There are various processes by which the phonetic form of a word or a phrase is not pronounced (e.g., ellipsis under identity with an antecedent, but others as well). For example, in VP ellipsis, we have:
(1)
John left early, and Mary did <leave early> too.
In this example, <leave early> is meant to indicate a constituent that is not realized phonetically, yet is present syntactically.
But consider the fact that there are three different kinds of information associated with the syntactic object [leaves early]. There is (a) the syntactic structure, (b) the phonetic string leaves-early, and (c) there is the semantic value of [leaves early].
From this perspective, one can ask if there are ellipsis operations that leave the phonetic string intact, but delete the syntactic structure (and hence the semantic value). For example, consider this putative case of ellipsis:
(2)
John ate fish, and Mary ate fish too.
In this example, the second occurrence of [ate fish] cannot mean simply [ate]. For example, it is impossible to continue (2) with “but I don’t know what”. (# stands for semantically anomalous):
(3)
#John ate fish, and Mary ate fish too, but I don’t know what.
Compare (3) to (4), where the second clause uses intransitive ate:
(4)
John ate fish, and Mary ate too, but I don’t know what.
In other words, the second occurrence of fish in (2) must be semantically interpreted. A similar contrast holds for adjuncts:
(5)
#John left at 3, and Mary left at 3 too, but I don’t know when.
So, we seem to have the following asymmetry:
(6)
a. Words and phrases may be unpronounced.
b. Words and phrases may not be uninterpreted.
We can put this asymmetry in terms of recoverability of deletion:
(7)
Recoverability of SM-Deletion
XP can be deleted (not pronounced at the SM-Interface) only if there a syntactically identical antecedent.
For example, in (1), the antecedent is [left early] in the first conjunct, and so in the second conjunct [leave early] can remain unpronounced. We put aside tricky issues here of how to define “syntactic identity”.
But no such condition exists on the CI side:
(8)
Recoverability of CI-Deletion
XP can be deleted (not interpreted at the CI-Interface) only if there is a phonetically identical antecedent.
Consider (5), from the point of view of (8). If (8) held, then yesterday in second conjunct would not have to be interpreted, since there is a phonetically identical antecedent in the first conjunct.
It may be possible to tell a learnability story here. How could the correct interpretation of (2) ever be learned? How would a child, seeing ate-fish ever learn that it did not mean the same as the phrase [ate fish]?
But I would like to suggest something more straightforward. It may be that the reason for the asymmetry is that that there is an important asymmetry between sound and meaning in the architecture of UG.
An I-language generates a set of syntactic objects, and they may be externalized. There are various ways by which a generated structure may not actually be externalized (as described in the literature on ellipsis). Hence the generalization in (5a).
Now suppose an I-language generates the syntactic object SO. Then it is impossible for SO to be externalized (at the SM-Interface) without being interpreted at the CI-Interface. Why is that? It follows from Full Interpretation (see Chomsky 1986: 98, whose statement is more general than (9) and covers both the LF an PF interfaces):
(9)
Full Interpretation at CI-Interface
Every element of the SO at the CI-Interface must receive an appropriate interpretation.
There is no possibility of ignoring or neglecting things at the CI-Interface.
What I am claiming is that FI holds at the CI-Interface, but not the SM-Interface. Phonetic form can in fact be ignored (deleted, unpronounced, neglected) at the SM-Interface. Syntactic structure cannot be ignored (deleted, unpronounced, neglected) at the CI-Interface.
These results point to the centrality of the derivation from the lexicon to the CI-Interface, and the secondary nature of externalization (which is outside the core derivation to the CI-Interface).
Very interesting ideas — but they inspired me to wonder a bit more about whether your description of what we do and do not find in the first place could be interrogated a bit more, or at least brought into sharper focus.
ReplyDeleteI wonder first whether it is actually true that an elided constituent is "not realized phonetically, yet present syntactically". While it is certainly true that ellipsis eliminates the expected segmental content of the elided constituent, there is interesting recent work by Danfeng Wu in her dissertation that asks whether prosodic information (timing, etc.) might remain after ellipsis (https://ling.auf.net/lingbuzz/006846) — so that, in a sense, ellipsis might be not a failure to phonologically interpret syntactic material, but a reduced or altered mode of phonological interpretation. Two copies, both of them phonologically interpreted, but differently.
With that in mind, I wonder if there might not be counterpart in the opposite direction wherever we find movement producing doubling, where a trace is pronounced as a full copy (e.g. predicate clefts of the "read I read a book" type). Here my guess is that we have two copies of the same PF material, both of them semantically interpreted, but not in the same way. Perhaps one is a variable and the other is its binder. Two copies, both of them semantically interpreted, but differently.
So might there be more symmetry than you suggest? Just a conjecture, not a conclusion, provoked by your really interesting post.