Le 02/04/2018 à 14:30, Thomas Beale a écrit : > > > On 02/04/2018 10:59, Philippe Ameline wrote: >> >> Le 01/04/2018 à 14:13, Thomas Beale a écrit : >> >>> >>> just by means of clarification for some readers, since I happen to >>> know how both openEHR and Philippe's system works, here's the way to >>> understand how openEHR would perform the same function as >>> Ligne-de-vie (which it can): >>> >>> * build a lot of CLUSTER archetypes, and probably more OBSERVATION >>> ones; each CLUSTER archetype would be one of the 'trees' >>> Philippe talks about >>> * each of those CLUSTER archetypes has slot nodes that indicate >>> where subordinate CLUSTER archetypes join, and which ones are >>> allowed, in the usual openEHR fashion; >>> o remember, a slot can allow multiple possible substitutions >>> * at run time, a form containing a top level Entry, usually an >>> OBSERVATION will be deployed on the screen, and by a process of >>> user choice / UI movements etc, the data will get filled in, and >>> subordinate CLUSTER archetypes will be chosen on the way, and >>> filled in along the way. >>> >>> This mode of operation is known by us in openEHR-land as 'dynamic >>> slot-filling' or 'runtime templating', as opposed to the more >>> typical design-time templating used in a lot of systems, where most >>> of the choices are made prior to runtime. But openEHR systems do use >>> runtime slot-filling as well, e.g. for writing discharge summaries >>> and referrals, where the data items are only knowable in the >>> encounter or report-writing session. >>> >> >> This trend allows me to discover that openEHR already became a rich >> ecosystem. Isn't this technique close from Gerard's vision of >> archetypes as "context for concepts" in a kind of ontology? >> >> However, I probably wrongly expressed what I wanted to say, and is >> more theoretical than comparing implementations, such as openEHR and >> Ligne de vie. >> >> When we talk to one another using a natural language, we just need a >> vocabulary and a grammar. The grammar is simply a set of rules, but >> not a physical pattern. We say "John sees the green house" and not >> "John as the subject sees as the verb the green as an adjective house >> as a noon in a position of direct object complement". >> >> In the same way, it is possible to express a structured discourse >> just using an ontology and a grammatical structure (say trees) >> without the need of any structure description. > > you are I think using 'grammar' in an unusual way - normally it means > the set of production rules that define legal utterances in some > language; this is an intensional definition, i.e. it can be used > computationally to parse actual utterances (including garbage) and > generate structures only for the utterances obeying the rules of the > grammar. > > In your usage, 'grammar' is what you call the trees, which are > extensional maps of legal utterances, or fragments of utterances, > which can only be connected together according to their rules, which > ensure correctness of larger utterances, e.g. a colonoscopy report. > > Structurally and computationally then, the fils guides (the trees in > Ligne de Vie) and archetypes are the same; they differ only in > representational details. However, there are two semantic differences. > > Firstly, the fils guides depend completely on the ontology (which is > an ontological terminology, to give it a more correct name, I think), > and the two things are built as a combined representational system. > Whereas elements in archetypes can have bindings made to ontologies > and/or terminologies, but don't rely on them, since they can rely on > their internal terminology to a reasonable extent (but not for value > sets like procedure or diagnoses etc). In theory, we should do what > fils guides are doing, and the reason we have not is only practical, > not theoretical: the development of bio-medical ontologies is still > young, and was almost non-existent 18 years ago when we started on this. > > The consequence has been that it is possible to construct archetypes > that say questionable or even wrong things with respect to ontologies > of those same things, say anatomical relationships. This rarely > happens in reality simply because archetypes are built by clinical > professionals and reviewed by many others, and mistakes tend to be > avoided, or if made, caught in review. But clearly it's not a > completely reliable strategy, and future versions of archetype tooling > should force live checking against suitable ontologies to detect > errors. Unfortunately, we still don't have such ontologies in anything > like the necessary detail - despite the existence of OGMS, and > numerous specialist OBO ontologies. SNOMED CT doesn't come close to > what is needed here. We still lack a comprehensive ontology covering > all of general medicine. > > Secondly, the 'utterances' represented by archetypes are not intended > to be directly linguistic expressions, but rather constitute an > underlying structural /reference/ 'terminology'. The fils guides on > the other hand express natural language utterances, i.e. they are like > a structural /interface/ terminology. With archetypes on the other > hand, the names of elements are used as a default to name fields etc > in the UI, but may always be overridden by some interface vocabulary, > or more likely a layer of language-level templates tied back to > templates based on archetypes. In openEHR we have no system for doing > the latter at the moment, although it is often mentioned as a nice idea... > > - thomas
Actually, I don't think that I use grammar in an unusual way. If I do it technically, lets assume for the sake of the discussion that I am really talking about a grammar, ie a set of rules that allows you to interpret an arrangement of concepts as a discourse. Typically, a dependency grammar is not just a tree representation, but a tree representation where you take as a rule that the sons of an element qualify this element. Since every natural language sentence can be represented as a dependency grammar tree and vice versa, it is possible to assert that a dependency grammar is a sufficient grammar. My point is that you have an ontology (say a terminology with terms grouped as concepts and concepts interrelated in a semantic network) and a true grammar, then there is no need for a "structural terminology"... one of the reason being that (part of) this terminology can find its place in the ontology. The first advantage is that practitioners can freely "tell whatever they want" in a structured way. For example using a tree interface with, for example, only the first elements already in place (say "encounter" as tree root and SOAP entries as sons). But it doesn't seem as the best interface for fully deterministic cases and archetypes (in their most basic meaning, ie flexible information schemas) are fit. Fil guides are used "in-between", as a way to help users fill trees with proposals of the kind "from the path you are currently located, you may benefit from this set of sons to carry your description one step further". I may elaborate on this. To sum it up, if you have a vocabulary and a grammar, you don't need anything more to tell things. In order to build systems that users can use, it is possible to use several techniques, such as "freely filling trees" (maybe guided by Fils guide) or using archetypes depending on the determinism level of what is to be told. Philippe
_______________________________________________ openEHR-technical mailing list [email protected] http://lists.openehr.org/mailman/listinfo/openehr-technical_lists.openehr.org
