Fodor and Le Pore's dissection of JP's book is, and is intended to be, an attack on a whole AI approach to natural language processing based on symbolic representations, so it is open to any other member of that school to join in the defence. IRS has its faults but also some technological successes to show in the areas of machine translation and information extraction (e.g. Wilks et al., 1993), but is it well-founded and philosophically defensible?
Many within IRS would say that does not matter, in that the only defence lexical or other machine codings need in any information processing system is that the system containing them works to an acceptable degree; but I would agree with those who say it is defensible, or is at least as well founded as the philosophical foundation on which FL stand. That is, I believe, one of the shakiest and most satirised of this century, and loosely related to what what Ryle (1957) called the ``Fido''-Fido fallacy: the claim that to every word corresponds a concept and/or a denotation, a view that has crept into everyday philosophical chat as the joke that the meaning of life is life' (life prime, the object denoted by ``life''2.
It is a foundation of the utmost triviality, that comes from FL
(op.cit., p.1) in the form:
(1) The meaning of ``dog'' is DOG.
They seem to embrace it wholeheartedly, and prefer it to any theory, like TGL, offering complex structured dictionary entries, or even any paper dictionary off a shelf, like Webster's, that offers even more complex structures than TGL in a form of English. FL embrace an empty lexicon, willingly and with open eyes: one that lists just DOG as the entry for ``dog''. The questions we must ask, though the answer is obviously no in each case, are:
The first of these points brings back an age of linguistic analysis
contemporary with Ryle's, in particular the work of writers like Lewy
(1964); put briefly, the issue is whether or not (1) expresses anything
determinate (and remember it is the mantra of the whole FL paper), or
preferable to alternatives such as:
(2) The meaning of ``dog'' is a domestic canine animal.
(3) The meaning of ``dog'' is a dog.
(4) The meaning of ``dog'' is ``domestic canine animal''.
not to mention
(5) The meaning of ``dog'' is ``dog''.
The two sentences (2) and (3) are perfectly sensible, depending on the
circumstances: (2) is roughly what real, non-Fodorian, dictionaries
tell you, which seems unnecessary for dogs, but would be more
plausible if the sentence was about marmosets or wombats. (3) is
unhelpful, as it stands, but perhaps that is accidental, for if we
translate it into German we get something like :
(3a) Die Bedeutung von ``dog'' ist ein Hund.
which could be very helpful to a German with little or no knowledge of
English, as would be
(2a) Die Bedeutung von ``dog'' ist ein huendliche Haustier.
To continue with this line of argument one needs all parties to accept the reality of translation and its role in argument: that there are translations, at least between close languages for simple sentences, and no amount of philosophical argument can shift that fact. For anyone who cannot accept this, there is probably no point in arguing about the role of lexicons at all.
Both (2) and (3), then, are sensible and, in the right circumstances,
informative: they can be synonymous in some functional sense since both, when
translated, could be equally informative to a normal fluent speaker of another
language. But (4) and (5) are a little harder: their translations would be
uninformative to a German when translated, since translation does not
translate quotations and so we get forms like:
(5a) Die Bedeutung von ``dog'' ist ``dog''.
and similarly for a German (4a) version of the English (4). These sentences therefore cannot be synonymous with (3) and (2) respectively. But (4) might be thought no more than an odd form of a lexical entry sentence like (3), spoken by an English speaker.
But what of (1); who could that inform about anything? Suppose we
sharpen the issue by again asking who its translation could inform and
(1a) Die Bedeutung von ``dog'' ist DOG.
(1a) tells the German speaker nothing, at which point we may be told that DOG stands for a denotation and its name is arbitrary. But that is just old nonsense on horseback: it implies that the English speaker cannot understand (1) either, since DOG might properly be replaced by G00971 if the final symbol in (1) is truly arbitrary. It is surely (3), not (1), that tells us what the denotation of ``dog'' is, in the way language is normally used to do such things.
DOG in (1) is simply a confidence trick: it is put to us as having the role of the last term in (3). When and only when it is in the same language as the final symbol of (3) (a fact we are confidently assured is arbitrary) it does appear to point to dogs. However, taken as the last term in the synonymous (1a) it cannot possibly be doing that for it is incomprehensible, and functioning as an (untranslated) English word, exactly as in the last term of (5). But, as we saw, (5) and (3) cannot be synonymous, and so DOG in (1) has two incompatible roles at once, which is the trick that gives (1) interpretations that flip uncontrollably between the (non-synonymous) (3) and (5). It is an optical illusion of a sentence.
In conclusion, then, (1) is a dangerous sentence, one whose upper case-inflation suggests it has a function but which, on careful examination, proves not to be there: it is either (case-deflated) a form of the commonsense (3), in which case it loses its capitals and all the role FL assign to it, since it is vacuous in English, or just a simple bilingual dictionary entry in German or some other language. Or it is a form or (5), uninformative in any language or lexicon but plainly a triviality, shorn of any philosophical import.
Those who still have worries about this issue, and wonder if capitalizing may not still have some merit, should ask themselves the following question: which dog is the real DOG? The word ``dog'' has 24 entries even in a basic English dictionary like Collins, so how do FL know which one is intended by (1)? If one is tempted to reply, well DOG will have to be subscripted then, as in DOG, DOG etc, then I shall reply that we will then be into another theory of meaning, and not one of simple denotations. My own suspicion is that all this can only be understood in terms of Fodor's Language of Thought (1975) and that DOG for FL is a simple primitive in that language, rather than a denotation in the world or logical space. However, we have no access whatever to such a language, though Kay among others has given arguments that, if anything like an LOT exists, it will have to be subscripted (Kay, 1989), in which case the role of (1) will have to be rethought from scratch. All the discussion above will still remain relevant to such a development, and the issue of translation into LOT will then be the key one. However, until we can do that, and in the presence of a LOT native speaker, we may leave that situation aside and await developments.
The moral for the rest of the discussion, and the role of IRS and TGL, is simple: some of the sentences numbered above are like real, useful, lexical entries: (3) is a paradigm of an entry in a bilingual lexicon, where explanations are not crucial, while (2) is very like a monolingual lexical entry, where explanations are the stuff of giving meaning.