Dan Everett: Word formation constraints (reply to Alec Marantz)

[Martha McGinnis]

OK, I've finally had a chance to read this thread carefully, and I
find I have nothing to add to Heidi's remark that DM looks like a
good theory to be adopting to try to account for these sentential
adjuncts, since DM doesn't allow us to sweep their unusual 'atomic'
behaviour under an all-purpose Lexicon.  With luck the comparison
with Wari' quotatives will shed some new light on the phenomenon.

Nevertheless, I thought I would comment on the discussion of DM, MP,
etc., in case another interpretation is helpful.

>A formal theory is, like DM, MP, etc, a theory in
>which FORM drives the model, rather than meaning. That is, meaning is not
>directly causally implicated in the model. In this sense. DM, like MP, are
>structuralist theories.

There may be similarities between structuralism and generative
linguistics (e.g. the focus on systematic properties of language,
rather than on prescriptive quirks, metaphor, etc.), but there's an
important distinction to be made here as well.  DM and MP, like other
generative theories, are concerned with systematic correspondences
between sound and meaning.  If I recall correctly, structuralists
were convinced that language could be captured simply in terms of
phonologically-based patterns, without reference to meaning. Thus
structuralism treats different languages as internally coherent but
fundamentally distinct systems.  Generative linguistics treats them
as different states of the language faculty, whose lexicons draw on a
universal set of semantically based syntactic features (tense,
aspect, definiteness, specificity...), as well as phonetically based
phonological features.

The heyday of structuralism is far enough back in history now that I
know of few explicit arguments against it. One is Morris Halle's
argument against the structuralist notion of 'phoneme'.  I also seem
to recall that Halle & Marantz argue against a structuralist approach
to morphology in their first DM paper -- i.e. that 'position classes'
based purely on morphological distribution don'tt capture the correct
generalizations about Potawatomi morphology.

>I have not yet seen in Andrew Carnie's thesis, for example
>(but I am working my way through it slowly, so if it is there, my
>apologies to Andrew), an attempt to ask the question of where the
>predicate node arises for the S to move into. It is just there in the
>phrase structure.

I'm not 100% sure I understand the question, but the issue here might
be the order of assembly of syntactic structures.  It's not possible
to assume that the syntax composes a single tree, step by step,
moving upwards (or downwards, for that matter).  For example, phrasal
specifiers have to be pre-assembled in a subtree before they can
merge with the main tree.

>Since there
>is no base component and since nodes are not so labeled in the lexicon, ex
>hypothesi, there is no way to have labeled nodes in the syntax, at least
>until after Merge and the postsyntactic DM work.

I think there's some flexibility here.  Virtually everyone assumes
the lexicon contains items of different categories, which have
different syntactic properties (both for Merge and Move).  Since
introducing Bare Phrase Structure, Chomsky has assumed that phrasal
units also have category labels, whose X-bar level can be read off
the structure.  Alec suggests that the syntax can get away without
making use of these category labels.  This sounds to me like a new
suggestion -- that labels need not be stipulated, but can be read off
the structure. By contrast, BPS uses different types of node labels
to distinguish between specifiers and adjuncts.

>On the other hand, I must admit that I *am* confused about node origins
>when those nodes do not dominate anything. Take movement of a phrase into
>the head of TP, for example. Why is the head/phrase there to serve as a
>target for movement? Either a node is required for lexical reasons or it
>is not there. What requires T?

Perhaps nothing, which is why we can just say "Gone fishin'."  The
notion that Merge has to be licensed by something also sounds to me
like a new proposal.  As far as I understand, Chomsky treats Merge as
free: it either does or doesn't succeed in generating semantically
interpretable structures.  It's possible that arguments are forced to
Merge by a local need to satisfy the theta-requirements of a
predicate, but even if so, this would not entail that there is no
unforced Merge.  What empirical benefits do we reap by taking the
position that Merge is always forced or licensed?

>But globality seems to be an
>embarassment here. Surely GREED or something in that spirit is more
>desirable, i.e. a local lexical requirement.

As you know, 'Suicidal Greed' is local, but Greed and Procrastinate
are the classic global constraints.  Am I allowed to say 'classic'
about the 1990's yet?

Cheers,
Martha


mcginnis at ucalgary.ca