Critically evaluate the Garden Path Model of sentence processing
Human sentence processing is a complex process governed by syntactic and language rules. The goal of all research is to discover how people understand language (Ferreira, Christianson & Hollingworth 2001).
There are a variety of theories about how sentences are processed and this account outlines the theory and evidence for the garden path model of sentence processing.
The garden path model of sentence processing
The garden path model of sentence processing itself was proposed by Frazier and Fodor in 1978 (Christianson et al. 2001). The garden path model of sentence processing suggests that, when encountering ambiguous sentences, only one meaning is initially processed. Then, upon reaching the end of, or a key point within, the sentence, if the meaning ascribed does not work the sentence is reparsed until a satisfactory meaning can be ascribed.
Using the sentence “While the man hunted the deer ran into the woods” Frazier and Fodor proposed that it was the words ‘the deer’ that presented the processing difficulty. When the word ‘ran’ is encountered the initial attachment of ‘the deer’ as an object of ‘hunted’ can be recognized as wrong. Therefore, rather than being attached to the subordinate clause, as would be normal, instead it is stolen by the matrix verb ‘ran’ in order to become the theme of the latter (Christianson et al. 2001).
The garden path model is a two stage model (hence the rather bizarre original article title “The sausage machine: A new two-stage parsing model”). The first stage involves syntactic information only, with analysis of the semantic information being part of the second stage.
The garden path model relies on two main principles being adhered to within sentence processing, those of late closure and minimal attachment.
The late closure principle underlies the misanalysis that initially occurs in garden path sentences. This principle shows that incoming material is attached within the clause or phrase being processed currently, rather than being attached to a separate phrase that has already been processed, hence ‘closed’. It is referred to as late closure as processing keeps options open as long as possible, and if a word was attached to a clause already processed, thus closed, it would not be available for later clauses, despite apparently being available for this (Christianson et al. 2001).
The minimal attachment principle states that “for any sequence NP-V-NP (noun phrase-verb-noun phrase), the second NP will be interpreted as a direct object” (Ferreira, Henderson 1990). Thus attachment of incoming material to the currently analysed phrase should take place using the simplest structure possible. The structure comes from the structural nodes within a sentence (see figure 1). It can be seen that processing method A utilises fewer processing nodes so is the one that would be adhered to according to minimal attachment. However, it does not allow for whatever occurs after the word defendant, as this might introduce an anomaly within the processing.
Figure 1. Two ways of analysing a garden path sentence, indicating the number of processing nodes required (Osterhout, Holcomb & Swinney 1994)
Fodor and Inoue expand that garden path model to suggest that sentences are processed according to their triage hypothesis. This is described as a form of reasoning about the probable revisability of a structure (Fodor, Inoue 2000). They indicate that the parser works on the method of saving resources as much as possible, thus will do the minimum processing possible, similar to the initial assessment of medical situations using a triage system.
It has been found that the longer the ambiguous phrase is, the more difficult processing is. So, if the error signal for a sentence is close to the head of the mis-analysed phrase it is easier to recognize, than if it is several further words distant (Ferreira, Christianson & Hollingworth 2001). For example in the sentence While the man hunted the deer ran into the woods is easier to reanalyze than while the man hunted the deer that was brown with speckles ran into the woods due to the distance between deer and ran in the second example.
Event related potential evidence
Event related potentials measure the changes in the electrical activity of specific areas of the brain that coincide temporally with specific events. When different individuals read similar sentences their brain potentials show similar effects. Three specific stages in processing are revealed by 3 specific anomalies in the trace, replicable across many individuals.
If there is a semantic anomaly in a sentence (eg the thunderstorm was ironed) then the ERP shows a deviation after 400 milliseconds in a trace taken in the middle of the brain (Datta et al. 2004). This is known as the N400 and if a sentence has no semantic anomalies (eg the shirt was ironed) then no N400 is produced.
If a sentence contains a syntactic anomaly (eg the shirt was on ironed) a deviation occurs after about 200 milliseconds on a trace taken over the left frontal lobe (Broca’s area). This is known as the LAN (left anterior negativity). LAN deviations are often produced in conjunction with the third trace deviation, the P600. This is a positive deviation after 600 milliseconds and is produced in response to grammatical violations - eg sentences that contain some ambiguous syntax or grammar.
Garden path sentences cause a P600 but no LAN deviation. This is because there is nothing ungrammatical about the sentence, it is just not as easy to understand as it first reads. The classic example of “the horse raced past the barn fell” and the anomaly arises from the final word ‘fell’. The first reading of the sentence will indicate that there is the horse that is racing past the barn of its own volition. However the final word ‘fell’ indicates that in fact there is an external factor at work upon the horse, eg a rider, that is causing the horse to be ridden past the barn, and as a result of this the horse fell. It is believed that the P600 is actually an electrophysiological marker of the garden-path effect (Osterhout, Holcomb & Swinney 1994) but unfortunately it hasn’t been explained exactly why or how the deviation occurs.
It has been shown via electrophysiological traces that older adults spend more time looking at and processing the 2nd verb in an ambiguous sentence than younger counterparts, and are overall less able to comprehend garden path sentences (Kemper, Crow & Kemtes 2004). This was theorised to be due to the impairment of working memory that occurs with age, reducing the ability to process the ambiguities within a garden path sentence.
Criticisms of the garden path model of sentence processing
There have been many criticisms of the garden path model of sentence processing since its instigation. In particular many view it as too simplistic, in that stating that initial analysis of a sentence only relies on the principles of late closure and minimal attachment.
It has also been argued that, according to the garden path model, once an ambiguous sentence has been analysed and the correct meaning ascribed, the initial incorrect interpretation is discarded. Recent studies have, however, indicated that initial misapprehensions do persist, and when questioned readers will retain some of these throughout (Christianson et al. 2001). For instance, even though on full processing of “While the man hunted the deer ran into the woods” it should be seen that the man and the deer are essentially separate entities with no connection, the impression persists that the man is actually hunting the deer in the woods (Kaschak, Glenberg 2004).
The constraint satisfaction model of sentence processing
Opponents to the garden path model of sentence processing support the constraint satisfaction model. This states that garden path sentences are analysed using all sources of information immediately. Thus the two stage element of the garden path model is obviated by the constraint satisfaction model as there is no need for a second stage if all possible analysis is undertaken immediately. Analysis of all the possible ways of interpreting a sentence takes place simultaneously, and the one with the most support ‘wins’ in terms of the output of the meaning of the sentence. Supporters of the constraint satisfaction model argue that the theory of minimal attachment does not work for all sentences. Altmann et al argue that ambiguous sentences are comprehended utilised multiple probabilistic and predictive constraints, eg all possible meanings of a sentence during and at the end of meaning are processed (Altmann et al. 1998).
The garden path model relies on the principles of late closure and minimal attachment. Whilst some have argued that these are far too simplistic it has been indicated that “minimal attachment operates during the parser's first pass through a sentence. Only in later passes does the parser consult information about verb biases” (Ferreira, Henderson 1990). This is actually a logical method of sentence processing as in all its working the brain seeks to be economical, thus minimal processing of a sentence would fit with this.
- Altmann, G.T.M., van Nice, K.Y., Garnham, A. & Henstra, J. 1998, "Late Closure in Context", Journal of Memory and Language, vol. 38, no. 4, pp. 459-484.
- Christianson, K., Hollingworth, A., Halliwell, J.F. & Ferreira, F. 2001, "Thematic roles assigned along the garden path linger", Cognitive psychology, vol. 42, no. 4, pp. 368-407.
- Datta, S., Lyon, I., MacKintosh, B., McLannahan, H., Murphy, K., Naish, P., Nettle, D., Romero, I., Toates, F. & Whatson, T. (eds) 2004, Learning and Language, 1st edn, The Open University, Milton Keynes.
- Ferreira, F., Christianson, K. & Hollingworth, A. 2001, "Misinterpretations of garden-path sentences: implications for models of sentence processing and reanalysis", Journal of psycholinguistic research, vol. 30, no. 1, pp. 3-20.
- Ferreira, F. & Henderson, J.M. 1990, "Use of verb information in syntactic parsing: evidence from eye movements and word-by-word self-paced reading", Journal of experimental psychology. Learning, memory, and cognition, vol. 16, no. 4, pp. 555-568.
- Fodor, J.D. & Inoue, A. 2000, "Garden path repair: diagnosis and triage", Language and speech, vol. 43, no. 3, pp. 261-271.
- Kaschak, M.P. & Glenberg, A.M. 2004, "This construction needs learned", Journal of experimental psychology. General, vol. 133, no. 3, pp. 450-467.
- Kemper, S., Crow, A. & Kemtes, K. 2004, "Eye-fixation patterns of high- and low-span young and older adults: down the garden path and back again", Psychology and aging, vol. 19, no. 1, pp. 157-170.
- Osterhout, L., Holcomb, P.J. & Swinney, D.A. 1994, "Brain potentials elicited by garden-path sentences: evidence of the application of verb information during parsing", Journal of experimental psychology. Learning, memory, and cognition, vol. 20, no. 4, pp. 786-803.
comments powered by Disqus