|AUTHOR: Randall, Janet H.
SUBTITLE: The Geometry of Argument Structure
SERIES TITLE: Studies in Natural Language and Linguistic Theory 74
Eva Wittenberg, Center for Cognitive Studies, Tufts University
Theories about argument structure have a central goal: To find a key that
correctly predicts which semantic participants map onto which syntactic
positions, for any given verb. The problem with this undertaking is that there
is a large number of semantic participant classes, commonly referred to as
''semantic roles''. In syntax, however, the range of available argument positions
these roles compete for is quite limited. Consequently, the subject position can
be taken by an Agent (''He jumps''), by an Experiencer (''He enjoys the food''), by
a Theme (''It floats''), and so on. The question is: Which semantic participants
can go where, and under which circumstances?
Researchers have tackled this puzzle in various ways. One way is to allow for a
variety of distributions, regulated by a hierarchy of semantic roles (Jackendoff
1992), resulting in something like ''if there is an Agent, put it into Subject
position; if there is not, take an Experiencer'', and so on down the queue.
Another way is to severely restrict the number of semantic roles (Dowty 1991). A
third approach posits ''hidden levels'' within a syntactic tree that accommodate
all possible semantic roles (Hale & Keyser 1993, 2002; Chomsky 1995, Kratzer
1996). The latter theories are often called ''structure-preserving'' (Levin &
Rappaport Hovav 2005), since there is a perfect one-to-one mapping: each node in
a tree expresses exactly one kind of semantic role. The meaning, the argument
goes, can be read off the tree directly, resulting in something like ''the
Specifier of little v is an Agent''. In these theories, the key to Argument
Structure is put into hidden syntactic layers (see Horvath & Siloni 2002,
Culicover & Jackendoff 2005 for discussion).
This new book is addressed to researchers and graduate students who are also
looking for the key to Argument Structure. It is grounded in a Government and
Binding theory of syntax, but it also incorporates early work in predicate
decomposition (Jackendoff 1972). Basically, Randall advocates structure
preservation similar to Hale & Keyser (1993, 2002), but with different means. In
her approach, there are no hidden levels of syntax. Instead, the syntax of a
sentence is a product of a hierarchy within Conceptual Structure. So the key
lies within the semantic level of grammar.
What follows is a short outline of the theoretical model. Due to space
restrictions, not all verb classes Randall deals with can be elaborated on; but
the reader may rest assured that the book, as one reviewer put it, takes on a
whole minefield of them.
Randall decomposes Conceptual Structure (CS) by using only the primitives CAUSE,
BE, and BECOME, plus a few locative prepositions (and later VIA, for
manner-of-motion verbs). In that, she adopts a strict ''localist'' position,
stating that every verb can be stated in terms of motion or location (Jackendoff
1983, see Levin & Rappaport Hovav 2005 for discussion). States, in Randall's
words, are also ''nothing more than abstract PLACEs'' (p. 30). The CS for ''The
sun made the ice cream into a liquid'' would look something like (1):
(1) [CAUSE ([sun], [BECOME ([ice cream], AT [liquid])])]
This structure can be written like a syntactic tree, and the semantic roles in
the predicate can be read off of it: the first argument of CAUSE is always an
Agent; the first argument of BE or BECOME is always the theme; and so on (p.
31). Note that in most theories, (1) would already violate this generalization
because ''sun'' is not volitional or animate. At least one of these notions is
usually seen as a requirement for agency (Dowty 1991). Randall, however, does
not discuss the precise nature of semantic roles. Also not accounted for are
psych verbs like ''fear'' and ''love''; Randall does not address these verbs, but
she states that the lack of a convincing analysis of psych verbs is due to the
lack of a good CS representation, not due to flaws in the theory.
Still, the general idea is that the CS tree predicts the syntactic tree, because
their geometry is the same. Any structure in CS will surface in the syntax; so,
the theory is strictly structure-preserving without positing hidden levels in
the syntax. Few steps are needed in order for this to work.
First, between CS and syntax (''Deep Structure''), there is an Argument Structure
(AS) interface level which filters the CS arguments and determines which
elements are visible to syntax, similarly to Wiese's (2004) semantic interface
level. The filtering process abides by three rules. First, it strictly follows a
modified version of the Theta Criterion (Chomsky 1981): For every predicate,
each semantic argument licenses one syntactic argument, which in turn can only
satisfy one semantic role. The Theta Criterion is enforced by the Bound Argument
Condition, which states that if there is competition between two bound CS
arguments, only the higher one is allowed to link to an AS position.
The meanings of individual lexical items in a sentence are combined in a process
called ''Argument Fusion'', where the conceptual structures of arguments fuse with
the conceptual structures of their heads, until the meaning of the whole
hierarchy is complete. It is also through Argument Fusion that selectional
restrictions are obeyed, since only CS constituents that are compatible with
each other can be fused. Fusion cannot apply more than once to a CS constituent,
as stated in the Prohibition Against Double Fusion (PADF) constraint.
This machinery is best illustrated with the example of Resultatives, which are
extensively discussed in Chapter 5. Resultatives are the product of two CSs. A
base predicate, like ''to water'', is subordinated under a CAUSE predicate as its
first argument. The second argument of this new predicate is the result. The CS
of a sentence like ''The gardener waters the tulips flat'' then looks like (2):
(2) CAUSE (CAUSE (the gardener, BECOME(water, PLACE(on, the tulips)),
BECOME(the tulips, flat))
Argument Fusion links the meaning of each lexical item to its head. The Theta
Criterion is satisfied, since for every predicate, there is exactly one semantic
role. How is *''The gardener waters the tulips the tulips flat'' prevented? The
Bound Argument Condition kicks in, allowing only the higher instance of ''the
tulips'' to be expressed (in this case, it is the first instance).
There are two possible problems with this solution. One could choose a different
but equally plausible event representation, like ''The gardener causes the tulips
to go flat by watering them'', illustrated in (3):
(3) CAUSE(the gardener, BECOME(the tulips, AT(flat))), VIA(CAUSE(gardener,
BECOME (water, ON (the tulips))
The argument fusion mechanism could not apply anymore, since ''by watering the
tulips'' would be an adjunct sister of the outermost CAUSE.
Also, there is a conceivable CS structure like (4) (discussed in chapter 6.1):
(4) CAUSE(CAUSE(the gardener, BECOME(water, PLACE(on, the tulips)), BECOME(her
The structure would translate into ''*The gardener watered the tulips her
sneakers soggy''. Since there are only three syntactic argument slots in each
sentence available, the parser would not link ''the sneakers'' in AS; but there
does not seem to be a mechanism that prevents ''The gardener waters the tulips
soggy'', which is not what is intended in CS. This problem is addressed in
Chapter 5.4. The suggested solution is that the result adjective is itself a
result clause, whose theme can only fuse with the theme of the superordinate
result clause. However, the logic (or perhaps only the exposition) of this
argument is somewhat opaque.
Let us review the problem of Argument Structure theories and evaluate this book.
Can Randall accomplish the task of predicting the mapping from semantics to
syntax for any kind of verb better than other theories? Mostly.
Searching for the key to Argument Structure in the semantics of a verb promises
far more success than positing invisible layers in the syntax for which there is
no evidence on independent grounds. However, by adhering to many traditions of
Generative Grammar without revising them, Randall robs her theory of many
possibilities. One of them is the Theta Criterion, which is treated as an axiom,
and not as merely a default case. This leads to positing invisible levels or
elements in CS trees, which is just like positing invisible levels in the
syntax: Not justifiable on independent grounds.
Another syntactic heirloom is the external/internal argument distinction, which
Randall translates into CS. This complicates matters unnecessarily. A
description in terms of semantic qualities would have sufficed (many phenomena
that are ascribed to unaccusativity could be easily solved in terms of agency,
telicity, or animacy). One cannot help but have the impression that in order to
satisfy the syntactic rules, many grammaticality judgments are based on the
theoretician's introspection, and are not entirely convincing: most native
English speakers tend to find ''the watering of tulips flat'', ''the rolling of
boulders smooth'', or ''The twisting of ropes taut'' equally bad or even worse than
the presumably ungrammatical *''the running of shoes threadbare''. Other
grammaticality judgments are proven questionable by quick Google searches (most
of the examples marked as ungrammatical on p. 76 yield more than half a million
hits; *''beachgoer'' even has an entry in the Merriam-Webster dictionary). Since
these data are a cornerstone of the unaccusative verbs' analysis, it might have
been better to obtain quantifiable speakers' judgments instead of relying on
Despite this criticism: Randall's theory succeeds in being strictly
structure-preserving, which has been a desideratum in many approaches for years.
She also succeeds in keeping the inventory of semantic primitives small. And
most importantly, she offers profound analyses and insights into an incredible
variety of verb classes, which researchers from all theoretical backgrounds will
Chomsky, N. (1995) The Minimalist Program, MIT Press.
Chomsky, N. (1981/1993). Lectures on Government and Binding: The Pisa Lectures.
Mouton de Gruyter.
Culicover, P. & R. Jackendoff (2005) Simpler Syntax, Oxford.
Dowty, D. (1991) Thematic Proto-Roles and Argument Selection, Language 67, 547-619.
Hale, K. & J. Keyser (1993) On argument structure and the lexical expression of
syntactic relations. In: Hale, Kenneth & Jay Keyser (eds.) The view from
building 20. Cambridge, MIT Press.
Hale, K. & J. Keyser (2002) Prolegomenon to a theory of argument structure.
Cambridge, MIT Press.
Horvath, J. & T. Siloni (2002) Against the Little-v Hypothesis, Rivista di
Grammatica Generativa 2002, 27, 107-122.
Jackendoff, R. (1992) Mme. Tussaud Meets the Binding Theory, Natural Language &
Linguistic Theory 10, 1-31.
Jackendoff, R. (1983) Semantics and Cognition. Cambridge, MIT Press.
Jackendoff, R. (1972) Semantic Interpretation in Generative Grammar. Cambridge,
Kratzer, A. (1996) ''Severing the External Argument from its Verb'', in Johan
Rooryck and Laurie Zaring (eds.) Phrase Structure and the Lexicon, Dordrecht:
Levin, B. & M. Rappaport Hovav (2005) Argument Realization, Research Surveys in
Linguistics 3, Cambridge University Press.
Wiese, H. (2004) Semantics as a gateway to language, in: Härtl, Holden, & Tappe,
Heike (eds.), Mediating between Concepts and Grammar. New York: de Gruyter,
ABOUT THE REVIEWER
| ABOUT THE REVIEWER:
Eva Wittenberg is a Visiting Fellow at the Center for Cognitive Studies at
Tufts University, and at the Laboratory for Developmental Studies at
Harvard University. Her research interests include the syntax-semantics
interface, specifically Argument Structure questions, and psycholinguistic
methods. She is currently preparing a dissertation on Argument Structure
mismatches under the supervision of Heike Wiese at Potsdam University, Germany.