Draft Proposal

Argument Structure: The Lexicon—Syntax Interface

Adele E. Goldberg

At the suggestion of Robert Van Valin and John Davey, I propose to write a book titled, Argument Structure: The Lexicon—Syntax Interface, to be published in the series of surveys on Syntax and Morphology edited by Van Valin. The book will aim to provide an accessible, critical, up-to-date source of information about the linguistic theory of simple clauses, or argument structure. The text will be aimed at graduate students and advanced undergraduates, and will be written in as accessible a way as the subject matter allows.

The book will synthesize work from a number of theoretical paradigms demonstrating a number of convergences in the field, including the following:

1)the trend toward more detailed semantic representations and the decline of atomic semantic roles

2)the incorporation of information structure features (e.g., topic, focus) into syntactic and/or argument structure representations

3)the recognition that the main verb does not always directly or uniquely determine clausal syntax and semantics

4)the recognition that the lexicon contains phrasal form-meaning pairings as well as word or root level pairings of form and meaning

The book further argues that the apparently wide divergences derive to a great extent from differing answers to a single question of how to treat linguistic paraphrases or alternations. The conviction that alternations should be generated from a single base form motivates derivational theories of argument structure. The view that alternations are best captured by lexical redundancy rules underlies current monostratal generative approaches (LFG, HPSG). The idea that each surface form of an alternation should be treated on a par and analyzed on its own terms, with alternations accounted for only by meta-grammatical description is the cornerstone of more cognitive and functional approaches. Today, these widely divergent approaches to alternations are generally assumed without in depth discussion. Debates typically center only around how to implement the assumed strategy, making cross-theoretical comparisons difficult. The book refocuses attention on this pivotal issue, offering new data and relying on a form of argument used by Chomsky (1970), that has been broadly accepted by researchers from various theoretical backgrounds.

Chomsky (1970) stands as a classic argument against deriving nouns from their verbal counterparts. Chomsky’s argument was basically that the NPs projected from derived nouns have exactly the syntax of NPs projected from underived nouns. To avoid an account in which this is mere coincidence, we need to recognize that both types are base-generated as nouns. With Williams (1991) we might call this the “target syntax argument:” if form B is clearly not derived and form A has the same target syntax as form B and is distinct from C, then any argument to derive A from C is undermined.

Williams (1991) makes a parallel “target semantics argument”, observing that the meanings of NPs projected from underived nouns fall into certain categories: the meanings of NPs projected from derived nouns fall into exactly the same categories; moreover, the categories of NP meanings are different from the categories of S meanings. That is, given that the syntax and semantics of “derived” nouns are like that of underived nouns, and unlike the syntax and semantics of clauses, it is clearly simpler to allow the nouns to be base-generated as nouns and not derivative of clause structures.

It is observed that the exact same reasoning leads to the conclusion that both instances of classic “alternations” should also be independently generated. Consider the “spray/load” alternation as in (1) and (2).

(1)Pat loaded the wagon with the hay.

(2)Pat loaded the hay onto the wagon.

Notice (1) has the same target syntax as the examples in (3): all have a DO and a with PP.

(3)a. Pat loaded the wagon with the hay.

b. Pat adorned the tree with lights.

c. Pat broke the window with a hammer.

The target semantics of the examples in (3) is also similar: the DO is affected in each case, and the with phase codes an instrumental-type argument.

At the same time, (2) has the target syntax and semantics as the examples in (4), each involving a DO and directional phrase and entailing caused motion:

(4)a. Pat loaded the hay onto the wagon.

b. Pat put the hay on the wagon.

c. Pat shoveled the hay into the wagon.

Notice also that the b and c forms of (3) and (4) cannot be derived forms, as demonstrated in (5) and (6) respectively.

(5)b. *Pat put the wagon with hay.

c. *Pat shoveled the wagon with the hay.

(6)b. *Pat adorned lights onto the tree.

c. *Pat broke a hammer onto the window.

That is, both the target syntax argument and the target semantics argument hold of these alternations as well. Potential counterarguments and implications will be discussed in detail.

Other chapters will focus on different aspects of relating form and meaning. Current linking theories will be surveyed, and their strengths and weaknesses critiqued (chapter 3). Tricky data for linking theories, including various types of complex predicates, will also be discussed (chapter 4). Differing perspectives on what is contained in the lexicon will be reviewed, and the current trend to include phrasal patterns as well as stems will be highlighted (chapter 5). Very different proposals for what the phrasal patterns and their associated semantics should be will be analyzed (chapter 6). An over view of the present state of the field will be offered (chapter 7). Finally, several key questions that have been the focus of research on argument structure are reviewed in light of the discussion in the book (chapter 8).

I feel I am in a strong position to publish a book that will be well received and widely read. I am fortunate to have received a broad training in the subject matter, having a strong background in both functional and lexical approaches to Argument Structure. My previous book, Constructions: A Construction Grammar approach to Argument Structure (University of Chicago Press, 1995) touches on many issues that will be dealt with in greater detail in the present volume. I also have the resources available to enable me to provide an up-to-date analysis of other generative approaches. Visits with several key researchers are being arranged to insure various perspectives are discussed accurately.

While Argument Structure has remained at the heart of linguistic theorizing for the past forty years, remarkably, no critical survey such as the one proposed currently exists. The possibility of highlighting convergent trends and clarifying existing divergences represents an exciting and timely opportunity that has the potential to broadly impact the field. Below I include a summary of each of the book’s chapters.

I would aim to submit a manuscript to Oxford before 2003.

1: Issues of meaning: What needs to be captured and how best to capture it?

This chapter discusses both the traditions of semantic or theta roles, and lexical decompositions, emphasizing the generalizations that these constructs are intended to capture. There is a general trend away from positing contentful semantic roles, favoring lexical decompositions. The reasons for this trend are outlined, including the fact decompositions allow for certain paraphrase relations to be represented explicitly, the fact that it has never been clear what the inventory of semantic roles is, and the fact that semantic roles fail to capture the relational nature of the slots. Work by Jackendoff, Van Valin, Wierzbicka, Levin and Rappaport, Pustejovksy, Harley, and others are discussed in this chapter. Criticisms of decompositions are also discussed (e.g., Fodor and Lapore 1997). Issues related to problems of granularity are also analyzed: how specific should the roles or decompositions be? The answer clearly depends on what aspects of form are meant to be predicted; for example, predicting simple transitivity requires less specificity than predicting the distribution of various adverbs.

The trend toward including information structure features, e.g., topic and focus, is also analyzed. Again the convergence in the field is striking (see, e.g. P&P approaches; Yoon, to appear; Jackendoff, to appear; Williams to appear; Van Valin and LaPolla 1997; Fillmore et al. to appear; Langacker 1987).

Inclusion of still additional aspects of meaning or function is a cornerstone of functional approaches. For example, Construction Grammar aims to incorporate semantic frames of world knowledge, even though frames are notoriously difficult to represent. Cognitive Grammar presents perhaps the most fully fleshed out attempt to incorporate fine-grained aspects of meaning and function into grammatical representation.

2: Alternations: a pivotal issue

The question that perhaps most neatly divides current generative and functional approaches surrounds the issue of how to account for close paraphrases involving the same heads, or alternations. Two broad approaches can be identified with divergent answers to this question. Generativist linguists, by in large, focus on the similarities between paraphrases or “alternations” such as those in (1)-(2) and (3)-(4):

(1)Pat gave Chris a book.

(2)Pat gave a book to Chris.

(3)Pat loaded the hay onto the wagon.

(4)Pat loaded the wagon with the hay.

The situations in which (1) is true are the same as those for which (2) is true, and vice versa; thus the sentences are truth functionally equivalent. The same is roughly true for (3) and (4). In generative approaches, alternative ways of expressing a truth functionally equivalent (or nearly equivalent) message are related explicitly by the grammar, either by derivation or by lexical rule. The intuition is that the sentences in (1) and (2) or those in (3) and (4) are so close semantically that their relationship must be captured by the grammar.

There have been various implementations. Early Generative Semantics pioneered this type of work. They aimed to derive any two close paraphrases from a single source. The early enthusiasm for “lumping” paraphrase relations together by derivation, as opposed to “splitting” them by allowing multiple base-generation, eventually led to derivations in which there was no morphological overlap. This trend was curtailed in the face of striking criticisms (Chomsky 1970; Fodor 1970). However, recent trends in the Principles and Parameters framework appear to be reviving the basic idea (Baker 1988; 2001; Hale and Keyser 1993). Functionalists on the other hand, focus on the long-recognized, subtle differences in meaning between near paraphrases and argue that each should be generated, or accounted for, independently. The nature of semantic representation, discussed in chapter 1, thus plays a role in determining different stances on the question of alternations. This chapter aims to refocus attention on this pivotal issue, offering new data suggesting that derivations are not always appropriate, using a form of argument that has been broadly accepted by researchers adopting various approaches (Chomsky 1970; Williams 1991).

3. Relating semantics to overt expression

Chapter 3 reviews well-known mismatches between semantic arguments and syntactic complements including dummy subjects; redundant reflexives, raising constructions (see Jackendoff 2002 for summary). This chapter also focuses on standard ways to relate semantics to overt expression, including role hierarchies and linking rules. The topic is directly related to the focus of chapter 2, as the question of whether the aim is to link near paraphrases to the same underlying structure, or directly to distinct surface forms plays a major role in determining how general the linking rules may be.

4. Complications in relating semantics to overt expression

The history of the notion of “government” and “head” are discussed; the traditional role of the main verb in predicting the overall form and meaning of the clause is detailed. The topic of complex predicates, including noun incorporation and serial verbs, is also discussed. Varied accounts of complex predicates have been offered which preserve the idea that the main verb is the locus for clausal linking; strengths and weaknesses of such accounts are addressed. Constructional issues, in which aspects of the overall form or meaning are not naturally attributed to the main verb are also discussed.

5. What’s in the lexicon anyway?

This chapter focuses on the current trend toward a much broader view of the traditional lexicon. Many have recognized the need to include stored phrasal patterns as well as stems or words including DiSciullo and Williams 1987; Fillmore, Kay and O’Connor 1988; Goldberg 1992, 1995; Williams 1994; Michaelis and Lambrecht 1996; Jackendoff 1996, 2002; Ackerman & Welbelhuth 1997; Marantz 1997.

6. Assigning meaning directly to formal structures

This chapter focuses on specific proposals for how to treat meaningful phrasal patterns. Differences between approaches are observed to depend on whether formal structures are taken to be surface structures or underlying structures. Further differences arise based on the type of meaning assumed (which relates back to discussion in chapters 1-3). The approaches of Borer (1994), Hale and Keyser (1993), Montague Grammar; Williams (1994); Jackendoff (1990; to appear); Goldberg (1992; 1995) are discussed in this chapter.

7. How successful have attempts been at predicting form from meaning?

The penultimate chapter is aimed at taking stock of major successes and failures in the domain of relating form and meaning. To discuss the title question coherently, we clearly must decide what we take to be “meaning” and what we take to be “form.” Several subquestions are discussed: Does truth-functional meaning predict surface form? The clear answer to this question is no, since a multitude of surface forms exist for a given truth-functional proposition; Does truth-functional meaning predict underlying form? The assumption among generative grammarians is typically yes, with variations often ascribed to complex derivational processes. Do highly specific functional specifications predict surface form? Different theories answer this question in different ways. Construction Grammar answers in the negative (e.g., Lambrecht 1994), observing that language-particular idiosyncracies and paradigmatic contrast play a role. Cognitive Grammar on the other hand, argues that, modulo phonology and word order, the answer is yes. This last theory is explicitly non-structuralist and weakly Whorfian.

ShelDOn and Strange (1984)

8. Conclusions

The final chapter provides an overview of the major issues raised.

How are semantic generalizations best captured? The answer to this question depends in part on what one wants to use the semantic generalizations for. A case was made that very specific functional information is critical if the aim is to predict or even motivate overt surface expression; it remains unclear whether sufficiently subtle semantics will be sufficient, and there are reasons to be skeptical; it is possible to stipulate that coarse semantics predicts underlying form, but at the cost of complicating derivational mechanisms that lead to surface form.

Are semantic constraints and syntactic constraints independent, or is one predictable from the other? In answer, we observe a tendency for semantics and syntax to be aligned; this tendency as been captured in various ways, for example, by “iconic” motivations (functionalists); a “faithfulness” constraint (OT); an “economy of shape” preference (Williams, to appear), or a preference for alignment (Jackendoff 2002). At the same time, we note that there are mismatches in which syntax and semantics are not aligned that must be captured as well. Mismatches are observed to arise from paradigm effects, historical changes and conflicting motivations (Goldberg, to appear; Francis and Michaelis, eds. to appear).

How should the grammar handle two surface forms with closely related meaning? This was the critical question focused on in chapter 2. The answer has often been assumed without reflection or discussion; differences in the answer to this question have spawned very different approaches to the phenomena; a summary of pros and cons of various approaches is offered

Is the main verb responsible for argument structure? We suggest that this is an idealization that had been the null hypothesis until recently; the growing consensus is that it is not always true. The current danger is that that the pendulum will swing too far the other way resulting in a view that semantics is “in” syntax with words being merely window-dressing.

How do we reconcile the strong tendencies for 1) syntax and semantics to align and 2) the tendency for the main verb to determine syntax and semantics, with the vast number of exceptions? A number of trends aimed at capturing these facts are discussed. These include an increased emphasis on the lexicon, OT syntax, default inheritance, and the “shape conservation” of Williams (to appear).