Applied Ontology (IOS Press)

REVIEW FORM

EB Member in charge: Christopher Menzel

Review done by: Christopher Menzel

Affiliation: Philosophy Dept, Texas A&M University

If you had any trouble using this review form, please identify here which questions raised a problem and what kind of problem they raised:

------

Paper number: AO-2006-94

Title of Paper: “Semantic, Classes, Ontologies and Taxonomies”

Author(s): Terje Aaberge

------

SECTION I. COMMENTS TO BE WITHHELD FROM THE AUTHOR(S).

(Please use this section only for comments that you want to be withheld from authors.

Use as many lines as needed.)

EVERYTHING BELOW THIS LINE WILL BE RETURNED TO THE AUTHORS

SECTION II. SUMMARY AND RECOMMENDATION

OVERALL EVALUATION - Overall quality is (check as appropriate):

Excellent / Good / Average / Weak / Bad
X

RECOMMENDATION:

Accept as it is
Accept, yet certain minor revisions should be made (see Section IV for recommendations)
Do not accept at this stage. Author(s) should prepare a major revision (see Section IV) and resubmit the new version for a second and final review.
X / Reject
Inappropriate for JoDS - Submit to another journal (suggested journal: )

SECTION III. OVERVIEW

A). CONTENT

a). How would you rate the technical quality of the paper?

Excellent / Good / Average / Weak / Bad / Don’t know
X

b). How would you describe the technical depth of the paper?

Too technical: Requires expertise in the domain / In-between / Right:
Is precise and rigorous, yet understandable by non-specialists of the domain / In-between / Too Low:
Remains superficial, making technical assessment dubious
X

c). Does the paper make a tangible contribution to advancing the state-of-the-art in its field?

Yes / Yes, but limited / No
X

d). Is the bibliography adequate?

Yes / Yes, but see Sec. IV / No
X

B). PRESENTATION

a). Is the abstract an appropriate and adequate digest of the work presented?

Yes / No
X

b). Does the introduction clearly state the background and motivation in terms understandable to the non-specialist?

Yes / Only to some extent / No
X

c). How would you rate the overall organization of the paper?

Satisfactory / Needs improvement / Poor
X

d). Relative to the technical content, is the length of the paper appropriate?

Yes / Too long / Too short
X

e). Is the English satisfactory?

Yes / No
X

f). Disregarding the technical content, how would you regard the quality of presentation?

Excellent / Good / Average / Weak / Bad
X

SECTION IV. DETAILED COMMENTS

(Please explain your suggestions and comments to help the authors understand what they could do to improve their submission.)

The author introduces two metalinguistic frameworks for characterizing two classes of rather weak, interpreted first-order languages, one for individuals, one for properties. Certain languages from these two classes (presumably ones where the predicates of the individual language are, or are correlated with, terms of the property language) can then be juxtaposed into a third type of language, a description language – the motivation here apparently being that such languages are similar in structure to the languages of description logic.

The framework characterizes the languages in these classes in terms of various metalinguistic maps, e.g., from objects in the domain to their names, from names to the predicates they are joined with to form facts about the object named, from sentence names to the truth values of the named sentences.

I had a number of questions about the technical details here (see below), but my more global concern was that I simply could not understand the motivation for the author’s framework. The framework makes explicit some rather obvious maps that are implicit in any semantics for a first-order language. And while I found the connections between these maps as indicated in the various diagrams appealing, they were much too obvious to be particularly revelatory. Moreover, the languages characterized by the individual and property languages are simply generic monadic predicate languages, and the juxtaposed description languages are essentially just 2-sorted languages with an instantiation predicate to indicate that a given property is instantiated by a given individual. As far as I could see, there is nothing technically novel here, nor could I think of any particular use these frameworks could be for doing ontology. They do not offer any additional expressive power over ordinary first-order logic – indeed, as noted, they are quite a lot weaker. Indeed, that they do not even have any binary predicates (save the implicit instantiation predicate) they are in significant respects less expressive than a description logic. Curiously, the author does not raise the issue of expressiveness at all.

The author’s emphasis on “observables” – i.e., atomic facts – suggests that atomic facts are particularly important or fundamental somehow. But, at least with regard to ontological engineering, many other types of facts not derivable from atomic observation are equally fundamental, e.g., negative facts (“The mouse did not eat the cheese”, “There are no wineries in Mississippi”) and quantified facts representing the implementation of policy (“Every automobile is shipped with a spare tire”). The author appears to be arguing for the importance of “observables” in Section 3 (“Operational Definitions”). But I found it hard to see the relevance of the discussion – which focuses on the complexities of observation – to either his metalinguistic frameworks or to ontological engineering generally.

The author applies his frameworks to ontological engineering in Section 5. But while the frameworks are able to capture a number of elements of ontological practice – property ascription, taxonomic classification – I could not see any practical or conceptual advantage to the analyses of the notions the author provides. (And I frankly found the chess game example of formal classification opaque. Perhaps it was just me, but after describing how to formalize the basic elements (e.g., pieces and positions on the table) the author jumped into the formal characterization without any reference back to these basic elements. Moreover, how the formalization represented an actual chess game was entirely lost on me.)

A couple of observations on the technical material. The definition of a map in the Appendix appears to identify maps with (extensional) binary relations – a map from A to B simply “associates elements in A to elements in B; that the author explicitly defines what it is for a map to be from A into B reinforces this impression. However, his notation entails that maps are functions only. Thus, he expresses that f is a map from A to B in the usual way as “f : AB”; but he also introduces notation to indicate that f maps a to b: “af(a) = b”, and this obviously implies that f is functional. Moreover, these functions all appear to be total.

Given this, the author’s object maps are rather puzzling. For instance, given an interpreted language, the author introduces a map  such that

 : D  N; d (d)

that maps the domain into the names of the language. Given the definition of a map as a total function on its domain, this seems to imply that every object in a domain has a unique name. That every object has a name, and that names are unique, are wholly untenable assumption for ontological engineering. Yet this very strong feature of the author’s frameworks goes completely unmentioned.

The framework also introduces a mapping  from names to predicates that are true of the denotations of the names:

 : N  P; n (n).

But again, the functional character of mappings seems to have unintuitive effects – in this case that, for each name, there can only be one atomic fact involving that name. And since from the previous map it follows that each object in the domain has exactly one name, it follows that there is exactly one atomic fact in the language about each such object. Again, this is a very peculiar. The author mentions it in Section 4.3, but does not motivate it at all. This will leave any reader accustomed to standard first-order logic very puzzled.

Finally, the author defines the function C in Section 5.2 on page 14. This function, applied to a single predicate, yields the objects in the domain to which the predicate applies. But the author begins using this predicate with two (and more) arguments subsequently this section. I think this is supposed to indicate the intersection of the set containing C applied each argument, but I couldn’t be sure. (Also, the ASCII diagram on page 14 is reproduced in the same proportional as ordinary text, and looks quite bad. At the least the author should use a nonproportional font like Courier; better still, a drawing program.)