Hi Pat, (01)
Thank you. I, for one, very much appreciate your taking the time
to state your world view(s). (02)
As someone not trained nor steeped in this material I continue to
look for coherence, pragmatism and common sense. I often find it
very helpful when you take the time to explain yourself in some
completeness. As mostly a lurker, I also appreciate source
references, with a preference for those available online. (03)
I would like to hear more of your views, especially about other
hot points for this forum including the semantic Web and Cyc, and
hot points in other forums such as http-range-14. (04)
To paraphrase that car bumper sticker: So many 'philosophers',
so little time. (05)
Thank you again. For what it is worth, I have not been
participating in this particular discussion, but insofar as I can
comprehend the issues I do agree with you. (06)
Mike (07)
Pat Hayes wrote:
> As I seem to be the only one in this discussion who agrees with me,
> maybe it would be useful to say what I mean directly, as I suspect my
> responses to other emails have been misunderstood.
>
> All these semantic issues seem to me to be quite simple and fairly
> obvious when one gets them clear, and should not require the use of
> words that honest engineers feel obliged to put into scare quotes. I
> think there is a sense among non-philosophers that when words like
> 'reality' or 'meaning' are used in technical discussions, that they must
> hide some mysterious, deep, even occult notions. But in fact, they mean
> just what they mean everywhere else. Good philosophy isn't trying to
> make things obscure or mysterious, its just trying to get things
> /clear/. So this message is a kind of beginners guide to my view of the
> general topic of semantics, and an attempt to shed some light on this
> apparently difficult topic; one which is, in fact, quite simple and even
> kind of obvious.
>
> BTW, nothing here is original. This is all absolutely standard textbook
> stuff. I learned it largely by reading Quine, but YMMV.
>
> ----------
>
> 1. First, we are talking about /semantics/, which is a fairly tightly
> circumscribed topic. Semantics is about the relationship between
> descriptions/languages/models/representations of something, and the
> thing or things they are supposed to be about: what they are
> descriptions/etc/ /of/. (I'll just use 'representation' and its
> cognates as a general word for the thing whose semantics we are talking
> about, everything from images to formal ontologies, though most of this
> is about the semantics of formal descriptions, in fact.) Semantics
> isn't about knowledge capture, or database management, or the Semantic
> Web, or about the future of Western Science, or any of that grand stuff,
> though it might well be relevant to those topics. Its just trying to
> analyze the relationship(s) between representations and what they
> represent. Or, as we say in the vulgar tongue, its about /meaning/.
>
> 2. In order to be useful, a semantic theory has to be able to cover the
> useful case, which is usually when the representation is supposed to be
> saying something about the actual world. That is, it isn't being used
> poetically or metaphorically or fictionally or whatever, just to
> represent something factual. Most representations are used in this way,
> and those that aren't seem to be parasitic on those that are (as when
> 'realistic' language is used in a novel to describe a fictional world),
> and certainly this is the interesting case for ontology engineering. So
> when I say that semantics relates a representation to reality, this is
> all I mean: that it relates the representation to what it represents,
> which in this case is, indeed, something real; which in turn simply
> means, not fictional or imaginary. I will call this the 'real case';
> that is, where the representation is about something real, and the
> semantic theory needs to deal with this.
>
> 3. Semantics talks about the relationship between two things: the
> representation itself, and whatever the representation is a
> representation of. But now, already, we have an issue, since we
> apparently need to say, in our semantic theory, what that second 'thing'
> is, and to do that in the real case, we apparently need a way to/ talk
> about real things/ in our semantic theory itself. But we have to use a
> language to do this talking in; and doesn't that beg the question? Well,
> yes, in a sense it does. If semantic theory were supposed to be able to
> /reduce/ meaning to something /else/, something different, something
> non-meaning-like, then we would seem to be stuck at this point in an
> intellectual trap. But semantics has a less ambitious goal. It is not
> reductive in this way: it doesn't set out to /eliminate/ representation
> and meaning in terms of something else, only to /describe/ them in ways
> that might be useful. (note1) And /of course/ it uses language to do the
> describing - what else could it do? - so in a way might be said to
> presume its own ideas in order to state them. Someone might adopt a
> skeptical position and say, /but your semantic theory can itself only be
> a description - a representation - which purports to describe reality,
> and so itself needs a semantics in order to be understood, which is
> circular/. And indeed, to repeat, if the goal were reductive this
> accusation of circularity might be a valid objection: but as the goal is
> only analytical, it can be simply accepted. Indeed, any semantic
> description - any description /at all/ - must in a sense presume a
> semantics, in order to be understood. That is inevitable, and should not
> be seen as a 'problem' or an 'issue' to be solved or avoided.
>
> 4. What it does mean, however, is that a semantic theory can be held to
> account in several ways.
>
> 4a. First, it should not /presume/ solutions to any semantic problems
> it itself purports to analyze. For example, to say that t/he symbol
> "is-a" means what "is a" means in English/ might be put forward as a
> small part of a semantic theory. But this is immediately subject to the
> objection that it tells us nothing, precisely because we don't have a
> satisfactory semantic analysis of the English phrase "is a". If someone
> feels that we do, or that none is needed, then they might find this kind
> of semantic analysis perfectly satisfactory, as Dick McCullough
> apparently does. But it does not meet the standards normally expected of
> a semantic theory precisely because of its immediate circularity: it
> presumes what it sets out to describe, and so provides no useful
> analysis at all. Apply it to English to see why. It would be a semantics
> for English which said: /"is a" in English means what "is a" means in
> English/. Which is not exactly false, but also isn't very useful.
>
> 4b. Second, and a related point, the semantics should use a way to talk
> about the second half of its topic - the reality described by the
> representation - in terms which are as semantically neutral as possible.
> That is, the semantic theory itself should make as few /presumptions/ as
> possible about the nature of the things being described by the
> representation. (note 2) This requires the semantic theory /itself/ -
> not the representation it is analyzing - to be minimal, which means to
> be very general, in its description of reality. It needs to make some
> assumptions in order to be a theory at all, but its own descriptive
> apparatus needs to be always cut to the very bone, as it were, and only
> use those terms or ideas which are absolutely necessary for the theory
> itself to function, no more. Part of the basic insight of semantics is
> knowing how to say enough but absolutely no more, how to get by with an
> absolutely minimal description of the reality-half of the relationship
> it is analyzing. Notice that /being described simply/ does not mean
> that the thing described must itself /be/ simple. If I describe someone
> by saying they are tall, I haven't said very much about them. But that
> does not mean that they do not have any other attributes.
>
> 4c. Third, the semantic theory should be correct when applied to itself,
> to the language in which it itself is written. This is the benign side
> of the circularity objection: at the very least, a semantic theory of
> how representations describe reality should be applicable to its /own/
> descriptions of reality (which, as I hope I have explained, it must
> somehow use, in order to be a semantic theory at all.) It should not be
> a counterexample to itself. Of course such a self-application will be
> circular, so would be subject to the first kind of objection, if put
> forward as an analysis of the theory's own ideas; but that is not the
> point. Rather, this self-application criterion is simply a basic test of
> the its own inner coherence, not a doomed attempt at an explanatory
> reduction of it to some simpler or more basic framework.
>
> 5. As I hope is now reasonably clear, any semantic theory must of
> necessity /use/ a representation (a description) to refer to reality. It
> - the theory - must be couched in a language, and so is itself a
> representation, of the kind that it itself purports to analyze. At this
> point we have rather a lot of representations to keep track of, so must
> speak carefully. There is the original representation that the semantic
> theory is about, and I will use this word "representation" to refer to
> this. It might for example be a formal notation of some kind, or an
> image, or a set of communication protocols, etc.. Then there is the
> descriptive framework in which the semantic theory itself is couched,
> which I will call the /metatheory/. The metatheory is the
> representational framework in which the original representation itself,
> the reality it describes, and the relation between them, is described.
>
> 5a. The question arises, how /complex/ must the metatheory be? Must it
> of necessity be more complicated than the representation, since it has
> to be able to describe all the reality that the representation does, and
> then some? But if the metatheory is more complicated than the original
> representation, have we not simply got ourselves into an endless loop
> requiring more and more complicated metatheories? No. Remember point (3)
> above: we are not trying to /reduce/ meaning to something else, only to
> analyze it. Our analysis can be couched in the same terms, and held to
> the same standards of precision and usefulness, as any other analysis;
> but that is all. It is not required to /explain itself away /- clearly
> an impossible task -/ /only to be useful. In this case, being useful
> means only providing some useful insight into the semantic relationships
> between representation and represented.
>
> 6. Given all the above, then, let me try to give a very quick sketch of
> how "Tarskian" model theory, the standard logical semantical framework,
> works. Here, the representation is a formal logical language, and we
> might as well take FOL as the example, since its pretty much the
> standard by which all others are judged. The basic idea is to say, for
> each 'meaningful' aspect of the formal representation language, what the
> world has to be like in order for this aspect to make exact sense; then
> to say exactly how the arrangement of the world determines the meanings
> of the expressions of the logic. The key idea here is that in our
> semantic theory we must be careful to say /as little as we possibly can/
> about the reality: /only/ what we need to say in order to make the
> semantics work, no more. This requires a little care.
>
> 6a. Look at the syntax. FOL expressions are sentences, which are
> supposed to be true or false. Sentences are made up from atomic
> sentences, boolean expressions (and, or, not, implies) which just tinker
> around with truth-values of smaller sentences; and from quantifiers
> (forall, exists). Take the /forall/ quantifier. Its supposed,
> intuitively, to say that something is true of /anything/. Any thing.
> What does this imply about the world being described? At the least it
> seems to assume that there are /things/, and that there might be more
> than one of them, and that a truth can be relativized to them, and it
> makes sense to talk about /all of them/. It doesn't seem to say or
> presume anything about /what/ they are, about their essential nature;
> only that it makes sense to speak of all of them. They can be
> /individuated/ from one another: if finite, they can be /counted/.
> Something might be true of this one but not that one. They comprise a
> /collection/ rather than a kind of amorphous cloud. The lot of them,
> together, can be described as a set, in fact. Set theory is just the
> tool we need here: it seems to capture exactly the necessary assumptions
> and no more. So our world must comprise at least a /set/ of /things/.
> Let us be modest and call this set the /universe/.
>
> 6b. Now consider an atomic sentence, say the sentence "P(a)". This is
> supposed, intuitively, to say that P is true of a; and (without going
> into too much detail) it is clear from the syntax that this 'a' has to
> be the name of one of those things in the universe-set. What does 'being
> true of' something really mean? Well, frankly, I don't really know; but
> in order to make a semantic theory I don't really /need/ to know. All I
> need is a /list/ of the things that P is true of, and then I can say
> that "P(a)" is 'true' if a is on the list and 'false' if not. So, in the
> spirit of minimizing our ontological commitment, we will say that a
> predicate like 'P' has to "mean" at least this list. It might have a
> much more complicated and mysterious other meaning, perhaps a 'real'
> meaning: no matter. As long as that other meaning can be used to extract
> this list, that is all our semantics needs; and indeed, it really does
> need this: it cannot make do with less, for then we would not know how
> to evaluate "for all". So this list it is. The real world must provide
> this much structure, and it need not (for our economic semantic
> purposes) provide any more: this is enough.
>
> So, here is our meta-description, in our semantic theory, of the reality
> represented by a FOL representation. It consists of a 'universe' set; a
> mapping from each logical individual name (like 'a') to something in the
> set; and of each predicate name (like 'P') to a list - a subset, as
> order isn't important - of the main set; and, although I havnt gone into
> it in detail, from each name denoting a binary relation to a subset of
> /pairs/ of members of the universe, and so on: in general, a
> /relational/ /extension/ - a set of tuples of things in the universe -
> for each relation symbol. (note 3) Put another way, the FOL semantic
> metatheory describes the real world as a /relational structure/; and for
> each world so described, it tells how to determine the truth-value (true
> or false) of any sentence in the FOL representation. It may be referred
> to as a model theory (the relational structure here being the 'model')
> or a logical semantics or a truth-functional semantics or a Tarskian
> semantics, after its inventor.
>
> 6c. Before proceeding, lets mention some of things this kind of
> semantics does /not/ do. It does not give any account of the meaning of
> anything outside strict FO logic. As an analysis of English meanings it
> is woefully incomplete and even plain wrong, since the English words
> 'or' and 'implies' and 'exists' have richer meanings in English than
> they do in FOL. It is entirely tailored to the syntax of this particular
> formal representational notation. None of this should be understood as a
> criticism of model theory, in my view. It does what it is intended to
> do. That it does not do more, is hardly good grounds for attacking it.
>
> 7. So, is the real, actual, world /in fact /a relational structure? Or
> is this just yet another /model/, another /representation/ which itself
> needs to be related to the /real/ real world by yet another semantic
> theory, this time one that applies to mathematical abstractions like
> relational structures instead of to formal logic? John Sowa has argued
> for this position many times. Let me call this idea the /semantic black
> hole/, for reasons which will become clear.
>
> 7a. Let us take this objection and ask how to overcome it. We would need
> a semantic theory for relational structures, one which relates them to
> the reality they represent. This second semantic theory will, in its
> turn, need to be able to describe this reality in a
> meta-meta-representation. Surely, then the same objection will apply to
> this second semantic theory: it must describe the world somehow, but
> whatever terms it uses to do so - whatever ontological framework it
> adopts - can be called yet another representation, not the real, real,
> real reality itself. We will /never/ get to the ultimate quiddity, the
> essential rock-bottom reality, this way. To insist that any description
> of reality must really be a description of another representation,
> amounts to a rejection of the entire semantic enterprise. To adopt this
> stance makes it impossible for any semantic theory to ever be stated:
> for as I hope I have made clear, /any/ semantic theory /must/ use some
> representation to describe the reality which is at the pointy end of the
> semantic arrow the theory purports to analyze. (Note 4)
>
> 7b. In any case, the semantic black hole objection is simply wrong.
> Model theory is stated using a metalanguage, of course, and the reality
> that the FOL representation is about is described in this metalanguage,
> of course, one that makes certain minimal ontological assumptions about
> reality, of course. But this inevitable fact does not imply that the
> reality so described is not real. Mathematical language is used all the
> time to describe real things: engineering and science would be
> impossible else, not to mention such mundane activities as sharing out a
> packet of candy between a group of children. So, to say that the real
> world being described by FOL is a relational structure, is not to say
> that it is some kind of Platonic abstraction, only that it is presumed
> by this description to have a certain minimal amount of structure, and
> that this structure is what is semantically relevant for the purposes in
> hand. So, consider the following example of a semantic relational
> structure for a first-order language containing the predicate 'P': the
> members of the universe are four pennies lying in a row on a table, and
> 'P' denotes the property of being heads-up. This is a relational
> structure made of pennies. Or consider, more ambitiously, the universe
> containing all sodium atoms in a person's body at a certain time, and
> 'P' is the subset of these that are sodium-24. Or, the universe
> comprising all the rivets in the Eiffel Tower, and 'P' the property of
> having been replaced at some time after the end of the second world war.
> You get the idea. All of these are relational structures in the required
> sense. The /general notion/ of relational structure is defined
> mathematically, using the language of sets, but examples of such
> structures - pieces of the world that satisfy the meta-description used
> by the FOL semantic theory - are all around us. The semantic theory does
> not relate FOL to another kind of representation: it relates FOL to
> something real, /using/ a (meta)language to /describe/ that reality
> (note 5).
>
> 7c. Replies to some common objections.
>
> 7c1. "This reply fails to address the main issue, which is that one can
> always think of the relational structure as a mere abstraction, without
> changing the truth-values assigned by the semantics at all. This
> semantic theory fails therefore to some to terms with something basic in
> reality-talk. According to this semantic theory, all of FOL /might as
> well/ be about pure abstractions. And this is revealed when people give
> examples of Tarskian models purely abstractly, which they often do."
> * A:* Indeed, they do, and one always can. However, this tells you
> something important about FOL: there is no way in FOL itself to be
> /certain/ that it is about something definite in the real world. In
> particular, FOL does not support /proper names/: symbols which are taken
> to be 'fixed' to something actual thing in the world and always identify
> that particular thing, like place-names or personal names. No amount of
> FOL axioms will ensure that "Cymri" means Wales, since if those axioms
> are satisfiable at all, they are also satisfiable in an isomorphic
> universe which contains no countries at all, only symbols. And this is
> not a weakness or limitation of the semantic theory, but rather a very
> basic weakness of FOL itself, one which is revealed by the semantic
> theory, and which it establishes beyond all possible doubt or argument.
> This is one major success of the semantic theory, in fact, to reveal the
> expressive limitations of the formalism. (BTW, in case you think this
> point is trivial, just check out the ongoing explosion of confusion
> surrounding the so-called "http-range-14" issue in the W3C Web
> Architecture (TAG) Group archives.)
>
> 7c2. "This semantic theory is vacuous and useless because it so obvious.
> It just says things like '/and/ means and' and '/forall/ means for all',
> but dressed up in fancy mathematical language."
> * A:* Well, yes, it is kind of simple, I agree. Quine discusses this
> point at length in his wonderfully written introduction to logic, where
> he treats logical semantics as a kind of careful translation into the
> meta-language. The semantics only works because we know what '/and/' and
> '/for all/' mean already and can use them, or some re-phrasing of them,
> in the meta-language itself. But still, it is not quite so circular as a
> direct translation, and it has some surprising consequences as a result.
> Logical beginners are often taken aback by things like DeMorgan's laws
> or material implication, which follow quite simply from the semantics
> but are hard to understand if all one has to fall back on are intuitive
> translations. And certainly, important results such as the compactness
> and completeness theorems are not at all obvious from a mere
> re-rendering of the logic into simplified English. The fact is, that FOL
> is kind of /half-intuitive/: it rests of course on an intuitive
> foundation, but it has surprises as well, and can extend the scope of
> reasoning far beyond what any human brain can achieve with unaided
> intuition. The simplicity and precision of logical semantics is a vital
> support in following these extended logical riffs: one simply cannot do
> it by trying to 'read' the logic directly. Four or five layers of nested
> quantification is beyond the mental attention span of the best human
> thinkers.
>
> 7c3. "Since the meta-theory uses mathematical language to describe
> 'reality', it is saying that the world is a mathematical abstraction,
> which is wrong. Mathematics only describes Platonic abstractions, not
> real things."
> *A: *(See earlier posts for more detail): No, the use of mathematical
> language does not, fortunately, imply that one must be talking only
> about mathematical /abstractions/.
>
> 7c4. (same as 7c3 but expressed differently): "But that semantics says
> that the real world is a relational structure (a 3-tuple/ an algebra/
> etc.). Surely you can't really mean to claim that reality /really is/ a
> relational structure (a tuple/algebra/etc/)??"
> * A: *Yes, I do really mean that. But it doesn't have the dire
> implications that you seem to think it has. Look at the four-pennies
> example again.
>
> 7c5. "There is something more basically wrong with your model theory,
> because it /assumes/ that the actual world /can/ be described as a
> relational structure. Maybe this is just metaphysically wrong: maybe
> reality isn't comprised of separate 'things' which stand in definite
> 'relations' to one another. What justifies this assumption, and what
> happens to the semantics if it is false? "
> *A:* Good question, and thank you for not mentioning quantum theory. If
> that assumption really is false, then this semantic theory simply does
> not fit the real world, as you say. But the moral would not be that the
> semantic theory is wrong, but that /FOL itself /is wrong. We began with
> FOL. These assumptions about the nature of reality were the minimal
> assumptions we were obliged to make in order to make sense of the FOL
> representational formalism itself. It is the representation, not the
> semantic meta-theory, which makes this very fundamental ontological
> assumption about the basic metaphysical structures of the real world. If
> reality really isn't made up of things standing in relationships, then
> FOL simply cannot be used to describe it. The semantics is the slave of
> the formalism at this point. The job of the semantic theory is to
> expound how the representation can describe a world. If it can't, and if
> this is revealed by the semantic theory, then that theory has done its
> job. Don't shoot the messenger.
>
> 8. Notice how the semantics was constructed by trying to capture
> minimally what the reality had to be assumed to be like, in order for
> the logical sentences to make sense, given their intended intuitive
> meanings. It begins with these intuitive meanings and uses them to guide
> the semantic construction; and the meanings it then provides are indeed
> very close to those original intuitions; perhaps, some feel, too close
> (cf 7c2 above). This means, though, that the 'formal' semantical
> meaning, and the 'intuitive' meaning, of a FOL expression are not
> sharply at odds with one another. Detractors of model theory often seem
> to assume that it gives a different, alien sense of meaning which is
> best ignored by ordinary users of logic, who simply 'read' the logic and
> understand it intuitively using plain old English tags or intuitions
> rather than this graduate-school-mathematics stuff. The graduate-school
> and the great-unwashed versions of logical understanding should mostly
> match up, however, and indeed they mostly do. The exceptions are always
> cases where the English-based intuition presumes too much of the
> relatively weak FOL expressivity (hence the puzzlement about the
> 'paradoxes' of material implication, such as a contradiction implying
> everything). Logical introductory text books are replete with exercises
> and cautions against these cases, and users of logic-based formalisms
> all acquire a grasp of meaning which corresponds more closely to the
> model theory, which is in fact /provably/ the appropriate one for FOL
> reasoning (note 6). But my point here is only to emphasize that formal
> semantics and intuitive semantics are not engaged in a battle for the
> soul of FO logic. The former is simply a more carefully expressed
> re-statement of the latter in terms which allow for exact analysis to be
> done and theorems to be proved.
>
> 9. OK, that's enough for now. TGIF.
>
> Pat
>
>
> --------
>
> Notes
>
> (1) The fact that logical semantics ("model theory") uses set-language,
> and has often been involved in questions arising in the foundations of
> mathematics, seems to have produced the idea that logical semantics is
> motivated by the same reductive concerns as those that exercise the
> foundations of mathematics (FOM). This impression is completely false.
> There has indeed been a huge effort in FOM, lasting nearly a century, to
> reduce all of mathematical language to pure set theory, where the whole
> universe (in this case the whole Platonic pure-mathematical universe,
> starting with the natural numbers) is re-built using constructions in
> set theory, so that the number three is defined to be the set {{} {{}}
> {{}{{}}}}, following a clever recursion in which zero is the empty set
> and every number is the union of the immediately preceding number with
> the union set of itself. (FOM is nothing if not ingeniously clever.) The
> ultimate purpose of such a reduction is to achieve some kind of
> guarantee of internal consistency for the mathematical edifice; to show
> that if at least set theory is consistent, then so is everything else.
> But this desperate search for some guarantee of mathematical consistency
> is not /intrinsic/ to set theory, which is simply a very general theory
> of abstract collections, and can be applied to almost anything. It is
> useful precisely because it is so very general and so simple. Model
> theory is one (rather elementary) application of set theory, and
> building a secure foundation for all of mathematics is another. They
> have virtually nothing to do with one another, apart from both using the
> same tool-kit. In particular, the fact that logical semantics is phrased
> using the language of sets does /not/ imply that it is a reduction of
> reality to some pure-set-theoretic construction. This common fallacy is
> often repeated, even by philosophers such as Barry Smith who should know
> better.
>
> (2) Actually this should be stated more carefully, as some
> representations are intended to have limited scope and only used to
> refer to particular kinds of thing or aspects of reality, e.g.
> Labanotation, and in such cases it is obviously OK for the semantics to
> make matching presumptions. But for general-purpose representations like
> most natural and logical languages, the point holds.
>
> (3) This is classical FOL, not Common Logic, which because of its more
> liberal syntax requires some modifications to the way the real world is
> described in the semantic metatheory.
>
> (4) There is one possible way to stop this regress, by insisting that
> the 'ultimate' semantics is stated entirely using demonstratives rather
> than descriptions. Instead of saying, <syntax> means <metadescription>,
> one says <syntax> means /this/, accompanied by a gesture towards an
> actual thing or part of the real (physical) world. For one the most
> telling critiques of this idea, see Gulliver's Travels, Book 3 (the
> Island of Laputa).
>
> (5) In the real case, that is. To be more exact: to something that /can/
> be real. It can also be an abstraction or imaginary, of course, which is
> appropriate since the formalism FOL itself can be used to describe
> abstract or imaginary things just as well as real ones.
>
> (6) The famous Goedel completeness theorem.
> (08)
_________________________________________________________________
Message Archives: http://ontolog.cim3.net/forum/ontolog-forum/
Subscribe/Config: http://ontolog.cim3.net/mailman/listinfo/ontolog-forum/
Unsubscribe: mailto:ontolog-forum-leave@xxxxxxxxxxxxxxxx
Shared Files: http://ontolog.cim3.net/file/
Community Wiki: http://ontolog.cim3.net/wiki/
To Post: mailto:ontolog-forum@xxxxxxxxxxxxxxxx (09)
|