Scrambling and Processing: Dependencies, Complexity, and Constraints

Two approaches characterize the current state of research in sentence processing. Both approaches aim to arrive at a well-grounded analysis of relevant constraints the construction is subject to. The first approach relies heavily on the hypothesis that a psycholinguistic investigation of a phenomenon can shed the light on some unresolved linguistic problem. The second approach advocated by cognitive science suggests that ideally we would study every construction in a language from both linguistic and psycholinguistic points of view. The first approach is firmly grounded in linguistics. Syntacticians are interested in the derivation of sentences and have ways of investigating how a sentence’s structural representation is defined by competence grammar. What distinguishes one syntactic theory from another is which the linguistic grounds it uses to argue for a particular structural representation. Often, theories compete with each other over long periods of time because linguistic methods do not strongly favor one theory over the other. One possible way to answer linguistic questions, suggested by Fodor (1993), is to investigate sentence structure using psycholinguistic methods that reveal how speakers and readers mentally represent such structure. The phenomenon of Scrambling is particularly fit to provide psycholinguistic evidence that may help to contrast different theories of grammar. These include generative grammar (Chomsky 1981; 1986; 1995) and the non-generative grammars such as Head-Driven Phrase Structure Grammar (HPSG) (Pollard and Sag, 1987; 1994), Lexical-Functional Grammar (Bresnan, 1998), and Categorial Grammar (Steedman, 2000). The Principles-and-Parameters version of the generative grammar assumes that arguments in a sentence derived by syntactic movement leave empty categories, that is, phonetically null placeholders in their original position. In contrast, the non-generative theories view empty categories, also known as traces, as theoretically undermotivated constructs and prefer accounts that do not require them. The second approach lies within the realm of contemporary cognitive science (Lepore and Pylyshyn, 1999). The syntactic study of word order variation cross-linguistically should contribute to a theory of the universal principles and constraints on language variation which govern Scrambling. It is also a necessary preliminary step in providing an account of how scrambled constructions are processed in sentence comprehension. Just as the theory of grammar has as its goal an account of Universal Grammar and parameters of language variation, the theory of sentence processing has as its goal the characterization of the universal parser, the human sentence processing mechanism. The theory of sentence processing has to address numerous questions. One question involves how movement constructions, usually referred to as filler-gap dependencies, are processed in general; another involves how scrambled constructions are processed in particular. Several theories of sentence processing argue for different accounts of Scrambling, but most of them start with a hypothesis that scrambled sentences are more complex than sentences in canonical word order. Complexity of a sentence can be broadly defined in structural terms (Fodor and Frazier, 1978; Frazier and Clifton, 1996), in terms of constraints it violates (MacDonald, Pearlmutter, and Seidenberg, 1994), in terms of computational resources (Gibson, 1998) or as working memory load (Just, Carpenter, and Hemphill, 1996). A considerable amount of experimental work has been done on Scrambling data from typologically diverse languages, using both the linguistic and psycholinguistic approaches. However, there is as yet no conclusive answer to the question of exactly how scrambled sentences are processed. Sentence processing experiments have been conducted in German, Japanese, Finnish, Serbo-Croatian, and Russian to establish the source of processing complexity of scrambled sentences in these languages, with the methodologies ranging from self-paced moving window technique to eye movement recording and cross-modal lexical priming. The studies will be compared in this chapter, and their implications for the linguistic and psycholinguistic theories of Scrambling will be discussed. The literature on this topic is growing rapidly, and in order to stay within the limits of a survey, this chapter will concentrate only on basic, clause-internal phrasal Scrambling of arguments. Other fascinating types of Scrambling, Long-Distance and Split Scrambling, will be left for another time. We will start with a brief introduction to sentence processing concepts and terms relevant to the subsequent discussion of processing of Scrambling.