CNL Lunch

Silvia Gennari (w/ Crain & Meroni)

Rapid Relief of Stress in Dealing with Ambiguity

Thursday March 8th, 12:30pm, 3416 Marie Mount Hall

This study used a head mounted eye-movement recording system to investigate the influence of contrastive stress during the on-line interpretation of spoken sentences in the context of visually presented scenes. Test sentences contained the focus operator only, as illustrated in (1) and (2). Such sentences assume that a set of alternatives to the focus element has previously been introduced in the context (the contrast set); the assertion is that the focus element has some unique property which members of the contrast set lack. Thus, (1) presupposes that the mother could have brought something other than milk to the boy, and asserts that in fact the mother brought him nothing but milk.

(1) The mother only brought some MILK to the boy. (contrastive stress)
(2) The mother only brought the boy some milk. (neutral stress)

Reinhart [1] claims that sentences like (1) tax the parser more than sentences like (2) because of an inherent preference to compute a contrast set based on the sentence's neutral stress pattern, which in (1) falls on the Indirect Object the boy. This initial computation must be subsequently abandoned, however, in response to the stress shift onto the Direct Object some milk. This predicts that sentences like (2) are easier than ones like (1) in contexts in which both sentences are true and receive the focus interpretation on the Direct Object (some milk). An alternative hypothesis can be generated, based on Rooth's [2] observation that phonological stress marks the intended focus element unambiguously, as in (1) (see also [3]). By contrast, multiple semantic representations may be built in (2), in the absence of any overt indication of the intended focus element. Therefore, two possible contrast sets may be generated, one for each post-verbal NP.

To evaluate these alternative hypotheses, subjects (n=25) were asked to judge the truth/falsity of auditorily presented sentences against visual scenes like (3).
(3) Entities in scene: boy's milk = Direct Object/focus element, boy = Indirect Object, Man = contrast to Indirect Object, Man's coffee = contrast to Direct Object, man's milk.
The dependent measures were overall response time and fixation duration on objects in the scene, including the focus objects and the contrast sets associated with the Direct Object (the milk) and the Indirect Object (the boy). On Reinhart's proposal, both contrast sets should be created for (1), but not (2), while the alternative account makes just the opposite prediction.

Results indicate that overall response time was greater for sentences like (2) than for sentences like (1) (p = .06). Moreover, subjects fixated longer on the contrast set corresponding to the Indirect Object in sentences like (2), (p=.008) indicating that they found these sentences ambiguous. The pattern of fixations makes it clear that sentences like (2) were interpreted with two contrast sets, one for each post-verbal NP. In the temporal regions corresponding to the two relevant NPs in (1) and (2), fixations on the Indirect Object contrast were longer for (2), indicating that the ambiguity was not resolved up until the response. By contrast, sentences like (1) appeared to be unambiguous for subjects, who fixated primarily on the contrast set corresponding to the Direct Object in the two NP regions. This pattern of fixations excludes a possible order effect because the same incoming NP information in (1) and (2) is processed differently (in its corresponding region) when stress is present. Finally, for both types of sentences, subjects fixated on the contrast sets more than any other entity in the scene in both NP regions, suggesting that the computation of the contrast sets occurred on-line.

The findings support the view that alternative meanings are computed in parallel in sentence processing, and they indicate that contrastive stress facilitates the resolution of ambiguities involving the computation of multiple contrast sets.
References

[1] Reinhart, T. The Processing Cost of Reference-Set Computation: Guess Pattern in Acquisition, Uil OTS working Papers.
[2] Rooth, M. 1992 "A theory of Focus interpretation", Natural Language Semantics, 1, 75-116.
[3] Sedivy, J. 1997, A Model-Base Psycholinguistic Study of Semantic Contrast, Ph.D. dissertation, University of Rochester.