LINGUIST List 5.825

Tue 19 Jul 1994

Disc: Claims about Harris and Chomsky

Editor for this issue: <>


Directory

  1. Bruce Nevin, Claims about Harris and Chomsky

Message 1: Claims about Harris and Chomsky

Date: Mon, 18 Jul 1994 16:59:49 Claims about Harris and Chomsky
From: Bruce Nevin <bnevinLightStream.COM>
Subject: Claims about Harris and Chomsky


At the beginning of this year, Alexis Manaster-Ramer made several claims
here about Zellig Harris's work on syntax, to which I have taken
exception in private correspondence. In the interest of avoiding
unproductive argument on the net, I have urged that we should come to
agreement about our disparate views and then post a summary. Alexis
acknowledged that his original claims were "excessive" and suggested
(email of February 23) that we might put together a joint paper for
publication. On April 21, he returned a more extended statement. I
returned an edited version with addenda of my own. Unfortunately, since
then he has been unable to find time for developing this discussion
further. With his permission, then, I am now posting a summary
incorporating the edited version of his text plus my additional
statements. I hope it is clear where we still disagree, and what our
respective views are.

First, the initial claims and objections. Alex had asked for information
about Harris' treatment of word order prior to 1956. In the Linguist
digest on 28 Dec 1993, Alexis said that Harris's

 treatments of both syntax and morphology in the form of explicit
 formulae did NOT handle word order AT ALL. They were intended to
 capture the facts of permissible morpheme sequences without paying
 attention to the actual order of the morphemes and morpheme
 sequences.

In private mail of 05 Jan 1994, I questioned this claim, saying in part
"How can you specify a sequence without specifying order?" I emphasized
the importance of word order for discourse analysis, which played an
important part in the development of transformational analysis when
Chomsky was a student of Harris long before 1956. [However, refer here
to correspondence from a fellow student with Chomsky in 1946, quoted
below, suggesting that discourse analysis may have been the later
development in Harris's work.] In the course of several exchanges of
email, Alexis agreed to give citations to back this claim.

In the Linguist digest 5.388 on 04 Apr 1994, Alexis said that

 the very definition of construction in Bloomfield's immediate
 constituent work was precisely a pairing of form and meaning, and
 there are many examples of this in [his] practice. . . .

 The dropping of the meaning component, as well as of several parts
 of Bloomfield's theory of the form of constructions (notably, his
 treatment of discontinuity, of free word order, of zeroes or ellipses,
 of crossclassification, and of constructions with an unbounded number
 of constituents) was just part of what happened when Harris and then
 Chomsky tried to "formalize" Bloomfield's theory.

I objected that for Harris (and for Bloomfield) it was not a question of
correlating linguistic form with linguistic meaning, because these were
considered two sides of the same coin (see _Historiographia Linguistica_
XX 2/3 355-398). I questioned the basis for asserting that Harris failed
to treat discontinuity, free word order, zeroes or ellipses, etc.

In his April revision and expansion of these claims, Alexis began
developing an argument that Harris watered down the "Bloomfieldian
teaching" on immediate constituents, and that this paved the way for
Chomsky's formalization of IC theory in the highly inadequate form of
PSG. In the edited version that follows only a few of my contributions
are explicitly set off with some sort of brackets. Generally, the more
detailed discussions of Harris' work are my contributions, for example,
the first sentence under (a), the second paragraph under (c), and much of
(e) and sequel (in the second occurrence of these letters in the
list of items discussing Harris' views).

-=+=-=+=-=+=-=+=-=+=-=+=-=+=-=+=-=+=-=+=-=+=-=+=-=+=-=+=-=+=-=+=-=+=-=+=-

Chomsky's campaign against the structuralist immediate constituent syntax
and in favor of transformational syntax began with his formalization of
IC syntax. This formalization he called phrase (or constituent)
structure grammar. However, it is apparent (see my paper with Kac in
Linguistics and Philosophy, 1990) that the formalization was highly
incomplete, leaving out the fact in the original formulation of immediate
constituent theory by Bloomfield. My [AMR's] claim is that Harris's
version of immediate constituent theory also omits important elements of
Bloomfield's teaching, and that this probably paved the way for Chomsky's
formalization of the IC theory in the highly inadequate form that we know
as PSG.

Bloomfieldian IC theory included the following characteristics which are
omitted from PSG:

(a) Every construction (roughly corresponding to a phrase structure
 rule) had to have an explicit meaning, while phrase structure rules
 do not have meanings.

(b) Constructions could have an unbounded number of ICs (for example, as
 Bloomfield mentioned, "serial" constructions, i.e., coordinate ones),
 while in the formulation of PSG that Chomsky used as late as the
 mid-60's, unbounded branching is not permitted.

(c) Bloomfield provided for discontinuous constituents, PSG does not.
 [However, Bloomfield's example of discontinuity involves paratactic
 interruptions of sentences, not the word order of yes-no questions
 and the like, which were handled by taxemes of word order. Bear this
 in mind in the discussion of discontinuity below. --BN]

(d) Bloomfield admitted zero constituents, PSG does not.

(e) Bloomfield provided for word order as a separable component of a
 construction, and a construction could have no word order at all;
 every PSG rule must place the elements to which it refers in a fixed
 order.

(f) Bloomfield gave clear indications of complex categories, that is,
 a single term B could refer to categories [AB] and [CB] in common;
 this is not possible in PSG.

 My question is whether Harris, Chomsky's teacher, had anything to do
with Chomsky's leaving all these things out. Now, it seems to me quite
fair to note that Harris, in his work, seems to have also left out SOME
of the same things. Following the same sequence of topics, we can make
the following observations:

(a) Harris does discuss the fact that constructions may have meanings
(_Methods in Structural Linguistics_ [MSL] 347), with reference to
Bloomfield's notion of tagmemes (_Language_ pp. 166, 276), and with a
cross-reference to his earlier discussion of word order as a morphemic
element (MSL 184). In his 1946 'From morpheme to utterance' paper,
Harris notes that "the formulae also cannot in themselves indicate what
meanings are to be associated with the various positions or classes",
adding that such information has "to be given in separate statements
accompanying the formulae". However, I have found no examples showing
what the form of such statements of meaning might be, and he does not in
practice refer to the meanings associated with e.g. his formulae for
Moroccan Arabic in MSL. In fairness, we must acknowledge that Bloomfield
also only rarely refers to the meanings of constructions.

(b) in his formulae for Moroccan Arabic morpheme sequences he provides
only for BINARY coordination:

 "Any morpheme class or sequence plus _u_ [that's 'and'] plus
 an equivalent morpheme class of sequence equals the morpheme
 class or sequence itself"

In other words, X -> X _u_ X

(Just as in all early transformational work, you can perhaps get
more-than-binary coordination by applying the rule again and again,
although recursion is not discussed here, but in any case you do
not get unbounded branching.) I have found no discussion of
unbounded branching anywhere in Harris's work, while the binarist
formulation recurs (e.g., in his 1946 paper "From morpheme to
utterance", where he gives a little sketch of English, and his 1963
paper "Immediate-constituent formulation of English syntax"). The 1963
paper, to be sure, is avowedly a restricted sketch presented solely for
the purpose of comparing different types of grammars.

(c) Harris recognized discontinuity, but nowhere in MSL does he cite
discontinuous syntactic constituents (or as he would have called them
then 'morpheme sequences'), only simple discontinuous morphemes. In his
1963 paper on IC analysis of English, he does not give a discontinous
analysis of English inversion sentences such as _Is John coming?_;
Hockett, for example, gave such an analysis some years earlier.

The analysis in the 1963 paper defines eight basic sentence structures
(including the yes-no question sentence structure singled out here as
well as others involving discontinuities) as simply strings of
constituents represented as nonterminal class symbols. Rewrite rules are
implicit in the definitions given for the constituents, but not in the
definitions of the sentence forms themselves. Harris' definitions of
these structures describe the discontinuous constituents instead by
informal statements about words being "moved" from the positions they
have in the assertion structure. This formulation is to facilitate
comparison with string analysis and transformational analysis (the aim of
this 1963 paper).

(d) Harris recognized zero alternants of morphemes, and the possibility
that all the morphemes of a constituent could be represented by zero
alternants, but it is not clear that this amounts to recognizing zero
constituents. However, he does seem to talk about zero constituents on
p. 340. <<This is a discussion of a grammar-regularizing technique that
has no bearing on the issue. An element (constituent) is "voided" when a
morphemic segment that is actually present is not assigned to any
constituent. This may be done when the morphemic segment would not
otherwise be independent of certain other morphemes. --BN>>

(e) The question of word order is the most vexed. Harris discusses the
possibility of treating order as itself a meaningful morpheme (pp.
184-186) and gives examples in Bengali and in English. He discusses free
word order as follows (p. 184):

 Finally, there are cases where the order of morphemic segments in an
 utterance is free; i.e. the morphemes occur in any order, with no
 attendant difference in the larger contextual environment or in the
 social situation.

 ... [W]e may find it convenient, in particular languages, to treat
 order on a par with morphemic segments, i.e. as just another segment
 in the morphemic constitution of the utterance. [fn: ... The new
 morphemic elements of order would consist of the relative order of
 these morphemic segments ... relative to each other in the utterance.]

He gives a Bengali example (fn. 61) in which a difference of order is
analyzed essentially as an allomorph of a "subordinate particle"
morpheme. In Moroccan Arabic he allows Adverbs to occur "at any point",
but there is no provision for this in his formulae. In fact, in his
discussion of the Morrocan formulae, he explicitly notes that in some
cases the actual word order is not correctly captured by the formulae.
For example he says: "N5V3 does not always mean that the N5 precedes".
This is the formula for NP VP basically [for noun subclass 5 and verb
subclass 3]. The problem is that he treats the subject prefixes and
suffixes on the verb as N5, and the suffixes at any rate obviously do not
occur before the V3 sequence. Likewise, on p. 355-356 he explicitly
talks about the diagrams he has introduced as sometimes giving only one
order of several that occur. In each such case, the orderings not
explicitly shown are provided for in "separate statements accompanying
the formulae".

This may be illustrated by another example of morpheme order not being
directly reflected in the formulae, namely, the position of t for tense
in the English sentence forms, e.g. N t V P N. Some members of the class
t occur as indicated, before V, e.g. "will":

 She will skate on it.

Other members of the class t occur after V, e.g. -ed:

 She skated on it.

Both orderings are represented by the class sequence N t V P N plus rules
("separate statements accompanying the formulae").

A clue to Harris' intentions here may be found in his discussion of
discontinuous phonological and morphological elements, where a mark
indicating the element may be written at the place where one of the parts
of the element occurs (generally, the first one), and the placement of
the other parts is specified by rule ("in separate statements
accompanying the formulae"). Thus, for a devoicing component over
English consonant clusters (MSL p. 134):
 _ _ __ _
 aez'be zd z vs. aez'bezdz "asbestos"

The alternants of a given component may differ considerably in form,
especially in the later, recursive stages of Harris's analysis. The
phonemic long component for devoicing may vary in length; another
phonemic long component may vary in its phonetic "substance" over its
length, or in different environments, i.e., the alternants may differ.
Similarly for morphemic long components and discontinuous morphemes.

It is a small step from this treatment of the several parts of a
discontinuous element to recognizing that alternative positions of the
various members of a morpheme class, e.g. Moroccan Arabic N5 or English
t, may be specified by rule, given the class-sequence in which the class
occurs (e.g. Moroccan Arabic N5V3 or English N t V P N). The rules (the
"separate statements") are an important part of how Harris' formulae
account for morpheme order.

This sort of complication of the descriptive statement occurs whenever
some particular sequence has a particularly free order, when two
sequences that are mutually exclusive occur in different places, and in
some other kinds of cases. A clear analogy can be drawn to the familiar
and relatively uncontroversial descriptive expedient of selecting one
morpheme alternant as basic and deriving the others from it by
morphophonemic rules: one ordering is selected as basic, that is, to be
represented explicitly in a syntactic formula, and the others are derived
from it by a rule of "extended morphophonemics" (Harris's term for it at
least as early as the late 1960s). In this way Harris's formulae and
associated statements were intended in combination to capture both the
cooccurence facts about morphemes and morpheme sequences and their order
of occurrence.

However, it is not easy to bring Harris's formulae into "alignment," as
it were, with PSG, so that they may be compared. Harris says explicitly
(e.g. in _Mathematical Structures of Language_, 1968) that he is not
interested in developing language-like mathematical systems like PSG and
trying to find some relation between them and language; he is interested
in describing language as a mathematical object.

-=+=-=+=-=+=-=+=-=+=-=+=-=+=-=+=-=+=-=+=-=+=-=+=-=+=-=+=-=+=-=+=-=+=-=+=-

Some addenda of my own follow:

If we are not to indulge in Whig history, we must some investigate what
the intentions were, severally, of Bloomfield, of Harris, of Bloch, of
Wells, and of Chomsky. Hilary Putnam has confirmed, in response to a
letter from me, that constituent structure was not featured in Harris's
seminars that Chomsky attended.

 My recollections about Harris' interests in 1946 support yours. What
 he was talking about and was very excited about in the seminars was
 what you described as the beginnings of transformational analysis.
 He did not mention, so far as I can recall, this "equivalence classes
 under substitution tests of morpheme class sequences" [that is,
 constituent structure] although he does of course return to that in
 some of his later work. I don't remember if he was already
 interested in discourse analysis, although of course much later he
 spoke of that. (Phone mail of 4:30 6/15/94)

There is no need to impute to Harris blame for Chomsky's "watering down"
Bloomfieldian immediate-constituent theory. It seems to me that there
are at least two motivations for it, one polemical and the other
theory-building.

As we know, Chomsky is a master of reframing the terms of debate onto
ground that is rhetorically favorable for him. I concur with Alexis'
claim (as I understand it) that Chomsky formulated PSG in the way that he
did essentially as a straw man. I suggest that Chomsky never intended in
good faith for PSG to represent IC analysis, or Harris's analysis by
substitution tests (possibly Chomsky never understood the distinction!),
or Wells' restatement of Harris' formulation as "expansions". I suggest
that PSG was a straw man from the outset, for the sake of promoting
transformational grammar. And if you can get your "opponents" (but would
I suggest that Noam was a combative and ambitious man?) to defend
themselves from your attacks on PSG, because it is after all a subset of
what they do, or looks like it, then in that measure you get them to
identify with it. A most effective tar-baby!

But PSG is a straw man only as regards polemic, for as Chomsky learned
about transformational analysis from Harris, he evidently saw that a
simplified version of IC analysis just might suffice for the much
simplified syntax of kernel sentences, with things like variable word
order, zeroes, discontinuities, unbounded branching, perhaps complex
categories (if he believe them descriptively relevant), and even
constructional meanings, being brought in through transformations defined
as operations on the trees (or labelled bracketings) produced in the PSG
base. Much could be said about why and how this formulation of
transformations as operations on trees--fundamentally different from
Harris's work--was a Bad Move that did linguistics no great favor, and
perhaps one day I will have the leisure to develop that theme.

Harris says that his morpheme-to-utterance procedure is not IC analysis
(1946, fn. 9):

 [T]he analysis into immediate constituents requires a technique
 different from that used in this paper, a technique based on
 comparing the apparent structures of utterances and parts of
 utterances. In this paper, on the other hand, we seek to arrive at a
 description of the structure of an utterance, without having any
 prior way of inspecting these structures or of saying whether two
 utterances are equivalent in structure. Therefore, the analysis into
 immediate constituents is not used here. ...

Bloomfield's method requires intuitive perceptions of the structure of
utterances and of parts of utterances. In 1933, Bloomfield was sanguine
that to anyone who takes the trouble just to ask the question, the
answers will be obvious (p. 161):

 Any English-speaking person who concerns himself with this matter, is
 sure to tell us [the immediate constituents of a sentence].

But as anyone knows who has taken the trouble to ask naive native
speakers about their judgments of constituency, this is, despite
Bloomfield's initial optimism in 1933, far from a transparent matter, and
even (perhaps especially so) with native speakers who are professionally
concerned, disagreements arise, hence, doubtless, at least some of the
"viva voce discussions" to which Wells alludes in his 1947 paper
"Immediate Constituents" (fn. 1).

Note that this is a methodological distinction, invisible to one who
merely compares Harris's results and Bloomfield's in Chomsky's PSG terms.
The problem is the fundamental historiographical obligation not to apply
the conceptual framework of one period to workers of another, and not to
set aside the concerns and aims that occupied them at the time. This is
difficult because it is a question of background assumptions and common
knowledge, to question which goes against the naive uniformitarianism by
which we daily make sense of our world.

-=+=-=+=-=+=-=+=-=+=-=+=-=+=-=+=-=+=-=+=-=+=-=+=-=+=-=+=-=+=-=+=-=+=-=+=-

Unfortunately, I haven't the "bandwidth" to write some sort of summary
and conclusion. The only antidote for the distortions of partisan whig
history is a working knowledge of methods and results that present real
alternatives. Sorting out real methods and solid results from mere
handwaving is a good purpose for historiography, surely a nobler and
more useful mission than the usual monolith-polishing apologetics.
Mail to author|Respond to list|Read more issues|LINGUIST home page|Top of issue