improved, uh, retrievability of that content. But I think Lance has a, uh, more interesting uh, story to tell around experimentation in this area.

Speaker D: Oh yeah. So I mean, I think, in short, there isn't really any, uh, research that at least that I found. Maybe it's out there and somebody can send it to me, but, um, that specifically addresses this. There are some uh, research done on chunking and RAG and things like that that are tangential to this, but don't focus on actual information types. Um, partly because a lot of those studies are done in computer engineering. And so they have a particular mindset when they're approaching those projects. Um, and so one of the goals that Rob and I have is to do this kind of study in a way that we can replicate it and share it with the rest of the industry and, and uh, as part of that, I've been working on a grant in my university with a class that, uh, with my writing with AI class. But we're also working with a computer engineering class on a disaster communications chatbot. Uh, so we're in Wilmington. We have hurricanes and stuff. Disaster communication is a big thing. And uh, what we did was we had students, we created a taxonomy last semester. Some of my colleagues of topics that would be in a chatbot. Uh, we adapted that and then we separated those topics among all the students. And their goal was to create system prompts together along and then to feed the AI the knowledge through their own rag systems. We actually used open source models, um, with LLM Studio and then to run tests to see how it works. Now I had asked my English students, um, to do two versions of the Knowledge. Um, the, the normal way that people are doing this, and especially in computer engineering, is going to the website and stripping it, um, with, uh, with Fire Crawl or whatever, or uploading the PDFs. Um, uh, and then I was asking my English students, okay, well take a couple of these key pieces and break it down into information. Rewrite it as a, as a topic with the information types in there and then add things in it that you think would be important for your UNCW audience. So it becomes more, more customized, um, and also more structured and organized. Not all my students did it. It was a very hard project to do. But, um, but not all my students did it. But I ended up with a group of chatbots that had the structured knowledge and a group that didn't. And there was a pretty big difference with how they performed. Um, you know, some chatbots were providing wrong phone numbers, um, that didn't have that.

Speaker A: You just.

Speaker D: They put in the UNCW websites, for example, um, or they, they weren't appropriately, um, giving context for like rules about feeding infants, um, things like that. Um, but the ones with the structured knowledge way outperformed, um, the other ones. We use like a perplexity to basically judge the answers according to human answers and using specific test questions that the students came up with. And it's not particularly a scientific study. It was a classroom project. Um, but I think it's something we could replicate. And it was very interesting to see those differences. And then the students who did structure their knowledge, seeing their eyes kind of light up when they finally, when they realize what they're seeing, um, uh, which is a, which is a really amazing thing and something hope to continue doing, you know, with Rob.

Speaker A: Yeah, what you both just said, I Also spend a lot of time in the knowledge graph world in the symbolic AI space. And there is a lot of evidence there that supports the notion of like ontologically structured knowledge graphs just way outperform. So it's. And you both have done experiments that are sort of like on the path to valid, further validating that. But, um, I can't believe we're kind of close to time already. But before we wrap up, I want to make sure I give each of you a chance. Is there anything last, anything you want to revisit from the conversation or that you just want to make sure we share before we wrap up? Uh, Rob, I'll let you go first.

Speaker C: M. Well, um, it's through looking at, you know, how are we going to create content that is going to improve machine and human performance. Right. So studying that. If you're not in structured authoring yet, um, you don't know what structured authoring is. You've got to dig into this because I think it's going to be highly, highly important to your careers as technical writers, technical communicators to be able to be well versed in how structured authoring works, um, and what sort of advantages you need to give your content so that it'll work well for machines as well. So I think we're on the cusp of just not understanding structured authoring is just not going to be an option for technical, uh, writers going forward.

Speaker A: Nice length.

Speaker D: Yeah, yeah, I would certainly agree with that. Um, also, uh, I mean the ability to see these patterns, uh, uh, I found our, you know, English students for example, or writers in general are really good at seeing these patterns, maybe not necessarily articulating them, which the information typing gives us, the verbs in the organization to say, uh, the words to labels to use to describe these, especially for AI. But I would say that information typing isn't just for technical writers. And that's kind of probably the case I'll be making on my substack, um, Cyborg's writing. Um, that certainly, uh, learning about this is I think going to be necessary for technical writers and content designers and all that, um, people working with AI. But I think it's useful for everybody to think about. I mean most people will be working with AI probably in the workplace, but even just thinking about how you structure your email, how you structure, uh, your syllabus or your assignment, uh, it's pretty useful in all those areas and uh, is kind of the case I'm making in my classes. And honestly, if I teach it in An Introduction to Professional Writing. I actually believe those students will learn more about writing and content, uh, than if they just were writing the traditional essays that we're always assigning in those classes, or doing reports and emails and things like that. Being able to think about, structure and use it and, uh, creative ways is really, I think, where, where we're at for at least professional writing, in my opinion.

Speaker C: No.

Speaker A: And as you were both talking, it occurs to me that, like, we're all, you know, the language is the interface now to everything. So we're all writing all the time, and it behooves all of us to, to develop these skills. Um, hey, one very last thing. What's the best way to connect with each of you?

Speaker C: Um, uh, Rob M. The best way to connect with me is reach out to me on LinkedIn, right? So I'm happy to make connections. Uh, I'm publishing there, um, multiple times per week, providing links into my substack. Uh, I'm only just getting started with, uh, my articles there, but I'd say the best way to reach me is by LinkedIn. And then from there you can DM me or send me an email or anything you want. So look for Rob Hannah. I'm also known as the Single Sorcerer, um, on, uh, on, on LinkedIn. So do a search and happy to connect. Great.

Speaker A: Lance.

Speaker D: Yeah, yeah, so I'm on LinkedIn a lot. That's where I, I, I post the most. And then I, of course, got my substack. It's called cyborgs writing, um, www.and, uh, that's where I'll be writing a lot about, um, using information typing and AI and, uh, content, things like that, and tech writing.

Speaker A: Cool, thanks. Yeah, and I'll put all that in the show notes as well, of course. Um, well, thank you both so much. I really enjoyed the conversation.

Speaker C: Thank you. It was great.

Speaker D: Thank you. It was fun.

Speaker B: Thank you for listening. If you can think of a friend who might enjoy this episode, please share it with them. And please join us again for our next content strategy interview.