Exploring the Syntax-Semantics Interface
Robert D. Van Valin, Jr.
Cambridge: CUP, 2005
Reviewed by Olivier Simonin
Exploring the Syntax-Semantics Interface is a clearly written general introduction to Role and Reference Grammar (henceforth RRG), by its leading proponent. Although the title highlights the study of the interconnection between syntax and semantics, which is a prominent feature of RRG, the book is primarily an up-to-date presentation of the theory. It follows the same structure as Van Valin and La Polla’s Syntax: structure, meaning and function (1997), while being more concise and easily readable. Naturally, the interaction between syntax and various aspects of semantics is emphasized throughout the book – i.e. those aspects which are taken as relevant within RRG. The range of phenomena surveyed and the typological diversity which illustrates them make it highly rewarding reading. It will be of much interest to all those who wish to broaden their conceptual horizons, not merely to linguists who are curious about how the interaction between syntax and semantics can be formalized.
The first four chapters are devoted to the three fundamental components of RRG: syntax, semantics (chapter 2), and information structure (chapter 3). The discussion of the syntactic component is broken up in order to introduce basic clause structure initially (chapter 1), before syntactic relations are investigated in detail (chapter 4). The second part of the book is concerned with how these three essential components interrelate. After proposing a device to effect just this in simple sentences (chapter 5), the author argues for a theory of complex sentences (chapter 6), on which he builds in order for his account to cover even such sentences (chapter 7).
Chapter 1: ‘Syntactic structure’. Clauses have a layered structure. Discounting its arguments, a predicate can be equated with a NUCLEUS, whereas it makes up a CORE with them. Adjuncts may occur in the PERIPHERY, and are then contained within a CLAUSE. There are thus several strata of syntactic construction. In Dana saw Pat yesterday, the verb saw belongs to the nucleus of the sentence, Dana saw Pat to its core, and the whole pertains to the clause it represents. Other entities are postulated: a/ precore or postcore slots, and detached positions—respectively appended to a core or a clause, and b/ operators, which have scope over a nuclear, core, or clausal layer. For instance, root modals—like can in You can see him—are core operators. Those operators ‘modify the relation between a core argument, and the action’ . ‘You’ is allowed to see someone, in the example just given. Quite significantly, ‘the ordering of the morphemes expressing operators with respect to the verb’ relates to ‘their respective scopes’ . After showing how some interesting linguistic phenomena can be captured within this framework (verb-agreement connections directly marked on verbs; the position of adverbs potentially changing the layer they pertain to, and thus their meaning), the layers found in clauses are argued to be partly reflected in the structure of NPs. This structure is similarly layered, judging by nominalizations such as the construction of the bridge in New York City, though it retains specificities (e.g. determiners are classified as NP operators).
Chapter 2: the bulk of ‘Representation and semantic roles’ deals with verbal semantics. Vendler’s aktionsart typology is first revisited. Semelfactives and active accomplishments are added to the four basic verb groups (which are: states, achievements, accomplishments, and activities). Active accomplishments imply the telic use of activity verbs—i.e. they are telic actions (e.g. I went to this place). Like accomplishments and unlike achievements, they are not punctual events. Semelfactives are punctual events with no result state (e.g. The light flashed; see Smith 1997, The parameter of aspect, 2nd edition). Predicates can also be classified according to whether causation underlies them (I moved the stone = I caused the stone to move). Though not foolproof, a battery of tests is offered to determine to which category a predicate belongs. Semantic roles are then explored. There are three types of semantic roles in RRG: 1/ verb-specific semantic roles, which are the most specific, being directly derived from a verb class (e.g. giver, believer, broken [thing]); 2/ thematic relations, which are a first step towards generalization (e.g. giver, runner, killer, are grouped together within the agent thematic relation); 3/ macroroles—actor and undergoer. Sometimes, a semantic category, say experiencer (somebody hearing, thinking, or liking something), is liable to be interpreted as either actor or undergoer, depending on whether the state of affairs is construed as an activity or a state.
Chapter 3: ‘Information structure.’ The discursive and pragmatic component of RRG draws extensively on Knud Lambrecht’s work (especially his 1994 Information Structure). Focus, topic, and information units are defined, and then given representational status. Two illuminating developments on the significance of information structure are provided in relation to syntax. First, the notion of VPs is re-analyzed in terms of focus patterns, along with the phenomena usually adduced to postulate a VP category (VP fronting, VP deletion…). Second, the author notes, with other linguists, that there exists a connection between focus structure and quantifier scope interpretation. In an utterance like Every girl kissed a boy, focus structure contributes to determining whether the girls kissed one boy only, or whether each girl kissed a different boy.
Chapter 4: ‘Syntactic relations and case marking.’ Grammatical relations, such as subject or object to a verb, need to be postulated as far as they signpost what semantic roles NP referents fulfil. Nevertheless, in a language such as Achenese (Austronesian family), semantic roles can be directly interpreted through the pre-verbal or post-verbal position of clitics—i.e. whether the NP refers to an actor or an undergoer, from which more specific roles are then deduced. Yet, in almost all human languages, some arguments are given privileged syntactic status. There are two types of PRIVILEGED SYNTACTIC ARGUMENTS (PSAs): a/ controllers may trigger verb agreement (like subjects in English), antecede a reflexive (the subject he in ‘He did it himself’), or supply the interpretation for a missing argument in an adjacent unit (it then roughly corresponds to a controller in generative grammar); and b/ pivots, which supply the missing argument in a linked unit, when an argument can be argued to bridge two constructions, like him in I saw him do it. The category subject can be seen as a PSA which has been generalized to many contexts. Not all languages with PSAs, however, have such a bias that makes it possible to single out a subject category. A formal procedure is then introduced to select PSAs in natural languages.
Chapter 5: ‘Linking syntactic and semantic representations in simple sentences’ is done through a linking algorithm, which differs depending on whether one goes from semantics to syntax, or the other way round. The first linking process has much to do with production, the second with comprehension. In both cases, discourse-pragmatics influences the linking. For instance, some focus structures may occur jointly with specific syntactic constructions. In John I saw, the topicalized element John is in focus. Focus structure is more generally shown to play a major role in linking, in addition to many other aspects in RRG.
Chapter 6: ‘The structure of complex sentences.’ The RRG description of complex sentences relies on two fundamental concepts: NEXUS and JUNCTURE. Nexus corresponds to the distinction between coordination and subordination, to which is added a halfway type, co-subordination, or dependent coordination: units are joined together in a coordinate-like relations, while still sharing some grammatical marking (languages with switch-reference are a paradigmatic example). The notion of nexus is thus both to do with logic and linguistic marking. On the other hand, JUNCTURE is concerned with the nature of the units being linked: nuclei, cores clauses, or sentences (= clauses + clausal operators + detached constituents). Juncture-nexus types are then graded according to the degree of binding between the propositional units linked, and related to a semantic hierarchy of states of affairs, indicating the cohesion obtaining between the states of affairs expressed by such units.
Chapter 7: ‘Linking syntax and semantics in complex sentences’ extends the linking algorithm to complex sentences. It provides an interesting discussion of the interaction between the syntax and the focus structure of constituents occurring at the boundary of propositional units.
Arguably, the strong points of the book lie in the discussion of discourse-pragmatics and its influence on the semantic and syntactic components, as well as in the impressive wealth, and unusual broadness of data that is covered. It is rather difficult to do full justice to a book which sets out to present a full-blown theory. Yet, most phenomena treated must have a refreshingly exotic aspect, for linguists trained in syntax and working on mainstream Indo-European languages. In other words, Exploring the Syntax-Semantics Interface contains a mine of useful, out of the way observations, which justifies careful reading. There is, in fact, little wonder that it does: RRG has been especially devised to account for little-known languages and what has long been deemed of minor interest.
The theory itself is original, and the formalization that underpins it thought-provoking, especially in the way the three major components of RRG are interrelated (syntax, semantics, and pragmatics). However, as is perhaps inevitable with such pro-actively formal attempts, many representations and mechanisms postulated can be questioned. Two main assumptions will be queried here: first, that macroroles (introduced in chapter 2) are legitimate semantic entities, and, second, that universal templates should exist for all sentences and phrases.
The major difficulty faced by the notion of macroroles is that there are only two of them (actor and undergoer), and at most two of them connected to a predicate. It is all well and good with canonical transitive verbs, like eat or hit (the two arguments are undergoers for eat, whereas there is an actor for hit). With a ditransitive pattern, however, one ‘logical’ argument is bound not to be interpreted as fulfilling a macrorole. For instance, in I gave the book to her, the subject is construed as an actor, and the direct object (the book) like an undergoer, whereas the indirect object is not linked to a macrorole. In I gave her the book, I is still understood as representing an actor, and the object occurring first, now her, an undergoer, while the book is not connected to any macrorole. RRG assumes so much in order to account for the fact that direct (i.e. first, in English) object arguments have privileged syntactic status, when compared to their companion objects.
The problem with this kind of reasoning is that one no longer knows whether one is still dealing with semantics, as opposed to information structure. All logical arguments are not interpreted as macroroles, which creates an imbalance, moving the semantic component away from a logical base. It is quite significant, in this respect, that the only variation noted between languages in the lexical phase of the linking algorithm, has to do with macrorole assignment and ditransitive patterns. Leaving those concerns out of the semantic domain might contribute to unifying the theory in a more satisfying fashion.
Like Dik’s functional grammar, to quote but one famous example, RRG postulates the existence of templates for sentences and NPs, whose slots are filled out, though a significant number of them can remain empty, unencumbered with zero constituents. Although the validity of each slot should be examined, I believe that a more fundamental issue is worth considering: is it legitimate to hypothesize a rich variety of slots, some of which are but rarely filled? The clearest example is perhaps to be found in NPs. NPs are argued to have a structure similar to that of clauses, on account of nominalizations such as their building a new bridge in New York City (chapter 1). Yet one might wonder whether classical NPs, like a new bridge, fit so neatly a template originally developed to account for propositional units.