LINGUIST List 3.784

Sat 17 Oct 1992

Disc: Unification vs Constraints

Editor for this issue: <>


  1. steven schaufele c/o elaine schaufele, unification- vs. constraint-based grammars
  2. anoop sarkar, Inheritance and Unification Grammars

Message 1: unification- vs. constraint-based grammars

Date: Fri, 16 Oct 92 20:18:03 CDunification- vs. constraint-based grammars
From: steven schaufele c/o elaine schaufele <AXSSCHAUICVMC.bitnet>
Subject: unification- vs. constraint-based grammars

Glad to see a discussion initiated on the relative merits of unification- and
constraint-based grammars; many thanks to Avery Andrews for bringing it up.
As it happens, my 1990 University of Illinois dissertation (Free Word-Order
Syntax: the Challenge from Vedic Sanskrit to Contemporary Formal Syntactic
Theory--available from University Microfilms) addresses some of the issues
raised in this discussion. In particular, i demonstrate that LFG, precisely
because it is a unification theory, is able to describe and account for
'scrambled' (i.e. freely discontinuous) phrases in Sanskrit. This is
because, in LFG, both interpretation and constraining is done not in terms of
structural relations in c-structure but by means of unification in
f-structure. The PS component can declare that both continuous and
discontinuous phrases are grammatical, while f-structure treats all
constituents as integral.

These same constructions present problems for the Principles & Parameters
Approach, precisely because, as Peter Svenonius (LINGUIST 3-774) notes, in
that framework constraints are necessarily defined in terms of constituent
structure. In particular, i argue that an adequate description of Sanskrit
scrambled phrases in PPA would require the weakening of the Structure-
Preservation Constraint to the point that it is no longer able to constrain
Local Transformations. Recent work in PPA (especially since what Peggy Speas
has called the 'functional big bang', in which every functional category or
 feature has its own set of projections) seems regrettably to be moving in the
direction of greater constraints to the point that the theory is losing
all falsifiability.

I am inclined to agree with John Coleman (LINGUIST 3-759) re. terminology.
Theory labels are often misleading (Darwinian and Einsteinian theory are
certainly not the only theories of evolution and relativity, respectively,
around). PPA certainly requires, at least implicitly, some form of
unification in order to get all the modules to agree on a derivation. And
how come Chomskyan theory gets to call itself 'Principles & Parameters'? Are
all other frameworks unprincipled? Don't linguistic and metalinguistic
parameters have a role in all linguistic theories? Likewise, 'constraints'
are, i believe, an essential part of any viable linguistic theory. But, as
Mark Johnson (LINGUIST 3-759) notes, there is a variety of logically possible
constraint-types, and the choice among them ought in principle to be an
empirical issue. Does anybody know of any psycholinguistic evidence in
favour of one type over others? It has long seemed to me that the
'Standard-Theoretical' school has gone overboard both in multiplying
constraints and in insisting that they must be in terms of constituent
structure, by making use of an unneccessarily limited sample of languages
(mostly modern Western European languages, with occasional glances at other,
more exotic breeds). The result has been the assumption that these languages
are normative and all others are weird, whereas if we had started with a
different, if not more diverse, sample we might logically end up with a
theory of UG that is not only simpler in terms of Occam's Razor but more
generally applicable.
Steven Schaufele
Mail to author|Respond to list|Read more issues|LINGUIST home page|Top of issue

Message 2: Inheritance and Unification Grammars

Date: Fri, 16 Oct 92 09:26:08+05Inheritance and Unification Grammars
From: anoop sarkar <>
Subject: Inheritance and Unification Grammars

Avery Andrews says (in Linguist 3-776),

> 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.

What indeed _is_ a linguistically right way to do anything?

If the right way refers to it being a psycholinguistically right way of
parsing NL, then lexicons which implement hierarchy operations like
inheritance, classification, unification, etc. are right on line with
current research in concept hierarchies/semantic nets, which are
considered to be psychologically interesting.

Besides, inheritance also makes computational sense; it increases the
deductive closure of a system, which means that a small set of rules
(or procedural knowledge) can project a large amount of information (or
declarative knowledge). Linguistics is about the characterization of
languages, but that does not obviate efforts towards a more _efficient_

Anoop Sarkar
Mail to author|Respond to list|Read more issues|LINGUIST home page|Top of issue