Date: Wed, 30 Oct 1996 02:11:03 +0100 (MET)
From: Tanya Reinhart <Tanya.Reinhart@let.uva.nl>
To: Linguistics Conference <LINCONF@tamvm1.tamu.edu>
Subject: Reply to Lidz
Message-Id: <Pine.SOL.3.91.961030020953.23488A-100000@cclsun01>
In-Reply-To: <9610281909.aa00627@monk.ling.udel.edu>
1. THE CHAIN CONDITION
Lidz brings up the following interesting case from Kannada, as an
argument against R&R's chain condition. In (a) the SE is case marked,
and still the sentence is allowed.
(a) Hari-(yu) tann-annu hode-du-koND-a
Hari-(NOM) self-ACC hit-PP-REFL.PST-3SM
'Hari hit himself'
Since the chain position normally disallows two cases in a chain,
it is not clear why (a) is permitted. Note, however, that the chain
condition in R&R does not mention case, as such. Rather, it requires
exactly one +R position per chain (at its head). +R is defined as
fully specified for features. The relevant features are not only
case, but also phi features (gender, no, etc). Any NP which is not
fully specified is -R. By this definition, 'tann-annu' in (a), is
still -R: Since the SE anaphor lacks phi features, though it has case.
(This, indeed, is an interesting exception to the common fact about
SE anaphors, that they lack also case, but it nevertheless is
consistent with R&R's formulation of the chain condition.) Note
incidentally, (in respond to another remark by Lidz) that +R is not,
itself, a semantic property, (like referential independence) but a
syntactic one. We do not assume that semantic properties like
reference are directly encoded in the computational system, but rather
the other way around: only if an NP is +R, i.e. it has full feature-
specification, it can refer independently (- A generalization of
Bouchard). One may loosely speak of +R as meaning 'referentially
independent', but this property is obtained only as entailment.
The real counter example for the chain condition would be the second
Kannada example Lidz cites in (b), where the object is a pronoun.
Hence, having both case and full phi features it should, indeed, be
ruled out.
(b) Hari avan-annu hode-du-koND-a
Hari him-ACC hit-PP-REFL-3SM
'Hari hit himself'
This, unfortunately, is another case where we have different informants
for the judgments. My data on Dravidian was collected in a course
I gave in Hydrabad's summer school in 1994. The structures in (b)
were discussed in detail, and were judged out. So, of course, if
my data is wrong, this, indeed, is a problem for reflexivity, as well
as for any other existing theory, as we shall see directly.
Lidz cites another example, where I have no qualms about the judgment,
and offers also a theoretical account:
Lidz says:
"Similar cases are found in various northern Italian Dialects,
like Piedmontese (Luigi Burzio, personal communication) and
Padovano (Christina Tortora, personal communication), given here:
(c) Gianni se varda lu
Gianni REFL sees him
'Gianni sees himself'
These facts suggest that R&R's chain condition is not at work
in determining the conditions on anaphora and thus that Reinhart's
claim that a theory with the chain condition and no binding
conditions is more parsimonious than one with both does not hold.
If there is a chain condition on A-movement, the
anaphor-antecedent relation does not appear to be subject to
it.
>From the perspective of the paper under discussion, these facts
suggest that the binding conditions do not apply if Condition
R is satisfied (i.e., since condition B would presumably be
violated in (b-c). The details of getting this to work are not
entirely clear. However, the generalization seems to be that
if the interpretative component can avoid applying the binding
conditions (by providing an interpretation to the predicate
directly from the lexicon), then it should."
Let us suppose that the details of getting this to work will get
eventually clearer, so the example in this dialect of Italian will
follow. Why then should this not yield precisely the same results
in all languages? E.G. In dutch (i) the verb has an intrinsic reflexive
entry, and the English 'behave' is intrinsically reflexive too. So
Lidz modified reflexivity rule (just like R&R's original) should allow
the sentences, now that the BT condition B is allowed to be ignored,
and the chain condition is given up.
i) *Max wast hem (Max washed him)
ii) *Max behaved him
So, we can decide to capture correctly Northern Italian dialects and
leave most other languages of the world to be handled by language
specific rules. Or we can go the other way around. In any case,
this is not a problem specific to the chain condition of R&R, but
to any existing formulation of the binding theory. We certainly will
not be saved by assuming here the traditional BT.
In fact, the pattern of Northern Italian is not that unique. It is
the same in Frisian, and in other languages and dialects. Eric
Reuland, and Coopmans & Philip have worked on some such cases and
argue that where this is allowed, the pronominal form lacks, in fact,
the full paradigm of phi features, hence it is allowed into a chain.
2. LIDZ' CONDITION R.
There is a problem with understanding the implications of Lidz'
proposal regarding near-reflexivity, which got further sharpened in
the discussion Lidz-Dubinsky.
The starting point of Lidz was to explain why (14b) does not have
the near-reflexive reading.
(14b) ze zag zich in een griezelige hoek staan
she saw self in a creepy corner stand
'She saw herself(=reflection, *statue) in a creepy
corner stand'
The idea is to derive this from the fact that lexically reflexive
predicates do not allow this reading. The problem was in what sense
can 'see' be a lexically reflexive predicate in this example. There
is a danger here that we will have a fully circular answer: This
would be the case if we have a theory that states something like the
following.
Theory iii:
a. SE can occur locally only when the predicate is lexically
reflexive.
b. Whenever SE occurs locally, then the predicate is lexically
reflexive.
There is no way to render this theory false, but this is one of these
cases of being right for the wrong reasons. (Theory iii is more or
less equivalent to a simpler theory that states that SE can occur
wherever it can occur.)
Of course, Lidz is aware of this danger, hence he offers an attempt
to show that 'see' is, indeed, a lexically reflexive predicate here,
even though this does not show in the familiar tests of intrinsic
reflexivization (nominalization).
Luckily, however, there is a clearer way to get out of the mystery
here, observed if we look at the minimal pair of (iv), as opposed
to (14b).
iv) *ze zag zich.
(14b) ze zag [zich in een griezelige hoek staan]
she saw self in a creepy corner stand
'She saw herself(=reflection, *statue) in a creepy
corner stand'
While 'zich' cannot occur as the object of 'see', it can as the subject
of the complement of 'see' (ECM subject). This pattern, with the
verb 'hear', in both Dutch and Norwegian is illustrated in full in
R&R (LI) p. 710. So what (iv) show, is that the tests for intrinsic-
lexical reflexivization (which Lidz mentions in his reply to Dubinsky)
work correctly in this case, and 'see' is not intrinsically reflexive.
Hence both R&R condition B, and its modified version in Rule R of
Lidz, correctly rule out (iv). Both also do not fail to allow (14b),
since no reflexive predicate is formed at all in this case.
But what we are left with is the interesting question Lidz brought
up, namely why (14b) does not allow, nevertheless, the 'statue'
interpretation, which he labels 'near-reflexive'.
The answer which suggests itself is that SE anaphors generally do
not allow this reading, independently of whether they occur in
intrinsic reflexivization contexts or not. (I am sorry, I no longer
remember if Lidz had any example of a SE anaphor which does allow
it. So, of course, this should be checked.) If this is the case,
then the most natural account is the one in terms of stress (neither
SE, nor any anaphor occupying a lexically absorbed position can carry
stress). But this is not the only conceivable answer. We do not have
to give up Lidz' important finding, just look for a more accurate
way to capture it. On the other hand, this does not strike me as
the type of problem that can justify changing the whole binding theory.
(Since, as I mentioned, what we are dealing with here is the
interpretation, rather than the distribution, of anaphora, and the
problem Lidz noted is not at all specific to reflexive contexts.
Jackendoff's original findings were about pronouns.)
3. THE REFLEXIVIZING MORPHEME IN KANNADA
Unrelated to the previous, Lidz makes an interesting observation in
his reply to Seely. He notes that the same morpheme (koL) which is
used to reflexivise a predicate in Kannada is used also in other
contexts, most notably the ergative structure in (ii).
(ii) baagil-u much-i-koND-itu
door-NOM close-PP-REFL.PST-3SN
'The door closed'
As he notes, this is not at all unusual. In many languages the same
morpheme can be used in both ergatives and reflexive contexts (Italian,
Hebrew)
This does not mean necessarily that koL is not a reflexivizer.
Chierchia has suggested, based on Italian, that, in fact, lexical
reflexivity and ergativity are derived by the same semantic operation
on a theta role, in the lexicon. (I pursue the same line in a work
in progress. The general correlation between the two was observed
also by Marantz.) The basic idea is that the ergative version (in
ii) is derived in fact from the transitive 'close', by an operation
which absorbs one of the roles. The only question is whether the
remaining argument will realize as the subject (reflexive) or as the
object (ergative). In many cases the realization is just identical.
E.g. in German, 'the door opened' happens to be 'Die Tur offnet sich').
--Tanya Reinhart