the child by using `con'. Symmetrically, the child would use `con' for
self-reference, and `me' (etc.) to refer to the mother. Thus, the ordered
pairs (me con) and (con, me) determine unique familial relations. [9]

Accuracy demands some qualification of the claims thus far put forward. It
might reasonably be suggested that Vietnamese does, in fact, contain
several self-referential expressions (such as `toi', `ta' and `minh') which
offer considerably less resistance to straightforward translation as `I',
which do not imply that a given familial relation obtains between speaker
and the person addressed, and which have an appropriate use within
monologue and private musing. Indeed, the English `I', employed within a
wide range of contexts, is almost invariably rendered in Vietnamese as
`toi', and the two words thus seem, in some respects, likely candidates for
synonymy. However, self-referential expressions such as `toi', `ta' and
`minh', while promising in the ways noted, nonetheless lack the crucial
property of transcontextual invariance. `Toi', for example, was originally
used self-referentially by a subject in addressing the king. [10]
Similarly, `ta', suggesting a `royal' plurality, is used by a superior in
addressing an inferior. While the correct use of such expressions does not
imply that the speaker bears a given familial relation to the person
addressed, it does bear the implication that the speaker is not related to
the person addressed via familial bonds. Thus, since there are contexts in
which such expressions cannot be appropriately used in self-reference, they
are not transcontextually invariant, and cannot be translated, without
serious distortion, as `I'.

There are, moreover, expressions in Viemamese signifying `self-'ban-nga'*
usually translates `ego' or the Vedantic `atman'; `vo-ban-nga' translates
the Buddhist 'egolessness'--but nothing which corresponds precisely to
`myself'. Letting `v' vary over V-expressions, the closest we can get to a
literal translation of 'myself into Viemamese is either `chinh v' (`the
very v'or `v itself') or `ban-nga cua v' ('the self of v'). But then there
are as many senses of 'myself in Viemamese as there are V-expressions.

　　　　　　　　 The `Reference' of Self-Reference

There is, it seems, no more fundamental epistemic principle governing
analysis in general, and the analysis of self-referential expressions in
particular, than that which demands that, in knowing the analysans, I
thereby know precisely what I know in knowing the original pre-analytic
analysandum. If:

(1) A am thinking is the correct analysis of

(2) I am thinking (where `A' represents an expression introduced by the
analysis to replace `I'), then surely, in knowing that (1), I thereby know
that (2), and conversely. If not, (1) cannot be a correct analysis.
However, the substitution of a referential term or phrase of any type
whatever for `A' in (1) will, I submit, cause the analysis
straightforwardly to fail the epistemic exam.

On this issue, if not on others, I firmly endorse Zemach's pellucid
argument:

　　 Reference can be achieved by descriptions, names, or demonstratives. A
　　 purely qualitative definite description will never do, since one may
　　 fail to believe it applies to the object it denotes. It may also fail
　　 the uniqueness condition, and denote nothing. A name will not do
　　 either: although it may refer to the right object, the believer may
　　 not know, or forget, what this name denotes. An indexical will not do
　　 at all: although a semantic rule may determine its reference, the
　　 believer may not know this rule or confuse it with another. [11]

A somewhat livelier and certainly more concrete appreciation of the
non-referential character of (purportedly) self-`referential' expressions
may be elicited by reflection upon the imaginary tragedy which Nozick
describes:

　　 Imagine that you and two other persons have been in an accident and
　　 are lying completely bandaged on three beds in a hospital, all
　　 suffering from amnesia. The doctor comes in and tells what has
　　 happened, that examinations have been made, and that (where the three
　　 persons boringly are named X, Y and Z) person X will live, Y will die,
　　 and Z has a 50/50 chance. [12]

Clearly, as this sombre scenario illustrates, self-reference cannot be
achieved in virtue of a definite description.

　　 The situation is not helped if the doctor gives the full life
　　 histories of each of the three. Nor are you helped if she says the