(14th in my logic of competence series)

Earlier this week I was at a meeting where we were talking about interoperability for abilities, and there was much discussion about the niceties of representation. Human readability is significant — whether the representation reflects what is in people’s minds. The same logic can be represented in radically different ways that are still logically equivalent (and so interoperable); there remains the question of what is identified by identifiers.

A relatively well-known example of variation of readability between different representations involves RDF. RDF/XML has a tendency to make people run a mile, as it can be difficult to comprehend what is represented. Triples formats (e.g. Turtle) at least have a great simplicity to them, because you can see clearly the mapping between the triples and the RDF “graph” (of blobs and arrows) that represents your little corner of the Semantic Web. (I am one of those who only started to appreciate RDF after getting over RDF/XML.) The problem with triples formats is that the knowledge structure is finely fragmented, so you don’t get a clear overview from the triples of what is being expressed: you still need a diagram that represents that overall structure. This is not a surprise — it is generally very hard to serialise a network structure in a comprehensible way. Only particular forms lend themselves to serialisation: e.g. strict tree structures.

In the case of the logic of competence, as I discussed in post 12 in the series, we want to represent both individual competence concepts (or abilities) and structures or frameworks that include several of them.

Published competence frameworks generally use plain text as a medium — they are not primarily graphical or diagrammatic — and have therefore in a sense already been serialised by the publishers. They come across overall rather like a tree structure, though there are very often cross-references and/or repetitions that betray the fact that the information is in reality more complex than a simple tree. But the structure is close enough to a tree to tempt people to want to represent it as such. This is nicely illustrated in Alan Paull’s comment on post 12. Alan’s ability definitions are nested within each other: a depth-first traversal serialisation of the tree if you like.

As Alan and I have agreed in conversation, it is possible to convert a tree-like competence structure to and from other forms. I’ll now give three other forms, and explain how the conversion can be done, and following that I’ll discuss the implications. I’ll call the forms: Atom-like; disjoint; and triples.

First, it can be transformed to a format similar to Atom, where each separate thing (for Atom: entry; here: ability) is given in a flat list of things, each thing including the links between it and other things. (Atom is the format we adopted for Leap2A.) To do this, you take each ability from the tree structure and put it into a flat list, replacing the relation to the nested ability with one using only its identifier, and adding a reverse link from the narrower ability to the broader ability it was removed from. It is also possible to reverse this procedure — start by finding the broadest ability definition (that is, the one which is has no broader links) and then replace narrower links by the whole narrower ability definition, removing the narrower ability’s link to the broader ability. If a narrower ability has already been put in place, leave the reference in place, to avoid duplication.

Second, it can be transformed into a disjoint structure with all the relationships separated out. It’s perhaps easiest to imagine this starting from the Atom-like format, as in the Atom-like format each ability has already been separated out, and there are fewer steps to reach the disjoint form. For each link within each ability, convert it to a separate relationship whose subject is the ability where it is defined, and whose object is the ability referenced. Separate the relationships, leaving the ability definitions with no relationship information included within their structure. An extra step of de-duplication can then occur, because probably the Atom-like format had two representations of each relationship: A narrower B and also the equivalent B broader A. Only one of each pair like this is needed to represent the structure fully.

As in the previous case, it is straightforward to reverse this transformation. For each ability, find the relationships which involve that ability identifier. If the relationship has the ability identifier in the subject position, include a link to the object ability within the ability. If the ability identifier is in the object position, include a link with the reciprocal relationship to the other ability.

Third, it can be transformed by being broken right down into RDF triples. As before, it is easiest to start with the nearest other form — in this case the disjoint one. Take each disjoint ability definition (without relationships). This should convert to a set of triples each with the ability identifier as subject, and probably a literal object. The separate relationships are already in a triple-like format, so they can be converted very easily. To reverse this transformation, examine each triple in turn. If subject and object are ability identifiers, turn the triple into a relationship. Then, for each ability, find all triples that have the identifier of that ability as the subject, and have a literal object, and build a single ability structure out of that set.

Now we’ve seen that these different formats are interconvertable, so which one you use does not impede the communication of a complete ability or competence framework. Where they do differ, however, is in what identifiers are seen to identify, and that does have implications, at least for human use.

Identifiers in RDF triples don’t really identify anything by themselves. An RDF resource is simply a node, with a URI as an identifier. RDF relationships have been called predicates or properties, which is nicely ambivalent about how tight the relationship is. RDF doesn’t tell you which relations relate to things that should be considered as part of the essence of the identified “resource” — or what is inside the “skin” of the resource, if you like. The only thing you can say, when grouping RDF triples together, is that literal properties don’t make any sense by themselves, so they can be seen as attached to, or hang off, a “resource”. In the discussion above, we have assumed that the abilities are the only kind of resources we are dealing with, and that will guide the conversion from the “triples” form to the “disjoint” form.

In the disjoint form, literal properties are grouped with the abilities they are properties of. These properties are likely to include the very well-known ones of title and description at least. The fact that relations are listed separately implies that the relationships are less essential to the nature of an ability than its title and description. In the Atom-like form, an identifier looks like it refers to an ability together with all of its immediate relationships. But in the tree-like form, the identifier of a broader ability seems to refer to the complete structure branching down from it.

Which of these is the most useful or flexible way to identify abilities? That is a real question, and I believe it was the question implicitly underlying much of the discussion at the meeting I participated in earlier this week.

One way of tackling the issue of what is the most useful way of doing identifiers is to look at when you would want to change the identifier. There’s not much one can say about this for RDF triples. For the disjoint form, an identifier would want to change typically when the title or description changes. For Atom-like form, the identifier might reasonably change if any of the direct relationships changed. For broader tree-like structures, the implication is that the identifier should change if any of the structure changes.

When an ability identifier changes is significant. Effective connection between what is taught, learned, assessed, required, claimed or evidenced is only assured if the same identifier is used. If different ones are used for essentially the same ability, extra provision needs to be made to ensure, e.g., that evidence for the ability under one ID can be used to fulfill requirements under a different ID. That provision might be in terms of declaring that two ability IDs actually are equivalent. So, generally speaking, it is reasonable to have ability identifiers changing only when necessary — when what the ability means in practice has actually changed.

So now we can ask: which approach to structuring ability or competence definitions delivers this outcome of needing changed identifiers no more (and no less) than necessary?

The first sub-question I’d like to address is: should changing structure always require changing identifier? My answer is clearly, no, not always, and this is the reasoning. Yes, of course you should change the identifier if the content has changed. But structure change does not strictly imply content change. After thinking a long time about this, I think the clearest example is with intermediate layers of structure. And, happily, this is illustrated in real life with several UK National Occupational Standards. OK, so imagine we have a three-layer competence / ability structure.

Top ability A has two sub-abilities, B and C. B is further divided into P, Q and R, while C is further divided into X, Y and Z. (In real life, there would usually be more.)

The body that defines the structure decides that the justification for the intermediate layer is rather flimsy, and removes it, leaving the structure that ability A has direct sub-abilities P, Q, R, X, Y and Z. The title and long definition of A are unchanged. Is A the same ability? I would answer, unequivocally, yes it is, because all evidence the former A is also equally evidence for the latter A.

Or apply this to the BSHAPM pizza making ability example. A stands for the ability to make pizza the BSHAPM way. B could be baking pizza. P, Q and R could be the three approaches to pizza baking. The BSHAPM could decide that, for simplicity, they wanted to eliminate the node of baking pizza as a separate ability, and instead represent the three approaches to baking pizza as direct sub-abilities of pizza making.

Now if you cling to the view that changes in structure must result in changes of identifier, this means that you will need to declare, and process, a whole extra kind of relationship: that the former A is equivalent to the latter A. This strikes me as unnecessary and quite possibly confusing. Possible: yes; ideal: no. The same example also goes against the Atom-style idea of ability identity. The immediate relationships of ability A change in this scenario, without the ability itself changing at all.

Thus, if we still want to deliver the outcome of changing the identifier only as much as necessary, not more, we are driven to the next type of structural representation, the “disjoint” one. But this comes with a caution. If we are not including the structure as an essential part of the ability or competence definition, we need to be sure that we aren’t cutting corners, and omitting to give a full description of the ability that we can use as a proper definition. Sometimes this may happen in cases where the structure is defined at the same time as the contained abilities. We may simply say that ability A is defined as the sum of abilities B, C and D. Then we risk not noticing that the substance, the content, of an ability has changed, when we change it to being composed of B, C, D and E. So, there is a requirement, to use this “disjoint” approach, that we properly define the ability, in such a way that if an extra component is added, we feel we need to change the definition, and thus the identifier with it. I would say that amounts to no more than good practice. At very least, we should have a long description that states that ability A consists of B, C and D (or B, C, D and E). Or we may choose to make explicit, in text that is not formally structured, the fact that ability A is actually made up of the things that are grouped together under the headings, B, C, etc. Usually, ability A will actually have more to it than simply the sum of the parts. One would expect at least that ability A would include the ability to recognise when abilities B, C, etc. are appropriate, and apply them accordingly, or something like that. So, again, failing to write a full definition or long description is laziness and bad practice.

This reflects back on what I said earlier about a structure doubling as a concept in its own right, or in other words a framework doubling as an ability definition (which also I have actually changed now so as not to have too much hanging around that I no longer believe). Perhaps that needs qualifying and clarifying now. The way I would put it now is that in authoring a competence structure, I am usually implicitly defining a competence concept, but good practice demands that I define that concept explicitly in its own right. It is then true to say that the structure “gives structure to” the concept, in the sense that it details a certain set of narrower parts that the broader concept “contains”. But that is certainly not the only way of structuring the concept. My example based on real NOS cases is only the tip of the iceberg — it is very easy indeed to make up endless examples where the same broad ability is structured in different ways.

It is also not true that a structure necessarily defines a clear single concept. In many cases (such as my BSHAPM pizza making ability) it may, but in very broad cases it may not do. We cannot have that as a requirement for a representation. Thus, contrary to what I wrote at one point previously, it is plausible to have a structure or framework title and definition that is independent of an ability title or definition. It’s just that you can’t use one as the other, and it’s more usual, in less broad cases, to have the structure and the ability concept closely related, perhaps even sharing the title. The structure should not, however, have a long description anything like an ability concept.

Thus, the structure “gives structure to” the concept, and the concept “is structured by” the structure.

Perhaps it is worth remembering that a major envisaged use of these structures, in their structured electronic (rather than less explicitly formatted document) form, is to give learners a set of discrete concepts to which evidence can apply, which can be self-assessed or assessed by others, which can be claimed, or required. Some kind of “container” element at least is necessary in which an ability can be contained or not. The container seems to be exactly the explicit framework structure. In the three-layer example above, the definition, identifier and title of ability A can remain the same, while the framework structure can change from containing in the first case A, B, C, P, Q, R, X, Y and Z and in the latter case A, P, Q, R, X, Y and Z. Applied to pizza making, the frameworks would have to be given better titles than “structure for baking pizzas (layers)” and “structure for baking pizzas (flat)”!

I’d like to conclude by pointing out the trade-off involved in taking the different paths.

• if you proliferate identifiers due to changing identifiers each time the structure changes, you’ll need extra mechanisms to pin together different “versions” of ability definitions that differ only in their structure, or the extent of their structure, not in their substance or content;
• if you want to keep the same ability identifier when the structure changes but not the content, you’ll need to take care to make long descriptions explicit; and to separate identifiers for structure and ability, pinning them closely together where appropriate.

With the latter option, I’m claiming that the ability identifier will naturally change in just those cases where one would want it to change. The cost is getting one’s head round the difference between the ability and the structure. I firmly believe it is worth the effort for system designers to do this, so that the software can handle things properly behind the scenes, while not needing to trouble the end users with thinking about these matters. I’m also suggesting that the latter option requirements embody good practice, where the former ones do not.

This gives us one step forward on the structure diagram from the last two posts, 12 and 13.

Next I’ll firm up on why allowing optionality in competences structures is a good idea, before going on to saying more about how to represent level attributions and level definitions.

(13th in my logic of competence series)

My recent thoughts on how to represent the interplay between competence definitions and structures seem to be stimulating but probably not convincing enough. This post tries to clarify the issues more thoroughly, at the same time as introducing more proposals about the structural relationships themselves.

Clear explanation seems always to benefit from a good example. Let me introduce my example first of all. It is based on a real ability — the ability to make pizzas — but in the interests of being easy to follow, it is artificially simplified. It is unrealistic to expect anyone to define a structure for something that is so straightforward. But I hope it illustrates the points fairly clearly.

I’m imagining a British Society of Home and Amateur Pizza Makers (BSHAPM), dedicated to promoting pizza making competence at home, through various channels. (The Associazione Verace Pizza Napoletana does apparently exist, however, and organises training in pizza making, which does have some kind of syllabus, but it is not fully defined on their web site.) BSHAPM decides to create and publish a framework for pizza making competence, liberally licenced, that it can be referred to by schools, magazines, and cookery and recipe sites. A few BSHAPM members own traditional wood-fired pizza ovens which they occasionally share with other members. There are also some commercial pizza outlets that have an arrangement with the BSHAPM.

The BSHAPM framework is about their view of competence in making pizzas. In line with an “active verb” approach to defining skills, it is simply entitled “Make pizza”. The outline of the BSHAPM framework is here:

• prepare pizza dough
  • with fresh yeast
  • with dried yeast
  • with non-yeast raising agents
• form dough into a pizza base
  • by hand in the air
  • with a rolling pin on a work surface
• prepare a pizza base sauce from available ingredients
• select, prepare and arrange toppings according to eater’s needs and choices
• prepare complete pizza for baking
• bake pizza
  • in kitchen oven
  • in a traditional wood-fired oven
  • in a commercial pizza oven

The framework goes into more detail than that, which will not be needed here. It also specifies several knowledge items, both for the overall making pizza framework, and for the subsidiary abilities.

Clearly the ability detailed in the BSHAPM “make pizza” framework is a different ability to several other abilities that could also be called “make pizzas” — for instance, the idea that making pizzas involves going to a shop, buying ready-made pizzas, and putting them in the oven for an amount of time specified on the packaging.

As well as the BSHAPM framework, we could also imagine related structures or frameworks, for:

• all food preparation
• general baking
• making bread
• preparing dough for bread etc.

So, let’s start on what I hope can be common ground, clarifying the previous post, and referring to the pizza example as appropriate.

Point 1: Frameworks can often be seen as ability definitions. The BSHAPM concept of what it takes to “make pizza” represents an ability that people could potentially want to claim, provide evidence for, or (as employers) require from potential employees. It could be given a long description that explains all the aspects of making pizza, including the various subsidiary abilities. At the same time, it defines a way of analysing the ability to make pizza in terms of those subsidiary abilities. In this case, these are different aspects or presentations of the same concept.

Point 2: Each component concept definition may be used by itself. While it is unlikely that someone would want to evidence those subsidiary abilities, it is perfectly reasonable to suppose that they either could form part of a course curriculum, or could be items to check off in a computer-based system for someone to develop their own pizza-making skills. They are abilities in their own right. On the other hand, it is highly plausible that some other curriculum, or some other learning tracking system, might want not to represent the subsidiary abilities as separate items, particularly in cases where the overall competence claimed were at a higher level. In this case (though not generally) the subsidiary abilities are reasonably related to steps in the pizza making process, and we could imagine a pizza production line with the process at each stage focusing on just one subsidiary ability.

Point 3: The structure information can in principle be separated from the concept definitions. Each ability definition, including the main one of “make pizza” and each subsidiary one, can be defined in its own right and quoted separately. The “rest” of the framework in this case is simply documenting what is seen as part of what: the fact that preparing pizza dough is part of making pizza; baking the pizza is part of making pizza, etc., etc.

Point 4: The structure information by itself is not a competence concept, and does not look like a framework. One cannot claim, produce evidence for, nor require the structural links between concepts, but only items referred to by that structure information. It is stretching a point, probably beyond breakage, to call the structural information by itself a framework.

Point 5: Including a structured ability in a framework needs choice between whether to include only the self-standing concept or to include all the subsidiary definitions. To illustrate this, if one wanted to include “make pizza” into a more general framework for baking skills, there is the question of the level of granularity that is desired. If, for example, the subsidiary skills are regarded as too trivial to evidence, train, assess, etc., perhaps because it would be part of a higher level course, then “make pizza” by itself would suffice. It is clearly the same concept though, because it would be understood in the same way as it is with its subsidiary abilities separately defined. But if all the subsidiary concepts are wanted, it is in effect including a structure within a structure.

These initial points may be somewhat obscured by the fact that some frameworks are very broad — too broad to be mastered by any one person, or perhaps to broad to have any meaningful definition as a self-standing competence-related concept. Take the European Qualifications Framework (EQF), for example, that has been mentioned in previous posts (4; 10). We don’t think of EQF being a single concept. But that is fine, because the EQF doesn’t attempt to define abilities in themselves, just level characteristics of those abilities.

There are other large frameworks that might be seen as more borderline. Medicine and the health-related professions provide many examples of frameworks and structures. The UK General Medical Council (GMC) publishes Good Medical Practice (GMP), a very broad framework covering the breadth of being a medical practitioner. It could represent the structure of the GMC’s definition of what it is to “practice medicine well”, though that idea may take some getting used to. The question of how to include GMP in a broader framework will never practially arise, because it is already quite large enough to fill any professional life completely. (Ars longa, vita brevis…)

It is for the narrower ranges of ability, skill or competence that the cases for points 1 and 5 are clearest. This is why I have chosen a very narrow example, of making pizza. For this, we can reflect on two questions about representation, and the interplay between frameworks and self-standing concept definitions.

• Question A: What would be a good representation of a structure to be included within a wider structure?
• Question B: What difference is there between that and just the main self-standing concept being included?

So let’s try to elaborate and choose a method for Point 3 — separating self-standing concept definitions from structural information. Representing the self-standing concepts is relatively clear: they need separate identifiers so that they can be referred to separately, and reused in different structures. The question to answer first, before addressing A and B is how to represent the structure information.

1. Option one is to strip out all the relations, and bundle them together separately from all the concept definitions. “Make pizza” at the broadest, and the other narrower abilities including “bake pizza”, would all be separate concepts; the “structure for” making pizza would be separately identified. The “make pizza” framework would then be the ensemble of concept definitions and structure.
2. Option two is to give an identifier to the framework, where the framework consists of the concepts plus the structure information, and not give an identifier to the structure information by itself.

Let’s look again at this structural information with an eye on whether or not it could need an identifier. The structural information within the imagined BSAPHM framework for making pizza would contain the relations between the various ability concepts. There would be necessary part and optional part relations. A necessary part of making pizza the BSAPHM way is to make the dough, but how the dough is made has three options. Another necessary part is to form the pizza base, and that has two options. And so on.

So, perhaps now we are ready to compare the answers to the questions A and B asked above. To include one self-standing concept in another framework requires that the main concept is represented with its own identifier, because both an identifier for the framework, and an identifier for the structural information, would imply the inclusion of subsidiary abilities, and those are not wanted. To include the framework as a whole, on the other hand, there is a marked difference between options one and two. In option one, both the identifier for the main concept and the identifier for the structural information need to be associated with the broader concept, to indicate that the whole structure, not just the one self-standing concept, is included in the broader framework. Even if we still have the helpful duality of concept and structure, the picture looks something like this (representing option 1):

If we had to represent the concept and structure entirely separately, the implementation would surely look still worse.

Moving forward, option two looks a lot neater. If the framework URI is clearly linked (though one structural relation) to the main concept definition, there is no need for extra optional URIs in the relations. It’s worth saying at this point that, just as with so many computational concepts, it is possible to do it in many ways, but it is the one that makes most intuitive sense, and is easiest to work with, that is likely to be the one chosen. So here is my preferred solution (representing option 2):

To propose a little more detail here: the types of relationship could cover which concepts are defined within the framework, and specifying the main concept, as well as the general broader and narrower relationships, in two variants — necessaryPart and optionalPart. (I’ve now added more detailed justification for the value of this distinction in post 15 in this series.)

[The following paragraph was altered after reconsideration, 2011-05-26]

One of the practical considerations for implementation is, what has to happen when something that is likely to change, changes? What can and cannot change independently? It seems clear (to me at least) that if a framework changes substantially, so that it no longer expresses the same ability, the main concept expressed by that framework must also change. Evidence for a “make pizza” concept whose structure does not include preparing dough doesn’t provide full evidence for the BSHAPM concept. It is a different concept. On the other hand, if the long description of the concept remains the same, it is possible to have a different structure expressing the same concept. One obvious way in which this is possible is that one structure could for BSHAPM pizza making could include just the abilities listed above, while a different structure for exactly the same ability concept could include a further layer of detail, for example spelling out the subsidiary abilities needed to make a pizza base in the air without a rolling pin. (It looks tricky: I’ve seen it once but never tried it!) Nothing has actually changed, but it is more detailed with more of the subsidiary abilities made explicit.

These arrangements still support the main idea, valuable for reuse, that the concept definitions can remain the same even if they are combined differently in a different framework.

There are two other features that, for me, reinforce the desirability of option 2 over option 1. They are, first, that various metadata can be helpfully shared, and second, that a range of subsidiary competence concepts can be included in a framework. Explanation follows here.

[The following paragraph was altered after reconsideration, 2011-06-08]

First, I am now saying that you can’t change a framework structure substantially without changing the nature of the main competence concept that stands for competence in the framework abilities together as one. The structure or framework would probably be called be something like “the … framework” where the title of the main concept goes in place of the dots. The two titles are not truly independent, but need differentiation, because of the different usage (and indeed meaning) of the competence structure and the competence concept.

Second, if we have an identified framework including a main concept definition as in option 2, there seems no reason why it should not, in the same way, include all the other subsidiary definitions that are defined within the framework. This seems to me to capture very well the common-sense idea that the framework is the collection of things defined within it, plus their relationships. Concepts imported from elsewhere would be clearly visible. In contrast, if the structural information alone is highlighted as in option 1, there is no obvious way of telling, without an extra mechanism, which of the URIs used in the structure are native to this framework, and which are imported foreign ones.

There are probably more reasons for favouring option 2 over option 1 that I have not thought of for now — if on the other hand you can think of any arguments pointing the other way, please preferably comment below.

If I had more time to write this, it would probably be more coherent and persuasive. Which reminds me, I’m open to offers of collaboration for turning these ideas into more tightly argued and defended cases for publication somewhere.

But there is more to finish off — I would like to cover the rest of the relationships that I see as important in the practical as well as logical representation of competence.

However, after writing the original version of this post, I have had some very useful discussions with others involved in this area, reflections on which are given in the next post.

(12th in my logic of competence series, heavily edited 2011-06-17)

One of the keys to a fuller understanding the logic of competence is the interplay between, on the one hand, the individual definition of an ability or competence concept, and on the other hand, a framework or structure containing several related definitions. Implementing a logically sound representation of competence concepts depends on this fuller understanding, and this was one of the requirements presented in the previous post.

A framework by its very nature “includes” (in some sense to be better understood later) several concept definitions. Equally, when one analyses any competence concept, it is necessarily in terms of other concepts at least related to competence. There would seem to be little difference in principle between these more and less extensive structures.

When we consider a small structure it is easier to see that the small structures usually double as a concepts in their own right. To illustrate this, let’s consider UK National Occupational Standards (NOSs) again. Back in post number 3 we met some LANTRA standards, where an example of a unit, within the general topic of “Production Horticulture”, is the one with the name “Set out and establish crops”. In this unit, there is a general description — which is a description or definition of the competence concept, not a definition of its components — and then lists of “what you must be able to do” and “what you must know and understand”. The ability items are probably not the kinds of things you would want to claim separately (there are too many of them), but nevertheless they could easily be used as in checklists both for learning, and for assessment. My guess is that a claim of competence would be much more likely to reference the unit title in this case.

From this it does appear that a NOS unit simultaneously defines a competence concept definition, and also gives structure to that concept.

It is when you consider the use of these competence-related structures that this point reveals its real significance. Perhaps the most important use of these structures by individuals is in their learning, education and training, and in the assessment of what they have learned. In learning a skill or competence, or learning what is required to fulfill an occupational role, the learner has many reasons to have some kind of checklist with which to monitor and measure progress. What exactly appears on that checklist, and how granular it is, makes a lot of difference to a learner. There can easily be too much or too little. Too few items on the list would mean that each item covers a lot of ground, and it may not be clear how they should assess their own ability in that area. To take the LANTRA example above, it is not clear to a learner what “Set out and establish crops” involves, and learners may have different ideas. The evidence a learner brings up may not satisfy an employer. At the other extreme, I really wouldn’t want a system to expect me to produce separate evidence for whether I can start a tractor engine, find the gears, and steer the machine. That would be onerous and practically useless.

Structures for practical use in ICT tools need, therefore, to be clear about the what is included as a separate competence-related definition within the structure, and what is not included, or included only as part of the description of a wider definition.

The “Set out and establish crops” LANTRA unit does have a clear structure, and the smallest parts of that structure are the individual ability and knowledge items — what someone is expected to be “able to do”, and to “know and understand”. And let us suppose that we formalise that unit in that way, so that an ICT learning or assessment support tool allowed learners to check off or provide evidence for the separate items — e.g. that they could “ensure the growing medium is in a suitable condition for planting” and that they knew and understood the “methods of preparing growing media for planting relevant to the enterprise”.

Then, suppose we wanted to include this unit in another structure or framework. Would we want, perhaps, just one “box” to be ticked only for the unit title, “Set out and establish crops”; or would we want two boxes corresponding to the elements of the unit, “Set out crops in growing medium” and “Establish crops in growing medium”; or would we rather have all the ability and knowledge items as separate boxes? None of these options are inherently implausible.

To put this in a different way: when we want to include one competence-related structure within another one, how do we say whether we just want it as a single item, or whether we are including all the structure that comes with it? The very fact that this is a meaningful question relies on the dual nature of a definition, as either a “stand-alone” concept, or structure.

The solution I propose here is that we have two identifiers, one for the concept definition itself, and one for the structure definition that includes the concept. I understand these as closely related. But the description of the structure would perhaps more naturally talk about what it is used for, and wouldn’t be the kind of thing that you can claim, while the description of the concept would indeed describe something that one could claim or require. The structure is structured, whereas the definition of the concept by itself is not, and the structure relies on definitions of subsidiary concepts, that are the parts of the whole original concept.

I illustrated a change of wording in the eighth post, raising, but not answering the question of whether the concept remained unchanged across the changed wording of the definition.

Let’s look at this from another angle. If you author a framework or structure, in doing that you are at the same time authoring at least one competence-related concept definition, and there may be one main concept for that structure or framework. You will often be authoring other subsidiary definitions, which are parts of your structure. You may also include, in your structure or framework, definitions authored by others, or at another time. Indeed, it is possible that all the other components come from elsewhere, in which case you will be authoring only the minimal one concept definition.

One more question that I hope will clarify things still further. What is the title of a structure? The LANTRA examples illustrate that there may be no separate title of a structure, from the title of the main concept definition contained in it. But really, and properly, the title of the structure should be something like “the structure for <this competence concept>”.

Contrast this with the subsidiary concept definitions given in the structure. Their titles and descriptions clearly must be different. They may be defined at the same time as the structure, or they may have been defined previously, and be being reused by the structure.

Exactly how all this is represented is a matter for the “binding” technology adopted. Representing in terms of an XML schema will look quite different from RDF triples, or XHTML with RDFa. I’ll deal with those matters in a later post. But, however implemented, I do think we have the beginnings of a list of information model features that are necessary (or at least highly desirable) for representing this interplay between competence definitions and structures. (I will here assume that identifiers are URIs.)

1. The structure (or framework) has its own URI.
2. The structure often has a “main” concept that represents the structure taken as a separate concept.
3. The structure cannot be represented without referring to the concept definitions that are structured by it.
4. Each concept definition, including the main concept and subsidiary concepts, has its own URI.

I’ve tried to represent this in a small diagram. See what you think…
(note that the colours bear no relation to colours in my other concept maps)

Of course, as well as the URIs, titles and descriptions, there is much more from structures or frameworks to represent, particularly about the relations between the concepts. So it is to the practical implementation of this that I turn next.

(11th in my logic of competence series)

Having discussed, defined, and mapped the principal features of the concepts of ability and competence, we are left with the challenge of working towards “the practical implementation of such competence structures” (ninth post) by looking at the “detailed structure and relationships of ability concepts and structures that contain several of them” (tenth post) and working towards a particular formalisation that represents those concepts adequately for the uses that are envisaged.

At this point, I’ll look back over the posts so far to collect what look like the principle requirements for implementing representations of competence in an interoperable way.

The first post in the series noted that the basis of “what is required” is logically the claim of, or the need for, an ability or competence. Thus an implementation should represent the analysis of “what is required” in terms of abilities. On reaching the sixth post, it was clear that the description of what is required can be formalised to an arbitrary degree, and analysed to an arbitrary granularity, so the formal structures used will need to be flexible rather than rigid.

The second post in the series briefly details the issue of transferability or commonality between different roles. Any formalisation should NOT try to answer questions of transferability, but rather provide a good basis for posing and answering those questions within their own domains.

The third post introduces the idea of abstractions in competence or ability definitions, and “common language between the outcomes of learning, education and training on the one hand, and occupational requirements on the other”. A common language is a language that is reused in different contexts. Particularly when concepts are used in different contexts, it is vital to identify them clearly, so that there is a minimum of ambiguity. Here is not the place to argue for the obvious choice for unambiguous identifier being the URI, but that is what I assume. A URI needs to be given to any ability or competence concept or framework that may plausibly need to be referred to across different contexts or applications. This obviously includes both the case from the second post of transferring between different occupational context, and the case from the third and later posts about what is being learned in education or training contexts being used in occupational ones.

The third post also started to look at some of the large body of UK National Occupational Standards (NOSs). One common sense requirement is that any common representation needs to relate to existing relevant materials. Doing this sets up the possibility of broad and fast adoption (politics and other factor being favourable, and with a fair wind) whereas failure to do this sets up the barrier of having to revise existing materials before adoption. Each NOS is clearly a hierarchically structured document, so a common representation must at least deal with simple hierarchical structures.

The fourth post on levels suggests that a simple hierarchy will often not be sufficient. Both claims and requirements need to be able to include levels, and the representation of levels must allow automatic inferences about higher and lower levels.

The fifth post proposes the requirement for a formal representation to cover the kinds of conditions cited in personal claims and job specifications, that go beyond and detail abstract definitions.

The sixth post starts to suggest some technology ideas for the formal structures, starting with SKOS.

The seventh post points out that decomposition is not the only way of analysing competence concepts. We also need the idea of style, variant, or approach to doing “what is required”. (Though this post did not finally resolve how variants, optionality and levels relate to each other.)

The eighth and ninth posts recognise the value in being able to represent equivalencies and comparisons, across different structures or frameworks as well as within them, and propose using the SKOS Mapping Properties for this purpose.

Listing these requirements in brief, we seem to have something like this:

1. represent competence concepts suitably for reuse
2. represent analysis of competence in terms of abilities
3. deal with levels helpfully
4. cover claims and occupational requirements
5. use SKOS as a basis
6. represent styles or approaches as well as decomposition
7. represent relations across different frameworks

Putting all my proposals for meeting these requirements here would make this post uncomfortably long, so instead I’ll break it down into more bite-sized chunks. (If I change my mind on how to structure the following posts, I’ll change it here as well, and in any case I’ll link from here to following posts when written.)

First, I’ll deal with how we can formally represent individual competence concepts and frameworks so that the structures contain existing materials, can work well together, and can be fully reused.

Next, I’ll put forward my developed ideas on how to represent the structural relationships between competence concepts, and tag on dealing with categories.

Later, I’ll deal properly with the tricky area of levels, for which up to now I have not come across any really convincing solutions.

I’ll do these all with the help of diagrams, representing not the conceptual connections of the previous post, but information modelling connections. This will come together in a big diagram.

I also want to compare and contrast with diagrams representing other past attempts to represent these things, but I haven’t yet decided whether to try to cover that bit by bit while first putting forward the ideas, or to do a big post that covers several alternatives.

After that, for real implementation, we’ll need to discuss the “binding” question — that is, the different ways of representing this emerging information model, particularly looking at XML, Atom, RDF triples, and XHTML+RDFa.

And I hope also to say a word about my great collaborators, the projects we have done or are doing together, and how this work relates to those projects.

At that point, I hope to be able to conclude the series, having outlined a fair solution to the practical representation of the logic of competence!

Now, on to the question of representing how definitions and structures relate to each other.

(10th in my logic of competence series)

In this series of posts I’ve used many terms as a part of my attempts to communicate on these topics. Now I offer definitions for or notes about both the concepts I’ve used in the blog posts so far, and related ones drawn from a range of other work, and I link to posts where the ideas behind these concepts are discussed or used prominently. Then, towards the end of this post (placed there solely for readability) there is a map of how the concepts I’ve used relate to each other.

There are two main sources for borrowed definitions: first, the European Qualifications Framework (EQF); and second, the European Standard that is currently in the process of being published, EN 15981, “European Learner Mobility Achievement Information”, and its published precursor, CEN Workshop Agreement CWA 16133. While I was nothing to do with the creation of the EQF, I am a joint author of CWA 16133 and EN 15981.

Definitions and notes

Criteria and conditions

One particular area that is harder than most to understand is represented by the box called "criterion or condition of performance or assessment" — and this is evidently fairly central to the map below, being the most connected box, and directly connected to the concepts which I originally proposed as logically basic: personal claims may be about meeting these conditions or criteria; job descriptions or requirements may have them included.

Assessment and performance criteria and conditions as general terms are fairly easy to understand in themselves. For assessment, they specify either the conditions under which the assessment takes place, or the criteria by which the assessment is measured. For performance, conditions in effect specify the task that is to be undertaken, while criteria specify what counts as successful performance.

What is less easy to see is the dividing line between these and the ability concepts and definitions themselves, and perhaps this is due to the same fact that we have reckoned with earlier — that how much is abstracted in an ability concept or definition is essentially arbitrary. One can easily read, or imagine, definitions of ability that include conditions and performance criteria; but some do not.

For the purposes of the concept map below, perhaps the best way of understanding this concept is to think of it as containing all the conditions or criteria that are not specified by the ability concept or definition itself; recognising that the boundary line is arbitrary.

To make common sense and to be usable, conditions and criteria have to be grounded in material or social reality — they have to be based on things that are commonly taken to be observable, rather than being based on theoretical constructs.

Concept map

The following diagram maps out several of the ways that the concepts above can be understood as relating to one other. Note that generic language is used in a neutral way, in that for instance the verbs are all in the present tense. However, many of these relationships are in fact tentative or possible, rather than definite, and they may be singular or plural.

Map of related competence concepts

The diagram is a concept map constructed with CmapTools, and includes various other concepts that I haven’t discussed explicitly, but on which I have suggested definitions or notes above. I reckoned that these other concepts might help explain how it all fits together. As always with these large diagrams, a few words of caution are in order.

• This is of course only a small selection of what could be represented.
• It is from a particular point of view, and cannot be perfect.
• Such a map is best looked at a little at a time. Focus on one thing of interest, and follow through the connections from that.

I hope that the definitions and the concept map are of interest and of use. What the map does not clarify sufficiently is the detailed structure and relationships of ability concepts and structures that contain several of them. This will follow later, but before that, I will review the requirements I have collected for implementation.

(9th in my logic of competence series)

My previous post covered how to do common competence features in different structures, typically where the structures share context. But what about when the two structures are from quite different starting points? Equivalences are harder to identify, but it may be useful to document other relationships.

My earlier post on the basic structures taken separately contrasted the UK LANTRA NOS’s with the QAA‘s Subject benchmark statement in the area. The way in which these are put together is quite different, and the language used is far from the same.

But there may be a good case for relating these two. What would happen if someone who has a qualification based on NOSs wanted to give evidence that they have attained Subject Benchmarks? Or, more likely, what if someone who has a vocational qualification in, say, agriculture wants to select modules from a degree course in agriculture, where the intended learning outcomes of the university’s degree course refer to the appropriate Subject Benchmark Statement? Even if there are no equivalences to note (as discussed in the previous post) we may see other useful relationships, such as when something in one structure is clearly part of something else in another structure, or where they are not equivalent, but they are meaningfully related. Let’s see what we can find for the (not atypical) examples we have been looking at.

Starting hopefully on familiar ground, let’s look at the generic skills related to the LANTRA unit CU5 that I’ve mentioned before. Element CU5.1, or unit CU5 in the 2010 Veterinary NOSs, is called “Maintain and develop personal performance”, and this seems related to the Benchmark’s “Self-management and professional development skills”. They appear not to be equivalent, so we aren’t justified in creating a skos:exactMatch or skos:closeMatch relationship between those two structures, but we could perhaps use skos:relatedMatch (another SKOS Mapping Property) to indicate that there is a meaningful relationship, even if not precisely specified. This might then be a helpful pointer to people about where to start looking for similar skill definitions, when comparing the two structures. The Benchmark seems to be generally wider than the NOS unit, and perhaps this would be expected, given that graduate level skills in agriculture should cover something that vocational skills do not. Here, “moral and ethical issues” and “professional codes of conduct” are not covered in the NOSs. Perhaps the closest correspondence can be seen with the Benchmark’s “targets for personal, career and academic development”, prefaced at “threshold” level by “identify…”, “typical” level by “identify and work towards…” and “excellent” level by “identify and work towards ambitious…”. In the NOS, the individual must be able to: “agree personal performance targets with the appropriate person”; “agree your development needs and methods of meeting these needs with the appropriate person”; “develop your personal performance according to your agreed targets, development needs and organisational requirements”; and “review personal performance with the appropriate person at suitable intervals”. They must also know and understand (among other things) “how to determine and agree development needs and personal targets”. Personally, I’m not sure whether anything deserves a skos:closeMatch property — probably what we would need to do would be to get the relevant people together to discuss the kinds of behaviour covered, and see if they actually agree or not on whether there was any practical equivalence worthy of a skos:closeMatch.

There is also a definite relationship between the Benchmark’s “Interpersonal and teamwork skills” and the NOS’s “Establish and maintain working relationships with others”. Again, it is difficult to identify any very clear relationships between the component parts of these, but despite this lack of correspondence at fine granularity, it seems to me that the five ability points from the NOS are more than covered by the five points appearing at the “typical” level of the Benchmark. There are two other SKOS Mapping Properties that might help us here: skos:broadMatch and skos:narrowMatch. These correspond to skos:broader and skos:narrower, but applied across different structures, rather than within one structure. Thus we could potentially represent that LANTRA CU5A (2010) has a “skos:broadMatch” in the Benchmark’s Interpersonal and teamwork skills, “typical” level. Conversely, that “typical” Benchmark component has a “skos:narrowMatch” in LANTRA’s CU5A.

On the subject-specific end, again there are plenty of areas where you can see some connection, but hard to see very clear distinct relationships. As you might expect, there is a tendency for the NOSs to deal with specific skills, while the Benchmark deals in more general knowledge and understanding. The horticultural PH16 NOS unit is titled “Respond to legislation affecting consumer rights”, while the Benchmark has various “subject-specific knowledge and understanding” to do with “social, economic, legal and technological principles underlying the business management of farm or horticultural enterprises”. Probably, people meeting this part of the Benchmark standard at a good enough level have skills that include that unit of the NOS, so we could in theory note a skos:broadMatch relationship between the NOS unit and that part of the Benchmark. But we could only do that (for any area) if we had URI identifiers available to mark the relevant sections unambiguously, and at present there are few if any competence structures where URIs have been officially assigned to the parts of the structure.

It seems unlikely that an agriculture graduate would be wanting accreditation of a LANTRA NOS unit, but if someone did, supporting systems could potentially make use of these relationships represented as SKOS Mapping Properties. More likely, someone who has covered the LANTRA NOS would be able to save a lot of time in putting together a shortened agriculture degree programme if all the skos:broadMatch relationships were documented, as it would be relatively easy to design a tool that allows efficient comparison of the relevant documentation, as a support to deciding whether a particular module at degree level needs to be taken, or not. This seems likely to be a similar process to Accreditation of Prior Learning (APL) in which the university accredits previous attainment in terms of their degree programme. It could also be adapted to APEL (E = “Experiential”) if the individual brought along a portfolio of evidence for attaining relevant NOSs. These processes are important in the likely future world where tailoring of degree courses becomes more common.

It looks like I have finished the coverage of the essential logical features of competence structures that I believe could usefully be incorporated in an interoperability specification. To repeat a point I have inserted in the introduction to this series, I would be delighted to discuss any of these posts one-to-one with interested people. It remains to bring all these points together in a way that is easier to follow, through the judicious use of diagrams, to discuss other emergent issues, and to talk about how we could work towards the practical implementation of such competence structures. The first diagram offered is a concept map, together with definitions.

(8th in my logic of competence series)

In the last two posts, I’ve set out some logic for simple competence structures and for more complex cases. But we still need to consider how to link across different structures, because only then will the structures start to become really useful.

If you look at various related UK National Occupational Standards (NOSs), you will notice that typically, each published document, containing a collection of units, has some units specific to that collection and some shared in common with other collections. Thus, LANTRA’s Production Horticulture NOSs (October 2008) include 17 common units that are shared between different LANTRA NOSs, and Agricultural Crop Production NOSs (May 2007) include 18 of these units. Ten of them appear in both sets. Now if, for instance, you happen to have studied Production Horticulture and you wanted to move over to Agricultural Crop Production, it would be useful to be able to identify the common ground so that you didn’t have to waste your time studying things you know already. And, if you want to claim competence in both agriculture and horticulture, it would be useful to be able to use the same evidence for common requirements.

How can what is in common between two such competence structures be clearly identified? There are currently common codes (CU2, CU5, etc.) that identify the common units; and units imported from other Sector Skills Councils (as frequently happens) are identified by their unit code from the originating NOSs. However, there are no guarantees. And if you look hard, you sometimes find discrepancies. CU5, for example, “Develop personal performance and maintain working relationships”, is divided into two elements, “Maintain and develop personal performance” and “Establish and maintain working relationships with others”. In both sets, “others” are defined as

1. colleagues
2. supervisors and managers
3. persons external to the team, department or organisation
4. people for whom English is not their first language.

But when the unit CU5 appeared in Veterinary Nursing NOSs in 2006, non-native English speakers were not explicitly specified. Do we have to regard the units are slightly different? We can imagine what has happened — presumably someone has recognised an omission, and put it what was missing. But what if that has been reflected in the training delivered? Would it mean that people trained in 2006 would not have been introduced to issues with non-native speakers? And does that mean that they really should be given some extra training? And later the plot thickens… LANTRA’s “Veterinary nursing and auxiliary services” NOSs from July 2010 has CU5, “Maintain and develop personal performance” and CU5A, “Establish and maintain working relationships with others”. This seems to follow a pattern of development in which the NOS units are simplified and separated. The (same) different kinds of “others” are now just included in the overview paragraph at the beginning of CU5A.

I hope it’s worth going through this exercise in possible confusion to underline the need for links across structures. Ideally, an occupational standard should be able to include a unit from elsewhere by referring to it, not by copying it; and there would need to be clear versions with clearly marked changes. But if people insist on copying (as they currently often do), at least there could be very clear indications given about when something is intended to be exactly the same, and when it is pretty close even though not exactly the same.

Back in the simple competence structures post, I introduced the SKOS relationships “broader” and “narrower”. There are other SKOS relationships that seem perfectly suited for this job of relating across different competence structures. These are the SKOS Mapping Properties. It would seem natural to take skos:exactMatch to mean that this competence definition I have here is intended to be exactly the same as that one over there, and skos:closeMatch would serve well for “pretty much the same”, or “practically the same”. If these relationships were properly noted, there could be ICT support for the kinds of tasks mentioned above — e.g. working out what counted as evidence of what competence, and what you still needed to cover in a new course that you hadn’t covered in an old course, or gained from experience.

And if all parts of competence structures were given globally unique IDs, ideally in the form of URIs, then this same process could apply at any granularity. It would be easy to make it clear even to machines that this NOS unit was the same as that one, right down to the fine granularity of a particular knowledge or ability item being the same as one in a different unit. An electronic list of competence concepts would have alongside it an electronic list of relationships — a kind of “map” — that could show both the internal “skos:broader” and “skos:narrower” relations, and the external “skos:exactMatch” and “skos:closeMatch” equivalencies.

This gives us a reasonable basis for reuse of parts, at any level, of different structures, but we haven’t yet considered comparison of competence structures where there aren’t any direct equivalence mappings.