Aronoff Face the facts - the Stony Brook Linguistics Department

advertisement
Mark Aronoff
Stony Brook University
Face the facts:
Reading Chomsky’s Remarks on Nominalization after forty
years*
1. Introduction
This short article is a response to a recent reading of Noam Chomsky’s
celebrated article, Remarks on nominalization (henceforth Remarks),
published in 1970, as well as the article that first picked up on
Chomsky’s analysis of nominalizations in Remarks in a serious way,
Alec Marantz’s No escape from syntax: don’t try morphological analysis
in the privacy of your own lexicon (1997).1 One point that Remarks
demonstrates is that the progress of a field is beyond anyone’s control:
the lesson that an author wishes to teach and the lesson that the field
learns are often far apart. Of course, the same may apply in the
present instance and perhaps I am mistaken about the lesson of
Remarks, but we may have to wait another forty years to find out.
I will concentrate here on argumentation, in particular the interplay
among facts, theory, description, and analysis as presented first in the
argumentation in Syntactic Structures and later in Remarks. The four
components have very different roles in the two works and it is this
contrast that I find so interesting. For me, and many others, what was
*
Thanks to Florence Villoing for inviting me to contribute this
article, and to Françoise Kerleroux, whose work I have long admired,
for providing the occasion. This article arose from a graduate
morphology course in Fall 2011 in which we went through classic
works in generative morphology. I am grateful to the members of that
class for very helpful discussion that modified my own thinking
considerably: Ginny Anderson, Emily Benjamin, Amanda Caffary,
Andrew Canariato, Sophia Kao, Hijo Kang, Yu-an Lu, Vahideh Rasekhi,
Aydogan Yanilmaz, and Oxana Yudina.
*
1
The Marantz article was published over a decade ago, has been
widely read,and has been influential, mistakenly, as I will show
towards the end of this article.
most important about Remarks was neither the theory, nor the
analysis, but the facts presented and described. Both of these, which
involved fine lexical semantic differences between distinct words and
even distinct senses of the same word, have had great influence on
many researchers working in word formation (derivational
morphology) ever since.
The analysis, which posits that the lexicon may contain items
unspecified as to lexical category whose category is provided only in
syntactic context, has had influence only within the school of
Distributed Morphology, which relies heavily on the notion of
categoriless roots. The theory in Remarks is much less developed,
except to the extent that it seeks to integrate lexical and grammatical
facts within a single account while allowing for non-transformational
solutions for certain types of lexical data. I will show that the theory
and analysis provide little insight into the data described and that at
least some of the data used to support the analysis contradicts the
theory.
By contrast, the importance of the analysis and theory in Chomsky’s
earliest book, Syntactic Structures, and their influence on the field
were much greater than were the theory and analysis of Remarks.
What matters most about Remarks are facts and description. My
overall point is that sometimes the balance of progress is driven more
by description and facts, sometimes more by analysis, and sometimes,
but much less frequently than we would like to believe, by theory. In
the end, though, what wins out is understanding. When we feel that a
work has helped us to understand a phenomenon, then we value the
work. Of course, understanding is entirely unmeasurable and there’s
the rub.
Linguists tend to devalue description in comparison to theory and in
most disciplines theoreticians stand highest in the pecking order. But in
actual practice, it is more often the describers who gain the greatest
honor. Since the turn of the millennium, the Nobel prize in physics, the
science universally considered to be the most theoretically driven, has
been awarded only three out of eleven times for theoretical work. Most
recently, in 2011, the award went to three scientists for the discovery
of the accelerating expansion of the universe (for which there is no
generally accepted theoretical explanation); in 2010, it went to two
scientists “for groundbreaking experiments regarding the twodimensional material graphene.” Still, even in physics, although they
don’t usually win the big prizes, theoreticians enjoy the greatest
esteem, especially the best ones.
2. Data, description, analysis, and theory in modern linguistics
Noam Chomsky is the premier theoretical linguist of modern times. He
emphasized the role of theory, and by doing so restored grammar and
discussion of grammar to the center of linguistics. In North America
before Chomsky, modern linguists saw their primary task as the
scientific analysis and description of languages, not the construction of
grammars or theories of grammar. They took pains to distinguish their
work, which they termed descriptive, from the prescriptions that had
preoccupied grammarians for the three centuries since the scope of
grammar had been widened beyond Latin and Greek to include the
vernacular languages. Accordingly, they avoided the term grammar
precisely because of its prescriptive overtones, overtones evidenced
from the beginning, as in the OED’s 1640 quote from Ben Jonson’s
English Grammar: “Grammar is the art of true, and well speaking a
Language.” Leonard Bloomfield, for example, on the very first page of
his Language, the bible of American descriptivism, distinguishes the
very new field of “Linguistics, the [scientific] study of language” from
“the ‘grammar’ and other linguistic instruction in our schools”
(Bloomfield 1933, p. 1).
Bloomfield’s most influential European precursor was Ferdinand
de Saussure, who certainly never shied away from the term
theory. Unlike the Americans, he also used the term grammar,
but much less centrally and a little oddly. For Saussure, any
complex system involving what he called values could be
described by a grammar, not just a language. In his posthumous
Cours, which follows fairly closely the sequence of his lectures,
discussion of the term grammar is delayed until the middle of
the book, Chapter VII of the section on synchronic linguistics
that forms the heart of the work, in a paragraph that starts as
follows:
Static linguistics, or the description of a linguistic state,
may be termed grammar in that very precise sense, by no
means uncommon, found in expressions like ‘the grammar
of chess’, ‘the grammar of the stock market’, etc. These
are all cases of complex systems, involving coexisting
values. (de Saussure 1916/1983, p. 185)
Chomsky, however, as noted above, has always viewed theory and
grammar as central to language in particular, not to just any socially
constructed system. Beginning with the title of his very first major
work, The Logical Structure of Linguistic Theory, completed in 1955,
Chomsky presented linguistic theory as his first and greatest
preoccupation and he reclaimed the term grammar in the very first
pages of that same work to denote the formal description of a
language, as well as on the first page of Syntactic Structures, the 1957
précis of the 1955 book that had begun as lecture notes for
undergraduates and that made his name in the field. Chomskyan
syntax is still known by the names generative grammar. By contrast,
the term descriptivist linguistics quickly came to be seen in opposition
not to traditional grammar but rather to modern Chomskyan
generative linguistics, although Matthews (1993) showed much later
that Chomsky’s work was deeply rooted in the American descriptive
tradition.
Ideological overtones aside, though, theory and description are never
opposed in practice, since they are inseparable in any discipline: no
facts or description exist outside a theory and an analysis. Theory,
analysis, and description form the three sides of a triangle and all
three serve together to account for facts. As has often been
emphasized, all are completely interdependent (including the facts!),
though individual works and scholars in any field may differ in their
emphasis on one or the other. Rhetorically, the most powerful way to
argue for a theory is to show that it provides an explanation for some
set of facts, where the explanation consists of an appealing description
couched within some analytical framework that is compatible with the
theory. Newton, for example, combined his theory (Newton’s laws)
with the analytical methods afforded by the calculus to describe
velocity and acceleration. Without the analytical tools of the calculus,
the theory would have not been capable of providing the deeply
satisfying description of the physical world that won over the scientific
community of his day and persists even today.
2.1.
Description, analysis and theory in Syntactic Structures
Within modern linguistics, the greatest example of how the
convergence of theory and analysis provide a description of facts so
convincing that it changed the entire field is the presentation of
English verbs in Chapter 5 of Syntactic Structures. This book is widely
acknowledged as among the most influential works of 20th-Century
cognitive science.
The heart of Syntactic Structures consists of a single long argument
about the form that a grammar must have if it is to describe the facts
of English in a revealing manner. Chomsky first shows that a onedimensional grammar that treats a sentence simply as a string and
moves from a given word to the next without recourse to a wider
context cannot describe English, because of dependencies between
words that are not directly adjacent to one another, as they need to be
in one-dimensional grammars. Chomsky next discusses phrase
structure grammar (or immediate constituent analysis in the
terminology of American structuralists), in which words are grouped
into larger phrasal units, showing that such a grammar is capable of
describing phenomena that the simplest grammars cannot handle,
because it can refer to phrasal nodes in a two-dimensional tree.
Chomsky then adds to phrase structure theory a third dimension,
transformations, which reorder strings of elements that are produced
by the phrase structure grammar. He notes that, although it is possible
to describe in purely phrase- structure terms certain of the phenomena
that he describes by means of transformations, the resulting grammar
will be “extremely complex, ad hoc, and ‘unrevealing’” (Chomsky
1957, p. 34). Very quickly, we move to the description and analysis of
English main and auxiliary verbs that made Chomsky famous, at
whose heart lies a transformation that later came to be called affixhopping. Chomsky’s basic insight was that, no matter how complex a
particular English verbal expression, every auxiliary verb determines
the form of the verb that follows it. Thus, if we compare John has read
the book with John has been reading the book, the auxiliary verb have
in both sentences is followed by a past participle form; simply put, the
auxiliary verb have must be followed by a past participle. Similarly, a
modal verb (can, may, must, etc.) will always be followed by a bare
verb form: John may read the book vs. John may have read the book
vs. John may be reading the book; and the progressive auxiliary be
must be followed by a present participle. Chomsky accounts for this
relation between an auxiliary and whatever follows it in two steps: first
with a phrase structure rule that pairs each auxiliary with an abstract
affix (have-en, be-ing, etc.), and then with a transformation that
switches each abstract affix with the following verb, resulting in a
verb-suffix combination that can then be spelled out.
The phrase structure rule works as follows:
Aux  Tense marker (Modal verb) (have + en) (be + ing) (be + en)
The transformation switches any of the bolded items (all of which are
abstract affixes) with the following verbal element and the result is
exactly what we need: each verbal element receives the proper
abstractly affixed form that is determined by the preceding auxiliary, if
there is one. Furthermore, the first element in the verbal cluster will
always bear the tense marker, which is also correct.
This description and analysis of English verbs is so beautiful that it won
the day for the theory, even though Chomsky admitted from the outset
that the whole system could be described using phrase structure and
selection, which Gazdar, Pullum, and Sag (1982) later did and which
Chomsky himself even advocated at one point (Chomsky 1995). But
what makes the affix-hopping analysis stand out is the beauty and
simplicity of its description. That is what made us all accept it as an
explanation and it was one of the principal causes of Chomsky’s
success. In other words, the beautiful analysis and description of some
very striking facts was the rhetorical weapon that drove the
acceptance of the theory, even though it is the theory that is most
remembered.
2.2.
Fact and description in Remarks
Moving ahead a dozen years or so, we come to 1970 and Remarks on
Nominalization, which appeared in a collection of original papers meant
to demonstrate developments in generative syntax in the brief period
since Aspects had been published in 1965, a period characterized
largely by the generative semantics wars (Harris 1993). Many of the
articles in the book reflect this war. Some were popular and influential
in their time but only this one piece achieved the status of a true
classic: it has been cited thousands of times in the literature. In some
sense, Remarks won the war for Chomsky.
The heart of Remarks consists of an analysis of English derived
nominals within a theory of lexical categories. This analysis has had
little influence, though the theory has survived in a very different
context. The lesson that most readers took away from Remarks was
inspired neither by the analysis nor by the sketch of a theory that
Remarks supplied, but instead by the description of the data, which is
more detailed. Much later, Marantz excoriated those who learned this
lesson. The lesson was correct, however, as I will show below. Spencer,
in his survey of morphological theory, puts it as follows: “The primary
importance of this paper for morphology was that it pointed to the
need for a separate theory of derivational morphology, distinct from
the theory of syntactic transformations” (Spencer 1991, p. 69).
Marantz calls this “an interestingly contrived reading of Remarks.”
(Marantz 1997, p. 214).
Both Marantz and Spencer are right. The actual analysis of
nominalizations that Chomsky supplied in Remarks was very different
from the lesson most of us took away from the article. The editors of
the volume characterize this analysis succinctly in their preface:
In the past, certain kinds of noun phrases had been related
to sentence structures bearing the same kinds of semantic
relations, by means of transformations which moved
strings from a deep structure, through a series of
derivations, to a surface structure noun phrase. In the
course of this, verbs, for example, were converted into
nouns. But, argues Chomsky, there is no a priori way of
knowing whether the noun forms underlie the verbs or vice
versa. Instead of relating alternative forms through
transformations, the base rules should be extended to
accommodate the alternative forms directly. This is the
lexicalist position…(Jacobs and Rosenbaum 1970, p. x)
This general position was preferable to what Chomsky called the
“transformational position,” which involves “some extension of the
transformational apparatus.” The general lexicalist strategy, extending
the base rules to accommodate derivational morphology directly, led to
a variety of theories that separated morphology from syntax, as
Spencer correctly notes, many of which still flourish today. Chomsky’s
specific analysis, however, extended the base rules in a way that never
caught on until it was picked up by Marantz (1997), and for good
reason: because it was very limited and never accounted in any
insightful way for the phenomena that Chomsky described. And if one
extends the data just a little bit, it fails completely.
2.3.
Underspecification in Remarks
Remarks is a very rich paper and many of the theoretical innovations
proposed in it have had wide repercussions in syntactic theory, most
prominently X-bar theory, which expresses common structures across
phrase-types, and the extension of the scope of the passive
construction to even simple nominals like Aristotle’s portrait by
Rembrandt. But I will concentrate here on the topic that lies at the
core of the article, the difference between what Chomsky called
(somewhat peculiarly, given previous use of the term gerundive in
traditional grammar) “gerundive nominalizations” and “derived
nominals”. His basic observation was that, while the meaning and
syntax of the gerund construction was always transparently related to
the meaning and syntax of the corresponding verb, the same could not
be said of the corresponding derived nominals:
The idiosyncratic character of the relation between the
derived nominal and the associated verb has been so often
remarked that discussion is superfluous. Consider, for
example, such nominals as laughter, marriage,
construction, actions, activities, revolution, belief, doubt,
conversion, permutation, trial, residence, qualifications,
specifications, and so on, with their individual ranges of
meaning and varied semantic relations to the base forms.
There are a few subregularities that have frequently been
noted, but the range of variation and its rather accidental
character are typical of lexical structure. To accommodate
these facts within the transformational approach
(assuming, as above, that it is the grammatical relations in
the deep structure that determine meaning), it is
necessary to resort to the artifice of assigning a range of
meanings to the base forms, stipulating that with certain
semantic features the form must nominalize and with
others it cannot. Furthermore, the appeal to this highly
unsatisfactory device, which reduces the hypothesis that
transformations do not have semantic context no near
vacuity, would have to be quite extensive. (Remarks, p.
189)
This description of derived nominals provides the empirical basis for
the lexicalist hypothesis and it has never been refuted. The analytical
question that Chomsky faces is, if we can’t use syntactic
transformations to deal with derived nominals, what device do we use
instead to relate them to the corresponding verbs, given “the
idiosyncratic character of the relation between the derived nominal and
the associated verb?” Chomsky’s solution is as follows:
Let us propose, then, as a tentative hypothesis, that a
great many items appear in the lexicon with fixed
selectional and strict subcategorization features, but with a
choice as to the features associated with the lexical
categories noun, verb, adjective. The lexical entry may
specify that semantic features are in part dependent on
the choice of one or another of these categorical features.
This is, of course, the typical situation within the lexicon;
in general, lexical entries involve certain Boolean
conditions on features, expressing conditional
dependencies of various sorts. Insofar as there are
regularities, these can be expressed by redundancy rules
in the lexicon. (Remarks, p. 190)
In simpler terms, the idea here is that the lexicon will contain an entry,
LAUGH, for example, that is not specified as either verb or noun, but
that will be given either a verb or noun category and form (laugh vs.
laughter) when it is inserted in an appropriate spot in the syntactic
tree. It might even be that the noun and verb are semantically
distinct, which can be encoded in the lexical entry by means of
Boolean conditions on features (presumably of the sort if noun, then
semantic feature q). This solution generated much less interest in the
quarter century following than the description of the problem laid out
on the previous page, in part because it was not worked out in any
detail in Remarks. Chomsky already began to find problems with the
solution in the footnote to the very paragraph in which he proposed it,
noting that it is not clear “whether a lexical entry is regarded as a
Boolean function of specified features or is to be replaced by a set of
lexical entries, each of which consists of a set of specified features”
(Remarks, fn 12, p. 217) and whether these two approaches are
empirically distinct. And ten pages later in the article, he remarks that
“only in the simplest case will exactly the same contextual (and other)
features be associated with an item as a verb and as a noun. In
general, lexical entries involve sets of shared features, organized in
complex and little understood ways, and we should expect to find the
same phenomenon in the case of derived nominals, given the lexicalist
hypothesis.”
2.4.
Underspecification meets the facts
In fact, the idea that a pair of lexical items sharing the same stem
might be distinct only in their category label is directly contradicted by
the extensive set of derived nominals that Chomsky had just cited on
the previous page and by his remark that “the range of variation and
its rather accidental character are typical of lexical structure.” If the
distinction were only in the category label, then one might want to
resort to the sort of transformational solution that Chomsky had
advocated for gerunds, which are nouns that truly share all the
relevant properties of the verbs that they are formed from. But the
same comment that dooms a transformational account applies equally
to an account involving semantically underspecified categoriless
elements that receive specification in context: “it is necessary to resort
to the artifice of assigning a range of meanings to the base forms,
stipulating that with certain semantic features the form must
nominalize and with others it cannot,” a stipulation that, Chomsky
reminds us, “reduces the hypothesis that transformations do not have
semantic context to near vacuity.” Instead of transformations, we have
lexical insertion, but the comment about vacuity applies equally.
Underspecification of the meaning of an item out of context (which is
what categorilessness demands), followed by specification in individual
lexical contexts, can never provide an explanation: it is an artifice
entirely equivalent to simply listing the meaning of the full expression
in the first place. If the meaning of an expression A + B is
unpredictable from the meanings of the expressions A and B but must
rather be stipulated as a property of the complex expression A + B,
sometimes completely, as Chomsky notes in the case of derived
nominals, then underspecification of A or B will never provide a
satisfying account of the stipulated differences, only a means of hiding
the truth.
Towards the end of the article, Chomsky mentions and then does not
discuss at all a type of nested structures that underspecification can
have nothing at all to say about: “As is well-known, processes of
derivational morphology are applicable in sequence – they may even
be recursive.” (Remarks, p. 212). The examples he mentions are
readable and readability (derived in sequence from read) and selfindulgent and self-indulgence (derived in sequence from indulgent). He
then goes on to argue for two pages against a transformational
analysis of these pairs. But nowhere does he discuss the problem that
these nestings present for his underspecification analysis.
Nesting of derivational affixes is quite common in English and in many
other languages. Consider the five following examples. All end in the
deadjectival abstract nominal suffix –ness but each contains a different
denominal adjective suffix. The general pattern is X]N Y]A ness]N. Each
example is drawn from a large set of words:
quirk]N i]A ness]N
boor]N ish]A ness]N
matron]N li]A ness]N
hope]N less]A ness]N
pain]N ful]A ness]N
Each example represents a triplet of words like quirk, quirky,
quirkiness. We cannot use underspecification to handle these triplets.
Let’s say we wanted to relate quirk and quirky through
underspecification: the basic entry will be QUIRK and it has no
category. When QUIRK occupies a noun slot, it gets no suffix and is
spelled out as quirk, but when it occupies an adjective slot it gets the
suffix –y. Now what do we do with quirkiness? It is based on a
category-bearing word, quirky. We can’t claim that quirky bears no
category, because we just gave it one. We might instead try to claim
that quirkiness contains a complex suffix –iness, based on the
categoriless QUIRK. But that would force is to say that –liness, lessness, -ishness, and –fulness and any other suffix combination are
all complex suffix. And what would it mean anyway to call something a
complex suffix? Does it have internal parts? Are they compositional?
How do these suffixes relate to the simple suffix –ness that occurs in
words like redness? The answer is that a theory with
underspecification of lexical category cannot handle nested
morphology.
Even when there are no affixes, it is still possible to show that
unspecified forms will not do the job. Consider the two noun/verb
zero-derived pair types exemplified by pilotn/pilotv and runv/runn. Both
Clark and Clark (1979) and Aronoff (1980) discuss the first pair at
great length, with very different analyses. Marchand (1969) has
discussed the second more briefly. Basically, the verb in the first
construction denotes some activity that has something to do with the
noun, so that the noun must be treated as basic in order to get the
semantics to work right (whatever one’s analysis). The second type,
which is much less productive, goes in the opposite direction. Here the
verb sense must be basic and the noun means, in Marchand’s words
‘single instance illustrative of the active process’ of the verb (Marchand
1969, 374). Marchand notes that the verb is usually intransitive. If one
operates with categoriless elements, then there is no way to even
begin to discuss the difference between the two types. And then there
are recursive cases like walkv/walkn/walkv, in baseball, where the last
sense means ‘To force in (a run) by giving up a walk with the bases
loaded’, as in Marshall walked home two more runs. If one tries to
derive the transitive sense directly from the intransitive, then one has
to deal with the fact that the object of the transitive is the inanimate
noun run. But an analysis in which the noun is derived from the first
verb as an instance and then the second sense of the verb is an
activity having something to do with the noun, makes perfect sense.
Of course, it is cases like these that led people to begin to work on
derivational morphology as a grammatical module whose function is to
form and analyze what Chomsky calls “the lexical categories N, A, V.”
(Remarks p. 210) There are too many theories of derivational
morphology to discuss here, but all share the basic feature of being
designed to deal with “the idiosyncratic character of the relation
between” any two lexical items.
3. Marantz’s analysis
Almost thirty years after the publication of Remarks, Alec Marantz
revived Chomsky’s analysis of English derived nominals in a paper
arguing against the notion that derivational morphology was a
separate module from syntax. Marantz defines a straw-man variety of
lexicalism as follows:
Lexicalism: words are created in the Lexicon, by processes
distinct from the syntactic processes of putting
morphemes/words together. Some phonology and some
structure/meaning connections are derived in the lexicon,
while other aspects of phonology and other aspects of
structure/meaning relations are derived in (and after) the
syntax. (Marantz 1997, p. 201)
According to Marantz, “Lexicalism claims that the syntax manipulates
internally complex words, not unanalyzable atomic units.” Marantz
provides in this paper what he bizarrely calls “the alternative that
allows us to dump lexicalism once and for all” (Ibid. p. 202). What
allows Marantz to ‘dump’ lexicalism is Chomsky’s analysis of derived
nominals in Remarks, but formulated within the theory of Distributed
Morphology, which incorporates three separate ‘lexicons’. First is a
lexicon of “atomic roots” that are similar to the unspecified “base
forms” of Remarks. There is a second lexicon, which Marantz calls the
vocabulary, where roots and complex entities apparently receive their
phonological forms, and a third, which he calls the encyclopedia: “The
Encyclopedia lists the special meanings of particular roots, relative to
the syntactic context of the roots, within local domains.” (Ibid. p. 204).
This last is equivalent to Chomsky’s listing of derived nominals and
their meanings and suffers from the same fatal conceptual difficulties,
which I will not rehearse here. In this theory “the grammar constructs
all the words in the syntax by the same general mechanisms” (Ibid. p.
205), so that there is no special morphological component. Marantz
notes first, in defense of this claim, that words have no particular lock
on idiosyncratic meaning. Phrases may also be “in the lexicon.” This is
true, but it has nothing to do with lexicalism, which has long noted
that the term lexical has two senses (Aronoff 1983, 1988; Disciullo
and Williams 1986; Williams 1981), one meaning ‘idiosyncratic or
unpredictable’, the other meaning ‘word internal’, which denote two
quite different things. Marantz confuses these two senses. The
lexicalist hypothesis that most of us use says only that what goes on
within the lexeme or word is different from what goes on outside it and
has nothing to say about unpredictability:
The lexicalist hypothesis, at least in its more common
weak version . . . is a hypothesis about syntax and not
about morphology. It consists simply of drawing a dividing
line along the syntactic dimension between lexeme
formation and supralexical syntax. To draw this line is to
claim only that syntax operates differently in the two
domains, above and below it . . . (Aronoff 1994, p. 15)
Williams (2007), puts the same hypothesis slightly differently:
The Lexical Hypothesis is about the organization of the
grammar into modules. It suggests that the system of
words in a language is independent of the system of
phrases in a language in a particular way. It is independent
of it, but communicates with it through a narrow channel—
the “top-level” properties of words. (Williams 2007, p.
353).
For Marantz, though, the line is not at the word or lexeme, but at the
root: “Things with special meanings are roots . . . Words are usually
also identified by their root content, where the roots are the items
subject to special contextual meanings.” (p. 13). Here is where the
analysis of Remarks comes in. Marantz starts his analysis by noting
that the idea that lexical items are complex items has its origins in Xbar theory. This is simply wrong. Chomsky introduced this idea in the
last section of Aspects (1965), which also contains the beginnings of
the analysis provided in Remarks. That aside, Marantz provides the
following “up-date” of Chomsky’s Remarks theory:
Nominalizations like “destruction” and “growth” . . .are
never “verbs” at any stage in the derivations, and thus DPs
. . . are not transformationally related to sentences… Roots
like DESTROY and GROW are category neutral, neutral
between N and V. When the roots are placed in a nominal
environment, the result is a “nominalization”; when the
roots are placed in a verbal environment, they become
verbs. (Marantz 1997, p. 215)
Marantz then recapitulates Chomsky’s argument for why the
nominalization growth is only intransitive, while the verb may be either
transitive or intransitive, while the opposite is true for destruction. This
is because “while certain operations cut across N-headed and Vheaded phrases, …. certain syntactic structures require the verbal
environment. In particular, the agent of transitive “grow” is …. a type
of causative agent projected only in a verbal environment.” (Ibid. p.
216).
3.1.
Two senses of grow
Marantz provides a complex syntactically based solution for why all
this should be the case. But, as I pointed out in the first work I ever
published (Aronoff 1974), the real problem here has nothing to do with
syntax, or even morphology. It is simply that the verb grow has two
quite distinct types of senses in Chomsky’s examples, which he did not
notice, while the noun growth has a sense corresponding to only one
of these.2 The intransitive verb can mean either GROW1 ‘exist as a
living plant in a specified environment’ (Corn grows well in this area)
or GROW2 ‘increase in size’: The corn grew a foot a day; his anger
grew by the moment. GROW1 has a related transitive similar in sense
to raise or cultivate: We grow corn in this area. GROW2 does not: *We
grew the corn a foot a day; *the incident grew his anger beyond the
point of containment. The nominal growth has many distinct senses
but is no longer used in a sense corresponding to GROW1: *the growth
of corn in this area (except in the expression the growth of crystals).
However, it is frequently used in a sense corresponding to GROW2
‘increase in size’, which we may call GROWTH2: the growth of the
economy.3 It also has a different, concrete sense GROWTH3 with no
clearly related verbal sense, ‘morbid formation’, as in his head was
covered with growths.
The table below expresses the facts:
GROW1 ‘exist as a plant’ GROW2 ‘increase in size’
intransitive (corn grows
here)
transitive (grow corn)
*nominal (the growth of
corn)
intransitive (corn grows
quickly here)
*transitive (*we grew
corn rapidly)
nominal (the amazing
growth of the corn)
GROW(TH)3 ‘morbid
formation’
*intransitive ??
*transitive ??
nominal (a growth on
her arm)
2
A quick glance at the OED reveals that grow has many senses,
as do rise and raise, discussed below, but it will suffice to discuss the
two senses in question here in each case.
3
Bill Clinton, in his 1992 campaign for President, used the
slogan grow the economy, which had the ‘cause to increase in size’
sense that was previously missing for GROW2 (OED Online has only
one citation for this causative sense, from William Caxton in 1481).
The causative sense has become more common in the last two
decades, as Emily Benjamin points out to me, but only with economic
terms as the direct object: jobs, business, etc. The intransitive verb
GROW2 and the noun growth have been used in reference to business
and the economy since the 1950’s. The wonderful term growthmanship
was popular in the 1960’s, but seems to have fallen into desuetude.
Chomsky is puzzled (Remarks 214) about why the growth of tomatoes
is unambiguously intransitive, while the growing of tomatoes can be
either transitive or intransitive.4 He posits an analysis in which the
transitive sense of GROW1 contains a causative component, but the
noun growth does not contain a causative component, which, he
admits, “involves an ad hoc step”. But the explanation for the noncausative reading of growth is simple once we understand that the
non-causative of growth corresponds only to GROW2 ‘increase in size’,
which has no transitive sense, while the gerund in the growing of
tomatoes has sense GROW1 ‘exist as a plant’, which has a transitive
meaning roughly ‘cause to exist as a plant’. This explanation has
nothing to do with causative features, but simply with the differences
between lexical senses. It is a purely lexicalist account. We are in fact
dealing with homonyms and so need no real explanation at all.
More generally, underspecification of root meanings accompanied by
specification in individual lexical contexts can never provide an
explanation, because it is equivalent to simply listing the meaning of
the full expression in the first place. In the worst case, the analysis
forces one to assume that the ‘root’ in isolation has no meaning at all,
because its meanings in specific lexical contexts vary so widely. The
only gain is self-deluding systematicity: the analyst believes that he or
she has extracted a generalization where in fact none lies.
Mine is a very particular account that depends on paying attention to
the different senses a single phonological form may have, but Marantz
has no use for particular accounts, only explanations, and then only
those that “explain the systematic behavior of nominals” (Marantz
1997, p. 216). He therefore decides that there is a “grow” class of
verbs whose members are either transitive or intransitive but whose
nominals are intransitive, and a “destroy” class of verbs whose
members are transitive but whose nominals may be either transitive or
intransitive. This classification conflates the two senses of the verb
GROW in the same way Chomsky’s original analysis did and is
therefore misconceived from the start. Based on this misconception,
Marantz goes on to concoct a fairly elaborate account of the facts at
hand. He translates Chomsky’s analysis of the intransitivity of growth
by saying that the root GROW is non-agentive, so that, when placed in
a nominal environment it can have no agentive argument, but in the
verbal environment, “a syntactically projected agent may appear,
4
The transitive sense is clearly favored by most speakers,
possibly because GROW2 is blocked here by growth.
yielding “John grows tomatoes.” But, as we just saw, the sentence
John grows tomatoes contains an entirely different sense of GROW
from the growth of tomatoes. The two expressions are unrelated and
Marantz’s systematic explanation is a chimera because there are no
systematic facts to be explained here. Instead, what needs to be
explained is the lack of systematicity.
Marantz extends his explanation to other phenomena, but since the
legs have been cut out from under it by the facts, there is little use in
pursuing it further. And in any case, he persists in the same confusion
of homonyms, and sometimes near homonyms . For example, he
discusses the verbs raise and rise, noting: “I think one can say, ‘I
raised the crane two floors,’ with the internal-cause reading preserved,
paralleling then, ‘I grew tomatoes’.” (Ibid. p. 222). What is interesting
here is that the verb raise, unlike grow, does have two transitive
senses. One means approximately ‘cause to rise’ which has an
analogous intransitive rise, but the other meaning, ‘breed (animals)’ or
‘produce (meat, seafood, etc.)’, has no phonologically similar
intransitive analogue. Again, the two verbs have nothing to do with
one another, so that any attempt to provide a syntactic account for
why one has an intransitive analogue while the other does not (which
Marantz does for the ‘breed’ sense of raise), is doomed from the start.
As Andy Canariato points out, rise is irregular (rise, rose, risen), while
raise is completely regular, making it hard to claim that they are the
‘same’ lexical item on any theory. Furthermore, the idea that raise is
somehow a “special pronunciation in its transitive use” of rise might
seem to make some sense for the ‘cause to rise’ sense of raise but the
‘breed’ sense of the word bears no relation to ‘rise’. Marantz has once
again been fooled by phonological similarity. The problem with all of
these explanations is exactly the one I identified in the conclusion to
my note in 1974:
It is striking, however, that no one had noticed the
semantic discrepancy before. Chomsky’s paper is not
unknown, the argument based on the fact of (2) [*the
farmer’s growth of the corn] is almost as celebrated, and
yet all a person had to do was look up the words in a
reputable dictionary, or think about the words a long time:
grow – growth. But in this fact of academic history lies the
greatest confirmation of the lexicalist hypothesis. It is the
characteristic of syntactically related forms that the
semantic differences between them remain constant…
Schooled in this assumption, the linguist who finds a set of
morphologically and semantically related items goes on
calmly with his business, believing that whatever
differences there are among the members of the set will be
constant, predictable, and for his present purpose,
disregardable. It is only within the lexicalist hypothesis
that we are warned to look more closely at the semantic
differences. In fact this hypothesis tells us that
morphologically related items that are members of distinct
Deep Structure lexical categories may be related
semantically in a somewhat haphazard fashion. (Aronoff
1974, p. 162)
4. Conclusion
Syntactic Structures revolutionized the field because it showed the way
to a new kind of linguistic analysis. I would argue that Remarks, in the
end, will be equally important, but because it did not provide any
methodological solution, its effects are taking much longer to be felt.
And that is because the lesson of Remarks is not about analysis or
even about theory, but about the nature of what is to be described.
Remarks shows that language is sometimes messy and that a good
description, analysis, and theory, must accommodate the mess, not
just step around it or cover it over. Among those who are looking only
for generalization, idiosyncrasy is always something to be avoided. But
by avoiding idiosyncratic facts, we run the risk of explaining a mirage
of our one’s own making.
Finally, though, one might ask what drives someone to the sorts of
theories he espouses, all facts to the contrary. The answer, I believe,
lies in the quest for a grammar in the spirit of Syntactic Structures. If
the lexicalist hypothesis is true, then the idea of a unified grammar,
what Marantz elsewhere calls a “single-engine theory” is no longer
very useful. The principals and the methods that have turned out to
provide insight into the internal structure of words, especially
derivational morphology, are quite distinct from those that apply in
syntax. Morphology turns out to be as much about exceptions as about
rules. This does not mean that it is not amenable to rigorous inquiry,
only that it does not present the same sorts of patterns that we find in
syntax. Brown and Hippisley (2012), for example, show how the
exceptions form a highly structured system of defaults within a
computationally implementable framework. Remarks, for at least some
of us, ushered in a new era in linguistics, one in which the methods of
inquiry have been driven at least as much by the data as by the
theory. That is its greatest lesson.
Some theoreticians may be disappointed by my conclusion. What it
shows, though, is that linguistics has matured as a science to the point
where theories can be proven wrong, no matter how appealing they
are, not by argument or mud slinging, but by facts.
REFERENCES
Aronoff, Mark. 1974. The semantics of growth. Quarterly Progress
Report of the Research Laboratory of Electronics 113: 159-62.
M.I.T.
Aronoff, Mark. 1980. Contextuals. Language 56. 744-58.
Aronoff, Mark. 1983. Actual words, Potential words, productivity and
frequency. Proceedings of the Thirteenth International Congress
of Linguist 163-171.
Aronoff, Mark. 1988. Two senses of lexical. ESCOL 1988, 1-11.
Bloomfield, Leonard. 1933. Language. New York: Henry Holt and
Company.
Brown, Duncan and Andrew Hippisley. 2012. Network Morphology: A
Defaults Based Theory of Word Structure. Cambridge:
Cambridge University Press.
Chomsky, Noam. 1957. Syntactic Structures. The Hague: Mouton.
Chomsky, Noam. 1965. Aspects of the Theory of Syntax. Cambridge,
MA: MIT Press.
Chomsky, Noam. 1975 (1955). The Logical Structure of Linguistic
Theory. Plenum Press: New York.
Chomsky, Noam. 1995. The Minimalist Program. Cambridge, MA: MIT
Press.
Chomsky, Noam. 1970. Remarks on nominalization. In Readings in
English Transformational Grammar. Ed. by Roderick Jacobs and
Peter Rosenbaum. 184 – 221. Waltham, MA: Ginn and Company.
Clark, Eve, and Herbert Clark. 1979. When nouns surface as verbs.
Language 55. 787 – 811.
Saussure, Ferdinand de. 1983 (1916). Course in General Linguistics.
Translated by Roy Harrris. London: G. Duckworth.
DiSciullo, Anna Maria, and Edwin Williams. 1987. On the Definition of
Word. Cambridge, MA: MIT Press.
Gazdar, Gerald, Geoffrey K. Pullum, and Ivan A. Sag. 1982. Auxiliaries
and Related Phenomena in a Restrictive Theory of Grammar.
Language 58. 591-638
Jacobs, Roderick, and Peter Rosenbaum. 1970. Readings in English
Transformational Grammar. Waltham, MA: Ginn and Company.
Marantz, Alec. 1997. No escape from syntax: Don’t try morphological
analysis in the privacy of your own lexicon. Proceedings of the
21st Annual Penn Linguistics Colloquium: Penn Working Papers in
Linguistics 4: 2, ed. Alexis Dimitriadis et.al. 201-225.
Marchand, Hans. 1969. The Categories and Types of Present-Day
English Word-Formation. Munich: Beck.
Matthews, P. H. 1993. Grammatical Theory in the United States from
Bloomfield to Chomsky. Cambridge: Cambridge University Press.
Spencer, Andrew. 1991. Morphological Theory. Blackwell; Oxford.
Williams, Edwin. 1981. On the notions ‘lexically related’ and ‘head of a
word’. Linguistic Inquiry 12. 245 – 274.
Williams, Edwin. 2007. Dumping lexicalism. In Oxford Handbook of
Linguistic Interfaces, ed. by Gillian Ramchand and Charles Reiss.
355 – 382.
Download