Markus Bader 1,2 & Josef Bayer2
1 University of Massachusetts at Amherst, 2 University of Konstanz
This paper will investigate the relationship between memory load and syntactic ambiguity resolution. Two accounts will be contrasted. First, the Dependency Locality Theory (DLT) of Gibson (2000) which claims that the HSPM will prefer a structure incurring less memory load when faced with a syntactic ambiguity. Furthermore, when memory load is measured by performance on unambiguous sentences, a positive correlation is predicted between performance on unambiguous sentences and garden-path strength in corresponding locally ambiguous sentences.
The second account (Meng & Bader, 2000) assumes a serial HSPM which prefers syntactically less complex structures in situations of ambiguity, with garden-path recovery being a function of diagnostic reasoning. Under this account, factors decreasing processing load can lead to worse performance on garden-path sentences. For example, if a structure is more complex in being more marked within the language under consideration, it might be dispreferred in situations of ambiguity but --- if unambiguously signalled --- might lead to better performance because markedness leads to higher visibility. Other factors, in contrast, might affect unambiguous sentences and corresponding garden-path sentences in similar ways. For example, an experimental manipulation might decrease processing load and aid the diagnosis processes in finding an alternative structure in case of a garden-path.
We will report two experiments testing the different accounts. In one experiment, German unambiguous and ambiguous object-subject sentences were investigated in a speeded- grammaticality judgments experiment (cf. the examples in (1)). Two factors were manipulated: (i) the case of the object could be dative (marked case) or accusative (unmarked case) and (ii) the subject was preceded by a focus particle or not. The results show that the presence of a focus particle resulted in better performance on unambiguous sentences and a less severe garden-path effect. The case manipulation, in contrast, affected unambiguous and garden-path sentences differently: Dative case led to better performance on unambiguous sentences but stronger garden-path effects than accusative case.
A further experiment showed a parallel pattern for the two factors that were manipulated: Increasing sentence length increased memory load and garden-path strength, but a case manipulation showed opposite effects on memory load and syntactic ambiguity resolution.
In sum, the experiments to be presented argue against a unitary relationship between memory load and syntactic ambiguity resolution as proposed by the DLT but favor serial, diagnosis-based models of the HSPM. We will present a detailed model of this kind and show how it accounts for a broad range of experimental findings.
Examples
| (1) | a. | Dative Sentence | |||||
| Ich weiss, | dass | {dem Lehrer-DAT | / | Eva-NOM/ACC/DAT} | (nur) die Tante-NOM/ACC geholfen hat-DAT. | ||
| I know | that | the teacher | Eva | (only) the aunt helped has | |||
|
"I know that (only) the aunt helped the teacher/Eva."
|
|||||||
| b. | Accusative Sentence | ||||||
| Ich weiss, | dass | {den Lehrer-ACC | / | Eva-NOM/ACC/DAT} | (nur) die Tanten-NOM/ACC unterstuetzt haben-ACC. | ||
| I know | that | the teacher | Eva | (only) the aunts supported have | |||
| "I know that (only) the aunts supported the teacher/Eva." | |||||||
References
Gibson, E. (2000). The dependency locality theory: A distance-based theory of linguistic complexity. In: A. Marantz, Y. Miyashita & W. O'Neil (Eds.), Image, Language, Brain. Papers from the first Mind Articulation Project Symposium. Cambridge, MA: MIT Press.
Meng, M., & Bader, M. (2000). Ungrammaticality detection and garden-path strength: Evidence for serial Parsing. Language and Cognitive Processes, 615-666.