Dear Pat,
Now we need a simple yes or no from you regarding your participation at
Vespucci between June 10 and 15. The week needs to be advertised asap, and
we will need to opt for a replacement for you by the end of this week.
Please remember that, if you accept, your presence during the entire week
would be very welcome, but is not a condition.
As for the paper, I had to return the manuscript to the author, who would
still very much welcome the missing review.
Thanks for your understanding.
Kind regards
Werner
Werner, I am sorry that you do not seem to be getting my emails. If you do not get this I do not quite know what I can do. I sent you the review some time ago, and declined your offer (with regrets). Here is the email I sent you on 11/15:
----------
Dear Pat,
Sorry that I have to bug you again – we need to make a decision now regarding your participation at <http://www.vespucci.org/>www.vespucci.org next June (10-15). Please let me know very soon what your possibilities are, if any.
Yes, of course. I think that I will have to turn you down, unfortunately. The issue is that, for a variety of reasons, partly personal and partly professional, it is rather tricky for me to plan this far ahead for a trip abroad at the present time. I am honored and flattered by your invitation (which is why I havered so long trying to find a way to accept it), and sincerely wish I could be there; but I already have had to regretfully withdraw from one workshop at the last minute, after accepting his invitation, and I do not want to put your schedule at risk.
All the best,
Werner
p.s. it would also be nice to receive your opinion on the Frank paper (though we may decide to reject it, but would like to give the author a full set of reviews)
Now I am confused: I sent you a review of Frank on 11/1/06. Did you not receive it? In case not, it is appended.
-----------------------------
DISTINCTIONS—A COMMON BASE FOR A TAXONOMIC
CALCULUS FOR OBJECTS AND ACTIONS
Andrew U. Frank
1 Is the subject of the article within the scope of the special issue?
yes , but see comments
2 Is this a new and original contribution?
no. See comments
3 How would you rate the overall scientific quality of the paper?
substandard (does not meet standards of strong national J's or
conferences)
4 What is your overall recommendation for publication?
__ Reconsider after major revision OR reject.
5 Please summarize the major contributions of the article followed
by any concerns or suggestions for improvement that you may
have. Please be as specific as possible in your comments. Points
to consider in your review are:
- Is there a clearly identified problem?
No.
The article can be read in several ways: as an attempt at an overview, summarizing and synthesizing some general ideas; as a broad proposal for an organized research effort; or as an exegesis and defense of a new idea. It is inadequate on any of these readings, but for slightly different reasons.
Nothing in the paper is new. The general ideas of taxonomic lattices, including a full mathematical theory, are well-known, and indeed the basic idea of using the presence/absence of 'qualities' to structure concepts was one of the earliest ideas in cognitive psychology, and has been thoroughly explored by, for example, Quillian, Collins and Loftus, Smith and others. (I have not traced it back to its original source, but the basic idea can be found in the early work of the neural associationists, for example Henry Head. A reasonable history can be found in any good textbook; a quick Google search turned up this summary: http://io.uwinnipeg.ca/~epritch1/ltmtheories2000.html) Not surprisingly, this extensive body of work has also revealed many problems with such a simple model of mentality, none of which are discussed in the paper. Many concepts - the paradigm examples are colors - are known empirically to not fit this model. (The author seems aware of this in his citation of Rosch, but does not acknowledge its importance as a refutation of the theory being presented: in fact, he uses 'red' as one of his examples, most unfortunately.) There is a huge body of work on, for example, rapid negative judgements, default overrides (birds fly, but penguins are flightless birds), typicality judgements, etc.., all of which strongly suggest that this naive model is simply wrong, or at best must be treated as a gross oversimplification of true mental structure. On a philosophical level, the entire notion of characteristic features for concepts was strongly refuted by Wittgenstein; one can disagree with Wittgenstein, but one should not ignore him. (The author cites Gibson frequently without apparently being aware that Gibson's theory also rejected this model.)
The author uses the earliest and most standard account of action and change, in terms of word-states, but makes no reference to the huge body of literature on giving ontological descriptions of processes and change, dating from the beginnings of AI, some of which is now part of internationally agreed standards for information exchange. He also seems to be unaware of the known problems of this naive model - which is often loosely referred to as the situation calculus, although that term also has a more exact meaning - such as the notorious frame problem (how to compactly describe the large amount of the world that is not affected by an action) and the less well-known but technically more challenging persistence problem (how to adequately describe on-going changes and non-changes). As he notes, to think of actions as changes of state is a natural extension of ideas used in describing computation; but he fails to note the fact that the actual world is importantly unlike the state of a computer, in particular that it has ongoing processes and events which take place concurrently with our actions, in some cases are caused by them, and affect their outcomes, all of which make the simple state-change model inadequate.
The text is careless and internally inconsistent in many places, using complex technical words without explanation and changing the meanings of its own technical terminology without warning. I give some examples in the detailed commentary below. Important and tricky ideas, such as 'rules' for internal consistency, and 'situations' in the world being conceptualized, are introduced without warning or explanation, and then used as though they were an accepted part of the overall account. Much of the text, particularly in later sections, is vague and sketchy to the point of being trivial. For example, section 7.9 notes that distinctions can be made by introducing distinctions, which is both trivially obvious and also trivially uninteresting as a theoretical observation, as the example given there suggests: the distinction between skipping and hopping, for example, is not given any elucidation by simply adding skip+ and hop- 'parameters' to one's world model; particularly if one takes the author's own Gibsonian view seriously, that such distinctions must ultimately be rooted in perception. The last paragraph of section 8 would not be acceptable in an executive summary of a research proposal.
The exposition is also careless in not distinguishing theory from fact. For one example - there are many throughout the paper - section 8 asserts that "A human constructs ... a mental model, of a situation, which is in terms of (very fine) taxa." But there is absolutely no justification for the claim that actual mental models consist of taxa, and indeed many reasons to presume they do not. Several claims are made without justification or supporting argument, in some cases apparently in opposition to other claims elsewhere in the paper. For example, "A taxonomic lattice allows mental inferences" (section 9.1): no such inference method has been suggested, and the taxonomic model was first contrasted with "complex, model-based deductions". It is known that simple lattice models do not support many intuitive inferences. "This is an economic and powerful method to build concepts" (section 9.2): no such method has been described. If the author were to attempt to implement his lattice-multiplication process, as others have done - see for example the work of John Sowa on lattice models for ontologies - he would quickly discover its limitations.
As an overview or exegesis, then, this paper lacks the necessary scholarship on several fronts, is poorly written, and is quite inappropriately and naively uncritical. It does not propose any new lines of research, and the technical exposition does not contain enough detail to evaluate it as a proposal.
Detailed comments on the text follow. These only extend through the first few sections, as a detailed commentary on the full paper at this level would take too long.
Section 1 paragraph 2. The historical point is tendentious. Although Linnaeus' botanic taxonomy is still in use (though much modified), this is largely due to his considerable botanic insight (especially its focus on plant sexual characteristics) than from the fact that it was a taxonomy as such. His taxonomy of human racial character is no longer in use, and arguably did the world a great deal of harm.
S1 pa3. "... dynamic aspects of reality, i.e. a classification of actions..." The identification of dynamics with action is naive and incorrect. Most of dynamics of the perceived world arises from events which have no agent, rather than from actions taken.
S1 pa3 list item 1. The statement is incorrect. Many, perhaps most, formal ontologies do not take (natural) language as a starting point.
item 3. While some ontological taxonomies are 'tree-like', many are not. Most modern ontology frameworks, including all the semantic web standards, allow unrestricted taxonomic structures, which are widely used in scientific ontologies, see for example the complex subclass structures in the SWEET ontologies (http://sweet.jpl.nasa.gov/ontology/)
S1 pa4+5. This seems inconsistent. If the focus is on physical reality (first line p4), why is the interest in mental concepts (last line p4)? And if it is on physical reality, why are psychological data of interest at all? It is important to bear in mind that psychological models of physical reality are not themselves accurate models of physics. Again, surely the concepts expressed in linguistic units are indeed mental concepts, so what exactly is the final distinction supposed to mean? And how is this negative claim to be reconciled with the last sentence of the next paragraph, which explicitly says that to use words in this way is 'viable'? None of this seems to have been thought through properly.
S1 Pa 7 'much of ontological reasoning can be achieved...'. This is a strong claim which is not justified by anything in the paper. Reasoning is nowhere addressed.
S2 pa 2. You need to say what you mean by 'object'. Is the space surrounding the bread an object? Is the surface of the bread an object? Do objects in your sense retain their identity through time, so that it is the same loaf of bread before and after the cut? If so, say so clearly. If so, by the way, at what point in the cutting does the loaf cease to exist? Ontological frameworks need to get things like this clear. You say the changes are only in the volume and mass of the loaf, but other things have come into existence, e.g. the loaf has a brand-new surface (whose texture is likely different from the older one.)
There is a huge literature, some of it essentially philosophical, on such matters. While it is not appropriate to survey this all here, the exposition should show some signs of being aware of it, or at least of the main issues that arise in philosophical ontology.
S2.3 pa 1. You mix up mass terms (butter) with object terms (table). This is incoherent, as mass terms do not denote either objects or classes of objects. If you mean, eg, 'piece of butter', then say that. Mass nouns do not indicate classes of objects in the usual sense of 'object'. If you mean to refer to mereological sums, say so.
S2.3 pa 3. Linguistic accounts do not allow a reader/hearer to 'reconstruct' the action, since they rarely contain enough detail. What do you mean by "context" here?
S3 pa 3. What do you mean by "situation" ? Are you alluding to the notion as exemplified in Barwise & Perry's 'situation semantics'? (http://www-csli.stanford.edu/~john/PHILPAPERS/sitsem.pdf) If so, say so. If not, clarify. What evidence do you have that linguistic accounts are less detailed than a mental model?
S4 pa1. This is confused. First, it is not at all clear that human conceptualizations correspond to sets (in the mathematical sense used here). Second, even if we ignore this complication, it is quite unclear that these sets are defined by a similarity criterion; in fact, it is known that in some cases they are not (color names); thirdly, even if they are, it is still not clear that these similarity criteria are perceptually salient, as earlier text in the paper seems to assume. And finally, even if all this is ignored, it is mathematically incoherent to claim that conceptualizations are sets and also that they are context- or situation (see above) -dependent. Sets have no contextual or situational freedom: they are mathematical structures.
S4 pa 2. Gibson is mis-characterized here subtly but importantly. He did not refer to 'objects'. For example, it is not the open doorway, but the space it frames - the lack of an object - which affords walking. To return to an earlier question: is that space on 'object'?
S4.1 pa 1. Properties now apply to "objects, situations or actions" ; previously they applied only to objects. Have properties grown, or are situations and actions considered to be kinds of object? (This is typical of the sloppiness which pervades the exposition in the paper. I do not have time to point out all the places where definitions are given and then mis-used.)
pa 2 "nouns and verbs describe intensional sets of.." Why "intensional' here? What distinguishes an intensional set from other kinds of set? (In the standard usage in formal philosophy, 'intensional set' is an oxymoron.)
pa 3. Why do equivalence classes yield a 'small number' of discrete values? It is trivial to define an equivalence class which yields a set of values of any cardinality, from small finite to uncountably infinite.
S 4.2. This discussion is confused and ambiguous. Does each taxon correspond to a single distinction/value pair, as in the Solid example? Or can a taxon have specific values for several distinctions? In the third sentence, is d' the defining value of the taxon, or any function? Must d' be a distinction, or any function? What if the elements of the taxon do not have a common value for that function (which could occur even if it is a distinction)? Why is there a biconditional in the formula corresponding to figure 2? The text implies only an implication.
I suggest re-writing this entirely.
S4.3. The usage of intensional/extensional here is non-standard. This is OK, but you should say so explicitly, and draw attention to the difference.
Sections 4.4 and 5 are at odds. 4.4 says that Boolean values are 2-valued: section 5 says they are 4-valued. Please be consistent.
The 4-valued logic shown in figure 3 uses 'top' and 'bottom' in exactly the opposite way to the standard usage, as in the cited Belnap work et. seq. , and also in theory of computation. Bottom is usually taken to mean 'no information' or 'undefined', corresponding to a tautology (A or notA), and top to a clash, error or contradiction, (A and notA). The choice is arbitrary, but you should draw attention to the nonstandard usage. The term 'semilattice' refers to the case without the conventional 'top': you would be obliged to treat this as an upper semilattice.
Section 5.1 pa 1. The centrally important notion of a 'product' of two lattices is here used without a shred of explanation of exposition. As this is the first mathematically non-trivial concept in the paper, some explanation is needed.
Section 5.4. Here "rules" appear without warning or definition. As the presence of these 'rules' has a huge effect on the theory, more exposition and careful definitions are necessary. A lattice with dependencies added is no longer a lattice, for example, so a large swathe of standard mathematical results no longer apply.
There is a casual remark here: "One could cast the semantic net ..somewhat wider, ..." which illustrates the way in which this entire project fails to address the central problems it is supposed to be facing. The same phenomenon occurs with any formalized framework put forward to encode intuitive propositions: the formalism under-determines the way it is to be used. And different such ways of course produce incompatible ontological frameworks. This central problem, which arises throughout all formal ontology work, is nowhere tackled in the paper, or even acknowledged as an issue. But contrast this with Linnaeus's taxonomy: instead of a merely mathematical technique which could be cast wider or narrower, he produced a set of criteria which were focused on the actual structural distinctions which had real meaning in his chosen field: the sexual characteristics of plants, rather than, say, their color or leaf shape. It was the fact that his notation could not be arbitrarily cast in any one of many ways which made it so useful and long-lasting.
Section 6. The 'taxonomy of relations' described herein is well-known in programming language theory as a type or sort structure. These have been thoroughly studied and their properties are well documented. However, in realistic ontology work it is very unusual to find relations which admit of clear-cut typing in this way. The relation indicated by the English word 'on', for example, does not, probably because it is in fact a complex of different relations, with various disparate conditions on their arguments. For more on such complexities, see Melissa Bowerman's work on spatial prepositions.
The citation of Gibson in the second paragraph is inappropriate: this is not what Gibson asserted.
S7 pa1. What is "JEPD" ?
pa 2. What is meant by "conceptual unit"? How does it differ from "concept"?
You assert that actions "change the state of objects in time". Do you mean that actions map from one state of an object to another, later, state of an object? Must it be the same object? (How then do objects come into existence or cease to exist?) Are these the only changes? Does the evolution of the world-state consist entirely of these action-mappings on object states, or can there be other changes happening? (If the latter, how do you describe them? If the former, how would you describe the action of pushing something off a high shelf, and then doing nothing to catch it or prevent it from falling? Or of locking a door to prevent it being opened? Or of lighting the fuse of an explosive, then retreating so as to be out of harms way?)
You say that the world is the "totality" of objects, and that its states are the "sum" of their states. What do the quoted words mean, exactly? Do you mean 'sum' in the merological sense? Given a set of individual object states, can there be more than one world which is their sum? If so, what distinguishes the different sums? If not, how do you distinguish A being on B from B being on A?
I could go on asking questions like this, but the point has probably been made. You need to describe your action/change model in a lot more detail, with more care and precision, and show how it differs from the multitude of other such models that have been proposed, and in some cases analyzed in enormous detail.
-------
=====CONFIDENTIAL============================CONFIDENTIAL=====
6 Are there comments about this article that you would like to
share with the editors of the special issue in confidence?
I apologize for the rather undiplomatic tone of the review, but this paper really is completely amateurish.
Name Pat Hayes
Date 1 NOvember 2006
--
---------------------------------------------------------------------
IHMC (850)434 8903 or (650)494 3973 home
40 South Alcaniz St. (850)202 4416 office
Pensacola (850)202 4440 fax
FL 32502 (850)291 0667 cell
phayesAT-SIGNihmc.us http://www.ihmc.us/users/phayes