LINGUIST List 3.776

Wed 14 Oct 1992

Disc: Postal's "Best Theory" and Unification

Editor for this issue: <>


Directory

  1. Randy Allen Harris, "The best theory"
  2. Avery Andrews, Unification vs. Constraint-Based Grammars

Message 1: "The best theory"

Date: Fri, 9 Oct 92 9:57:41 EDT "The best theory"
From: Randy Allen Harris <rahawatarts.uwaterloo.ca>
Subject: "The best theory"

Just some points of clarification following Geoff Nathan's
comments on "Postalian Best Theory."

Postal's best theory was one with only one type of rule,
alright, but it was the transformation, which in early
generative semantics did both syntactic and semantic duties.
He contrasted this with the _Aspects_ model, which had
transformations and semantic interpretation rules ("projection
rules" at the time), and with Chomsky's post-_Aspects_ suggestions
that yet other rule types would be required--in particular,
lexical redundancy rules and the new varieties of semantic
interpretation rules that blossomed a few years later in
Jackendoff's _Semantic Interpretation_.

This best theory, Postal called "Homogenous I"--'homgenous' because
of the single rule type, 'I' because it didn't hold water.
Though Homogenous I was preferable on a priori grounds, the
facts dictated otherwise, and Homogenous II was born in the
later stages of the same paper--HI with global rules.

Global rules were super-transformational in the way that Ross'
island constraints were/are super-transformational, and Perlmutter's
constraints, and Postal's cross-over principle, and recoverability-
of-deletion, and so on. In fact, Postal (and Lakoff, from whom the
term originates) considered all of these devices (or virtually all,
one can't be entirely sure) as instances of global rules. Global
rules were meant to be a bag of devices, in parallel to the bag
of devices (for deletion, insertion, and movement) under the name
"transformations". (All of these devices--or virtually all--were
equally accepted by 'interpretive semantics', by the way.) Postal
considered global rules a necessary evil.

So, Postal's final argument was that Homogenous II (=generative semantics
c1970) represented a less heinous departure from the best theory than
a theory which had transformations, global rules (without the label),
and several other rule types as well.

I leave it to someone else to figure out how this position bears on
unification and/of constraint-based grammars.

--Randy Harris
Mail to author|Respond to list|Read more issues|LINGUIST home page|Top of issue

Message 2: Unification vs. Constraint-Based Grammars

Date: Sat, 10 Oct 92 11:21:12 ESUnification vs. Constraint-Based Grammars
From: Avery Andrews <andalingdurras.anu.edu.au>
Subject: Unification vs. Constraint-Based Grammars

It's interesting that there's disagreement about whether unification
refers to a process (Henry Thompson, Linguist 3-756) or a mathematical
operation (John Coleman, Linguist 3-759), but the fact that one hears
people talking about `unification method_s_' suggests the latter.
What typically happens in the frameworks that use it is that it is
found that two descriptions are found to be descriptions of the same
thing (a `long component' manifested at two positions in the structure,
opne might say), and the unification method combines the descriptions in
such a way as to reveal the presence of contradictions. So perhaps
one could say that unification is `and' in logical regimes (restricted
ones, of course) where consistency is decideable.

Perhaps `unification-oriented' would be a more accurate term than
`unification-based', expressing the idea that unification is a preferred
method in these frameworks, but not necessarily their only basic idea.
But would it catch on?

As for other imports from computer science to linguistics (Rich Hillard,
Linguist 3.756) , it seems to me that some people are trying to do
something with inheritance, but it isn't clear to me that this has had
much impact on mainstream linguistics (but then it's not so clear that
unification has either). I've seen plenty of articles arguing that
inheritence is a good way to implement things involving the lexicon,
but I haven't noticed much in the way of attempts to show that it
is the linguistically right way to do anything.

Avery.Andrewsanu.edu.au
Mail to author|Respond to list|Read more issues|LINGUIST home page|Top of issue