It is my intention in this paper to take up where Professor Quine has left off in his essay on ontological relativity. In order to gain momentum and to focus attention on certain points in his observations, I will first summarize the issues raised there. It is also hoped that, if necessary, this recapitulation will resolve any differences in word usage. While it must be remembered that I will have different emphases from those of Professor Quine, the first part of the paper is intended only as a paraphrase of his points.

Radical translation, as one might expect from its name, is the most extreme, and hence the clearest, case of ontological relativity. We find that translational hypotheses for an unknown language are directly dependent on the world view that we ascribe to the native speakers of that language. This is so because we can interpret the language only in a relative sense: relative to some background theory of the world that we choose to ascribe to the natives. We can translate a particular word as "rabbit" only if we suppose that the natives view the world as consisting of objects unified in space and constant over time. Should we suppose a different view of the world in our native, we could just as easily propose alternate translations incompatible with our original conjecture. There is no way of finally deciding for any among the various alternatives.

A given set of analytic hypotheses for translation can be made adequate by an appropriate ascription of world-view and, alternatively, a given world view can be ascribed as long as we choose the right analytic hypotheses.

These observations are easily transferred to more familiar situations, for ontological relativity begins at home. In theory, the problem we confront in interpreting the remarks of another speaker of our own language are no different from those we face when attempting an obviously new language, although in practice we are not as aware of the difficulties. Reflecting on the ubiquitous nature of the problem we might expect some principle which gives some order to the communication game.

Quine mentions the principle of charity as such a rule. Its content may be formulated as a general position assigning a preference to an ascription of nonstandard usage of common terms over a "silly" world-view on the part of our interlocutor.

This is not a mechanical principle and it does not result in the ascription of a uniform world view to everyone we meet. It seems to come from a desire to minimize our interlocutor's silliness by locating it in his word usage rather than in his beliefs about the world, but it is certainly imaginable that in some cases he may come out better if we assume minor deviations in his view of the world rather than major eccentricities of usage.

In any case, we are confronted with a knotty and disconcerting problem. Our difficulties are obvious when we do a radical translation, for we are unsure how to begin. Do we prefer elegance in the language or rationalism in the world? Even if we could devise some general rule for deciding such extreme choices, who is to say that there is no third choice -- a happy medium -- which makes both the empirical and the linguistic efforts of the natives very reasonable?

At home, although somewhat less knotty, our difficulties are more disconcerting. We are usually confident in our conjectures on the meaning of the speech of those who speak in our native tongue, but Quine's observations cannot but leave us somewhat uneasy (at least in an intellectual way) that we might have a radically different view of the world from "everyone else," if indeed "everyone else" has a single view of the world. For the lesson of the indeterminacy of translation is that analytic hypotheses of translation form a system only together with a background theory of what the world is like. This theory may be supplied explicitly or it may be implicit, but it must be there.

*

For another perspective on the situation we may consider things from a formal point of view. Formally, translating one language into another involves setting up a mapping from the expressions of one language into the expressions of another, from an unknown language into a known language. But not just any mapping is acceptable (even if the observations about indeterminacy are true, which they are).

We may identify three parts of any operating language: the expressions, the universe, and the mapping of the expressions onto the universe (called the model). Although the translation itself is represented merely as a mapping between expressions, it is the other two components that constrain the choice of mappings. All three components of the known language are available, as are the expressions of the unknown language. By the principle of charity we at first assume that the unknown universe is the same as ours. If we also knew the model, our procedure for translation would be very simple: given an expression in the unknown language, map it into the universe and use the inverse of the model of the known language to get the result.

However, we do not know the model in the cases considered by Quine (we do not know the universe either, but I will come to that later). Since the model of the known language is given, a conjecture for the mapping of an expression of the unknown language onto some of the known universe, is also (or perhaps mainly) a conjecture of part of the model of the unknown language. We can describe our aim in doing a complete translation as a full account of the model of the unknown language.

The problem that Quine poses is that our known language includes a number of different ways of partitioning the whole universe. The universe is partitioned into classes by equivalence relations, but we have a number of these in English which are adequate to partition the whole (material) universe. Our main check on translational conjectures is through the use of these equivalence relations. But since our checks are only within a set of equivalence classes, we can only tell if one ostended object, associated with a particular expression, is in the same class as another ostended object associated with the same word -- but there is no way to tell if we have the right set of equivalence classes!

We can certainly check on differences and similarities within a set of equivalence classes, but there is no way to check for equivalences across classes. In most cases we assume that the partitioning used by the speaker of the unknown language is the same as our own. In other cases, most notably when dealing with the languages of apparently primitive peoples, we are less inclined to assume for those speakers the same partitioning that we use.

The thing is that we can do a systematic translation of the unknown language using any one set of equivalence classes, but we must recognize that we have no claim to any absolutely correct translation. Our translation may be perfectly adequate, relative to the equivalence class we have chosen or, in Quine's terminology, relative to our analytical hypotheses of translation. But many other hypotheses, or partitionings, would also serve.

This is the situation that bothers Quine. We do not seem to be able to determine the correct equivalence classes.

It is to be emphasized that within such a class it is easy to distinguish the correct translation. Once we have made some choice of analytic hypotheses, ostension will be adequate to discriminate between any alternatives for translation of a particular expression. The only problem is that we cannot be sure that we had originally chosen the correct set of equivalence classes (i.e. the correct analytic hypotheses), but again, within a given set of equivalence classes there is one, and only one, correct translation from an unknown to a kown language, or in general between any two languages. Also, given a list of expressions and its model in one set of equivalence classes there is a unique transformation of that model for any (other) given set of equivalence classes (with Quine we are assuming such classes to be determined by equivalence relations).

For example, we can tell that a particular object (a book, say) is associated with a particular word ("book"). However exactly what the speaker has in mind by the book object is not fully determined. He may be pointing to a coherent material object, or he may be intending to point to what he sees as a material expression of some mystical force. We cannot ever find out for sure which he intends.

*

We may call a model with its transformation rules a translation form. There will of course be no unique such form for any pair of languages, nor can any preference be assigned a priori between the various alternatives even if there were some way of getting an exhaustive list of them. However, various translation forms can be compared from the perspective of a given, known language and, for an unknown language, they should all turn out to be equivalent in the sense that the various transformation rules convert all different translation forms into the same model for the same choice of equivalence class.

It may be instructive at this point to ask about the importance of determining the correct set of equivalence classes. That it is important and necessary to the task of the translator is assumed by Quine with little justification except his pointing out (with something of a shudder) how radically the various classes may differ. We have pointed out that once a set of equivalence classes is chosen to work with, we can construct a unique translation (within that class) and that there is also a unique conversion from a model (in the sense given to that term in this paper) in one class to a corresponding model in any other such class. (There are some constraints on the choices of other universes, such as size and richness of structure, but as long as we stay with these set of equivalence classes, which are only alternate partitionings of what is "basically" the same universe we will not run into any of these problems.)

The consensus seems to be that this does not complete the job of the translator, that he must still seek to determine the correct class. However, if Quine's observations are correct, his job is definitely done in one sense. He will do no more because he can do no more. Ultimately in the set of equivalence classes identity is inscrutable; there is no standard against which we can measure particular classes and of course no other way of identifying them except against some standard or other.

To reiterate, why is it that we care about which is the "right" set of equivalence classes? Why are we not satisfied with what we can definitely (though not necessarily easily) achieve, namely, the best translation within some chosen set of equivalence classes?

There would seem to be two types of answers. One would expostulate on the "radical differences" between the various distinct classes, pointing out that since the various classes are so different it is odd to suppose that they are all simultaneously suitable. The other would focus attention on the difficulty of choosing one class to work with; even if we were satisfied with translation within one equivalence partition, how are we to choose that particular partition?

Continues . . .