Abstract
A theory of human sentence processing is a characterisation of how people recover an interpretation for a given utterance. Ultimately such a theory should detail the processor’s organisation, the representations employed, and the algorithms used to construct an analysis — where such a characterisation is isomorphic to that of the mind. Furthermore, we must situate the sentence processor with respect to the other cognitive systems with which it interacts, outlining the nature and degree of communication at all interfaces.
This is a preview of subscription content, log in via an institution to check access.
Access this chapter
Tax calculation will be finalised at checkout
Purchases are for personal use only
Preview
Unable to display preview. Download preview PDF.
Notes
Even if something like neural networks prove to be a good approximation of low-level mental architecture, there is reason to believe they are an inappropriate level at which to characterise the high-level symbol manipulation processes (Fodor & Pylyshyn 1988). For further discussion of parallel distributed processing models see (McClelland, St.John & Taraban 1989), (Mitchell 1989), and references cited therein.
This experiment has been replicated using an on-line task in (Altmann & Steedman 1988), but the results remain open to either interpretation.
Their aim was to replicate the results achieved in (Rayner, Carlson & Frazier 1983) which ere conducted using only a null-context: an approach justifiably criticised in (Crain & Steedman 1985).
This has recently been made explicit as the Lefl-to-Right Constraint in (Frazier & Rayner 1988).
Pereira notes this problem in his attempts to implement MA and LC using a shift-reduce parser with an oracle (Pereira 1985).
In most cases LC is used to explain the preferred ‘low attachment’ of constituent in multiple clause sentences. Consider, for example, “I told you John bought the car yesterday.”. In this sentence, yesterday may modify either the main or embedded clause, but there is a general preference for the latter, which is accounted for by the LC strategy, since this is the clause “currently being parsed”.
See (Cuetos & Mitchell 1988) for evidence suggesting that LC does not hold in Spanish. This is particularly problematic given the constituent order for the relevant constructions is similar for English and Spanish.
This highlights Frazier’s rather traditional assumptions concerning how structures are built. There do exist well-known techniques which permit the monotonic ‘insertion’ of structure into a tree (Marcus, Hindle & Fleck 1983).
In Pereira’s implementation, the oracle simply chooses the rule which is minimally complex (Pereira 1985). In cases where multiple rules must be invoked to attach a lexical item, more sophisticated techniques for comparing the relative complexity of candidate analyses will be required.
In addition, to the lack of binary branching, the analysis presented for Dutch (Prazier 1987a, pages 532–533, for example), are completely at odds with current analyses involving movement to the head and specifier positions of CP (see chapter HI). As a further example, Frazier’s analysis of PP modifiers is inconsistent; they are attached as adjoined phrases for NP’s (crucially introducing an extra node), but are attached directly into VP’s — an analysis which was reserved for complements even before binary branching was introduced.
Frazier (1985), however, presents a metrical system for evaluating syntactic complexity with respect to processing which assigns relative weights to different syntactic categories, suggesting some abstract link between their content and processing complexity.
We will not digress here into discussion of the so-called Derivational Theory of Complexity (DTC), which when refuted was taken as concrete evidence against the psychological reality of TG. Not only has this been dealt with thoroughly elsewhere (see (Berwick & Weinberg 1984)), but it also bears little relevance to current principle-based theories of grammar since it was developed in the context of a rule-based TG (the Standard Theory). We will take this point up briefly in chapter VII.
For a complete exposition of his analysis of the range of conscious garden-path constructions, see (Pritchett 1992).
We will discuss this issue shortly, but the reader is referred to (Mazuka & Lust 1990) for a thorough discussion of the data.
See (Pritchett 1993) and references cited therein for details distinguishing the two analyses.
Author information
Authors and Affiliations
Rights and permissions
Copyright information
© 1996 Kluwer Academic Publishers
About this chapter
Cite this chapter
Crocker, M.W. (1996). Perspectives on Sentence Processing. In: Computational Psycholinguistics. Studies in Theoretical Psycholinguistics, vol 20. Springer, Dordrecht. https://doi.org/10.1007/978-94-009-1600-5_2
Download citation
DOI: https://doi.org/10.1007/978-94-009-1600-5_2
Publisher Name: Springer, Dordrecht
Print ISBN: 978-0-7923-3806-2
Online ISBN: 978-94-009-1600-5
eBook Packages: Springer Book Archive